Repsol YPF-Harvard Kennedy School Fellows 2003

1 downloads 0 Views 5MB Size Report
formulación de un nuevo e innovador modelo .... cipalmente por miembros externos y de un Comité Directivo integrado ...... La perspectiva de la elección racional supone que los ..... Rodríguez, Armando Jiménez SanVicente, en revisión para su publicación en un libro dirigido por ...... .energy.gov/BlackoutFinal-Web .pdf).
Center for Business and Government John F. Kennedy School of Government Harvard University

Repsol YPF-Harvard Kennedy School Fellows 2003–2004 Research Papers

Repsol YPF-Harvard Kennedy School Fellows 2003–2004 Research Papers William W. Hogan Editor

Cambridge, MA April 2005

Center for Business and Government John F. Kennedy School of Government Harvard University

Repsol YPF-Harvard Kennedy School Fellows 2003–2004 Research Papers William W. Hogan Editor

Cambridge, MA April 2005

JOHN F. KENNEDY SCHOOL OF GOVERNMENT HARVARD UNIVERSITY 79 John F. Kennedy Street Cambridge, MA 02138 USA REPSOL YPF Paseo de la Castellana, 278 28046 Madrid, Spain FUNDACIÓN REPSOL YPF Juan Bravo, 3B 28006 Madrid, Spain

Copyright © 2005 by the President and Fellows of Harvard College Center for Business and Government John F. Kennedy School of Government 79 John F. Kennedy Street Cambridge MA 02138 USA Printed in the United States of America

Table of Contents About the Repsol YPF-Kennedy School Fellowship Program. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Board of Advisors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Faculty Steering Committee . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Repsol YPF-Harvard Kennedy School Fellows 2003–2004 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Juan Rosellón . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Joseph Aldy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 Darby Jack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 Cynthia Lin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 Collected Research Papers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 1.

Pricing Electricity Transmission in Mexico . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 Juan Rosellón

2.

A Merchant Mechanism for Electricity Transmission Expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 Tarjei Kristiansen, Juan Rosellón

3.

Different Approaches to Supply Adequacy in Electricity Markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 Juan Rosellón

4.

The Mexican Electricity Sector: Economic, Legal and Political Issues . . . . . . . . . . . . . . . . . . . . . . . . . . 89 Victor G. Carreón, Armando Jimenez San Vicente, and Juan Rosellón

5.

Strategic Behavior and the Pricing of Gas in Mexico . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 Dagobert L. Brito, Juan Rosellón

6.

Price Regulation in a Vertically Integrated Natural Gas Industry: The Case of Mexico . . . . . . . . . . 127 Dagobert L. Brito, Juan Rosellón

7.

Implications of the Elasticity of Natural Gas in Mexico on Investment in Gas Pipelines and in Setting the Arbitrage Point. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 Dagobert L. Brito, Juan Rosellón

8.

An Environmental Kuznets Curve Analysis of U.S. State-Level Carbon Dioxide Emissions. . . . . . 149 Joseph E. Aldy

9.

Divergence in Per Capita Carbon Dioxide Emissions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175 Joseph E. Aldy

10.

Income, Household Energy and Health . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193 Darby Jack

11.

The Multi-Stage Investment Timing Game in Offshore Petroleum Production: A Framework for an Econometric Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201 C.-Y. Cynthia Lin

12.

Estimating Annual and Monthly Supply and Demand for World Oil: A Dry Hole?. . . . . . . . . . . . . 213 C.-Y. Cynthia Lin

13.

Optimal World Oil Extraction: Calibrating and Simulating the Hotelling Model . . . . . . . . . . . . . . . 251 C.-Y. Cynthia Lin

Repsol YPF-Harvard Kennedy School Fellows

Profesores de Repsol YPF-Harvard Kennedy School

Research and Program Summary Center for Business and Government John F. Kennedy School of Government Harvard University April 2005

Resumen del Programa y de las Investigaciones Centro para Negocios y Gobierno Escuela de Gobierno John F. Kennedy Harvard University Abril, 2005

The Repsol YPF-Harvard Kennedy School Fellows Program for energy policy research, established in the Spring of 2003, provides opportunities for outstanding scholars to develop careers that will address the most challenging problems at the center of energy policy debates.

El Programa de Profesores de Repsol YPF y de la Kennedy School de Harvard para la investigación de políticas energéticas fundado en la Primavera de 2003, brinda oportunidades a académicos sobresalientes para desarrollar carreras que abordarán los problemas que plantean los desafíos más arduos centrales a los debates sobre política energética.

In the aftermath of the oil shocks in the 1970s, there was an explosion of activity in energy markets and a burst of activity in energy policy research, which launched a new generation of energy experts throughout the world. In the new century, another accelerated pace of change presents renewed demand to develop both ideas and people who can explain and clarify the complex topics that permeate energy policy. The transformed setting of international security and the expanded challenge of international terrorism profoundly affect the operation and importance of energy markets. Dramatic changes in energy industries, advances in technology, and policies to restructure energy markets present new choices and problems. There is a sense that an era of easy solutions of excess supply capacity, robust infrastructure, and low cost environmental improvements may be coming to an end. Stresses as diverse as global climate change, energy market liberalization, and greater integration of energy companies produce extended policy debates. Supported by a generous gift from the Fundación Repsol YPF, the Fellows Program responds to these opportunities and challenges by making an investment in intellectual capital through support of the research of another generation of energy policy scholars. The Fellows spend a year in residence at Harvard, and meet regularly with both faculty and visiting experts, such as Adrian Lajous, former Director General of Pemex, who was also in residence at the Center for Business and Government during academic year 2003–2004. In addition, the Fellows attended the Repsol YPF-

Luego de las violentas alzas petroleras de la década de los años setenta, hubo una explosión de actividad en los mercados energéticos y un arranque de energía en la investigación sobre las políticas energéticas, que hicieron destacar a una nueva generación de expertos en energía por todo el mundo. En el nuevo siglo, otro acelerado ritmo de cambio nuevamente exige el desarrollo de ideas y personas que puedan explicar y aclarar los complejos temas que introducidos en la política energética. El escenario transformado de la seguridad internacional y el crecido desafío del terrorismo internacional afectan profundamente la operación e importancia de los mercados energéticos. Los cambios dramáticos en las industrias energéticas, los avances tecnológicos y en las políticas para reestructurar los mercados energéticos presentan nuevos problemas y opciones de acción. Hay una sensación de que podría estar concluyendo una era de fáciles soluciones de capacidad de oferta excedente, infraestructura resistente y mejoras ambientales de bajo costo. Las tensiones tan variadas como el cambio climático global, la liberalización de los mercados y una mayor integración de las compañías de energía producen prolongados debates sobre política energética. El programa de Profesores, apoyado por una generosa donación de la Fundación Repsol YPF, responde a estas oportunidades y desafíos con una inversión en capital intelectual, mediante el apoyo de la investigación hecha por otra generación de académicos de las políticas energéticas. Los Profe1

2

Repsol YPF-Harvard Kennedy School Fellows Research and Program Summary

Harvard Energy Policy Seminar in A Coruna, ˇ Spain, June 2004. The Program provides for investigations by researchers who have already achieved high standards of excellence, and who are pursuing topics with direct relevance to current energy policy challenges. As the program continues, it will help to build an international network of energy policy research and analysis talent with the common experience as Repsol YPF-Harvard Kennedy School Fellows. By sharing expertise across borders and across generations, Repsol YPF and Harvard University are investing in a new assembly of scholars who will be capable of developing solutions to this century’s pressing energy challenges. The first year of operation demonstrated the strength of the vision and the value of the Fellows Program. The research summarized here and published in full in the companion volume, Repsol YPF-Harvard Kennedy School Fellows: Research Papers, April 2005, covers many topics, from the complex details of restructuring energy markets to important conundrums in environmental policy. Juan Rosellon built on his research from Mexico with a strong focus on restructuring to rely more on markets for electricity and natural gas. His collaborators include Tarjei Kristiansen of Norway, a former visiting scholar at the Center for Business and Government. Central problems are the incentives created by different pricing methodologies, development of rules to support infrastructure investment, and protocols for addressing strategic behavior by participants with market power. The several papers reflect completion of work begun prior to the year spent in residence as a Fellow and work that continues into the future building on the collaborations established under the program. Joseph Aldy devoted his research to investigations arising from his first-hand experience on the staff of the U.S. President’s Council of Economic Advisors, where he had been involved in the work on climate change polices. Aldy’s two papers address critical assumptions about international trends in carbon emissions and the degree to which economic development would lead to a convergence in the intensity of emissions. Both from a theoretical and empirical perspective, Aldy examined the relative experience for individual states in the United States as a possible analogy to the relative behavior of countries. The results illu-

sores pasan un año de residencia en Harvard, y se reúnen regularmente tanto con la facultad como con expertos que nos visitan, tales como Adrian Lajous, antiguo Director General de Pemex, quién también fue residente en el Centro de Negocios y Gobierno durante el año académico 2003 a 2004. Además, los Profesores asistieron al Seminario Repsol YPF-Harvard de Política Energética en La Coruña, España, en junio de 2004. El Programa apoya las investigaciones realizadas por estudiosos que ya han alcanzado altos estándares de excelencia, y que indagan sobre temas directamente pertinentes a los desafíos actuales sobre política energética. A medida que el programa procede, ayudará a establecer una red internacional de talento en la investigación y el análisis de la política energética, con la experiencia común como Profesores Repsol YPF-Harvard Kennedy School. Al compartir la pericia a través de las fronteras y las generaciones, Repsol YPF-Harvard University están invirtiendo en una nueva colectividad de académicos que serán capaces de crear soluciones para los apremiantes desafíos energéticos. El primer año de operaciones demostró la potencia de la visión y el valor del Programa de Profesores. La investigación resumida aquí y publicada en su totalidad en el volumen acompañante, Profesores de Repsol YPF-Harvard Kennedy School: Informes de Investigación, Abril de 2005, cubre muchos temas, desde los detalles complejos de cómo reestructurar los mercados energéticos hasta los acertijos de la política ambiental. Juan Rosellón aprovechó su investigación de México con un fuerte enfoque en cómo reestructura para confiar más en mercados para la electricidad y el gas natural. Sus colaboradores incluyen Tarjei Kristiansen de Noruega, un antiguo académico visitante en el Centro de Negocios y Gobierno. Los problemas centrales son los incentivos creados por metodologías diversas de determinación de precios, creación de reglas para apoyar la inversión en infraestructuras, y protocolos para responder al comportamiento estratégico de participantes con fuerza en el mercado. Los varios informes expresan la conclusión de trabajo comenzado antes del año pasado en residencia como Profesor y trabajo que prosigue hacia el futuro aprovechando la colaboración establecida bajo el programa.

Fellows 2003–2004

minate why the common assumptions of environmental convergence may not hold, which could substantially complicate the task of crafting international agreements among developing and developed countries. Darby Jack addressed a troubling environmental problem for developing countries regarding fuel choices and the implications for health. Using original data collection and subsequent analysis for Peru, Jack presents fuel choice models to consider standard welfare maximization to isolate income and price effects. He was motivated by an interest in the relative roles of information and family structure in determining the selection of fuels that create substantial indoor pollution and resulting degradation of health, especially for women and children. The data point to dilemmas in explaining the relative low percentage of expenditures on clean fuels given the high risks associated with the indoor emissions. Cynthia Lin explored issues that arise in alternative aspects of oil markets. Through the assistance of Bijan Mossavar-Rahmani, chair of the Board of Advisors for the program, Lin’s introduction to these topics included interviews with oil and gas industry experts in Houston and a visit to an oil production platform in the Gulf of Mexico. This field work led to formulation of a new and innovative structural model to characterize the role of information and interaction among owners of adjacent tracts. The model provides a new attack on the old problem of incentives in a multistage investment decision. Working at the frontier of econometric practice, Lin is pursuing empirical implementation of the model as part of the research agenda following on from the year in residence as a Fellow. A common thread is the high standard of excellence and relevance to energy problems of recognized importance. The authors’ reflections on their experience as Fellows describe benefits that they derived from the program, including an opportunity to focus exclusively on research; feedback from peers and faculty associated with the program; and exposure to experts from the energy industry. William W. Hogan Chair, Faculty Steering Committee Repsol YPF-Harvard Kennedy School Fellows Program

3

Joseph Aldy dedicó su investigación a estudios que surgen de su experiencia directa en el estado mayor del Consejo de Asesores Económicos del Presidente, en el cual participó en el trabajo sobre políticas concernientes al cambio climático. Los dos informes de Aldy tratan suposiciones críticas acerca de tendencias internacionales de emisiones y el grado al cual el desarrollo económico llevaría a una convergencia en la intensidad de las emisiones. Tanto desde una perspectiva teórica, cómo de una empírica, Aldy examino la experiencia relativa de los estados individuales en los Estados Unidos como una posible analogía al comportamiento relativo de países. Los resultados explican por qué las suposiciones comunes de convergencia ambiental podrían no conservar su validez, lo que complicaría mucho la tarea de preparar convenios internacionales entre países desarrollados y aquellos en vías de desarrollo. Darby Jack abordó un problema ambiental preocupante para los países en vías de desarrollo concerniente a las decisiones y sus consecuencias sanitarias. Mediante la recolección de datos y el análisis subsiguiente sobre el Perú, Jack presenta, para aislar efectos de ingresos y precios, modelos de elección de combustibles para considerar la maximización de bienestar estándar. Lo motivó un interés en los papeles relativos de la información y la estructura familiar en la determinación de qué las hace elegir combustibles que crean una considerable contaminación del interior de las viviendas y el consecuente deterioro de la salud, particularmente en el caso de las mujeres y los niños. Los datos señalan dilemas en la explicación del relativo bajo porcentaje de gastos en combustibles limpios dado los altos riesgos asociados con las emisiones producidas bajo techo. Cynthia Lin exploró los problemas que surgen en aspectos alternativos de los mercados petroleros. Mediante la asistencia de Bijan Mossavar-Rahmani, presidente de la Junta de Asesores del programa, la introducción de Lin a estos temas incluyen entrevistas con expertos de la industria petrolera y del gas en Houston y una visita a una plataforma de producción petrolera en el Golfo de México. Este trabajo de campo llevó a la formulación de un nuevo e innovador modelo estructural para caracterizar el papel de la información y la interacción entre los propietarios de áreas adyacentes. El modelo ofrece un nuevo enfoque al viejo problema de los incentivos en una decisión de

4

Repsol YPF-Harvard Kennedy School Fellows Research and Program Summary

For further details, see the Repsol YPF-Harvard Kennedy School Fellows Program web page at: http://www.ksg.harvard.edu/cbg/repsol_ypf-ksg_fellows/

inversión de múltiples etapas. Trabajando en los confines de la práctica econométrica, Lin lleva adelante la realización del modelo como parte del programa de investigaciones que ha seguido al año de residencia como Profesora. Un hilo común es el alto estándar de excelencia y pertinencia a los problemas energéticos de importancia reconocida. Las reflexiones de los autores sobre sus experiencias como Profesores en residencia, describen los beneficios que obtuvieron del programa; una oportunidad para enfocarse exclusivamente sobre la investigación, retroalimentación de sus pares y la facultad asociada al programa, y contactos con expertos de la industria energética, inclusive. William W. Hogan Presidente, Comité de Dirección de la Facultad Profesores de Repsol YPF-Harvard Kennedy School. Para obtener mayores detalles, consulte la página Web del Programa de Profesores de Repsol YPFHarvard Kennedy School en: http://www.ksg.harvard .edu/cbg/repsol_ypf-ksg_fellows/

Fellows 2003–2004

5

The Repsol YPF-Harvard Kennedy School Fellows Program is supported by a gift from the Fundación Repsol YPF. Program oversight and management includes an Advisory Board of primarily outside members and a Steering Committee of primarily Kennedy School faculty. Board of Advisors Bijan Mossavar-Rahmani, Mondoil Corporation, Chair Juan Badosa, Fundación Repsol YPF (2003–2004) José Luis Díaz-Fernández, Fundación Repsol YPF William Hogan, Harvard University Ramón Pérez-Simarro, Instituto Superior de la Energia (ISE) (2004–) Paul Portney, Resources for the Future John Ruggie, Harvard University Faculty Steering Committee William Hogan, Harvard University, Chair Juan Badosa, Fundación Repsol YPF (2003–2004) William Clark, Harvard University José Gomez-Ibanez, Harvard University John Holdren, Harvard University Dale Jorgenson, Harvard University Ramón Pérez-Simarro, Instituto Superior de la Energia (ISE) (2004–) Robert Stavins, Harvard University

REPSOL YPF-HARVARD KENNEDY SCHOOL FELLOWS 2003–2004 Senior Fellow Dr. Juan Rosellón Professor Centro de Investigación y Docencia Economicas (CIDE) Mexico City Pre-Doctoral Fellow Mr. Joseph Aldy Ph.D. Candidate in Economics Harvard University Pre-Doctoral Fellow Mr. Darby Jack Ph.D. Candidate in Public Policy Harvard University Pre-Doctoral Fellow Ms. Cynthia Lin Ph.D. Candidate in Economics Harvard University

6

Repsol YPF-Harvard Kennedy School Fellows Research and Program Summary

El Programa Becarios de Repsol YPF y de la Kennedy School de Harvard recibe una donación de la Fundación Repsol YPF. La supervisión y manejo del programa incluyen una Junta Consultiva integrada principalmente por miembros externos y de un Comité Directivo integrado principalmente por el profesorado de la Kennedy School. Junta de Consejeros Bijan Mossavar-Rahmani, Mondoil Corporation, Presidente Juan Badosa, Fundación Repsol YPF (2003–2004) José Luis Díaz-Fernández, Fundación Repsol YPF William Hogan, Harvard University Ramón Pérez-Simarro, Instituto Superior de la Energía (ISE) (2004–) Paul Portney, Resources for the Future (Recursos para el Futuro) John Ruggie, Harvard University Comité Directivo del Profesorado William Hogan, Harvard University, Presidente Juan Badosa, Fundación Repsol YPF (2003–2004) William Clark, Harvard University José Gomez-Ibanez, Harvard University John Holdren, Harvard University Dale Jorgenson, Harvard University Ramón Pérez-Simarro, Instituto Superior de la Energía (ISE) (2004–) Robert Stavins, Harvard University

BECARIOS DE REPSOL YPF Y DE LA KENNEDY SCHOOL DE HARVARD 2003–2004

Jefe de Cátedra (Senior Fellow) Dr. Juan Rosellón Profesor Centro de Investigación y Docencia Económicas (CIDE) Ciudad de México Becario, Pre-doctorado Sr. Joseph Aldy Candidato a Ph.D. en Economía Harvard University Becario, Pre-doctorado Sr. Darby Jack Candidato a Ph.D. en Política Pública Harvard University Becario, Pre-doctorado Srta. Cynthia Lin Candidato a Ph.D. en Economía Harvard University

Juan Rosellón Senior Fellow and Fulbright Scholar, Harvard University Professor of Economics at the Centro de Investigación y Docencia Economicas (CIDE)

Juan Rosellón, Senior Fellow and Fulbright Scholar at Harvard’s Kennedy School, is Professor of Economics at the Centro de Investigación y Docencia Economicas (CIDE) in Mexico City. As director of the program on energy economic regulation at CIDE, he researches regulatory policy problems that decision-makers face in Mexico. From 2000 to 2001, he was the editor of Economía Mexicana, one of the leading journals on the Mexican economy. He served as secretary of the Mexican chapter of the IAEE from 1999 to 2001 and has been a member of its advisory board ever since. Professor Rosellón was Chief Economist at the Mexican Energy Regulatory Commission (1995–1997), and was a faculty member in the Program on Privatization, Regulatory Reform and Corporate Governance at Harvard University (1997–2000), and at Princeton University (2001). He has been a member of the Mexican National System of Researchers (SNI) since 1994, the same year in which he received the National Award in Economics from Mexican president Ernesto Zedillo. Professor Rosellón earned a Ph.D. in Economics from Rice University. He won the Gabino Barreda Medal, the highest student honor granted by the National University of Mexico. Professor Rosellón has published widely on energy regulatory reform issues in journals such as Energy Journal, Energy Economics, and Review of Network Economics. At the Repsol YPF-Harvard Kennedy School program, he researched the reform of the Mexican electricity industry.

Juan Rosellón, Senior Fellow y Becario Fulbright en la Kennedy School de Harvard, es Profesor de Economía en el Centro de Investigación y Docencia Económicas (CIDE) en la Ciudad de México. Como director del programa en reglamentación económica de la energía en el CIDE, él investiga problemas de política reglamentaria que enfrentan los responsables de tomar decisiones en México. Desde el 2000 hasta el 2001, fue editor de Economía Mexicana, una de las principales publicaciones sobre la economía mexicana. Sirvió como secretario del capítulo mexicano de la Asociación Internacional de Economía Energética (IAEE, por sus siglas en inglés) desde 1999 hasta el 2001, y ha sido miembro de su junta asesora desde entonces. El Profesor Rosellón fue Economista Jefe en la Comisión Reglamentaria Energética Mexicana (1995–1997), y fue miembro de la facultad en el Programa de Privatización, Reforma Reglamentaria y Gobierno de las Sociedades Anónimas en Harvard University (1997–2000), y en la Princeton University (2001). Ha sido miembro de Sistema Nacional Mexicano de Investigadores (SIN) desde 1994, el mismo año en el cual recibió el Premio Nacional en Economía del presidente mexicano Ernesto Zedillo. Recibió un Ph.D. en Economía de la Rice University. Ganó la Medalla Gabino Barreda, el honor estudiantil más alto otorgado por la Universidad Nacional de México. Ha publicado extensamente sobre asuntos de reforma reglamentaria energética en publicaciones tales como Energy Journal, Energy Economics, y Review of Network Economics. En el programa Repsol YPF-Harvard Kennedy School, investigó la reforma de la industria eléctrica mexicana.

Dr. Rosellón Reviews His Year as a Repsol YPF-Harvard Kennedy School Fellow During my year as a Repsol-YPF Harvard Kennedy School Fellow, I developed various research topics that materialized in eight research papers and in two future research projects. My completed papers addressed electricity transmission expansion mechanisms, resource adequacy, a

El Dr. Rosellón Refiere Su Año Como Becario de Repsol YPF-Harvard Kennedy School Durante mi año como Becario Repsol-YPF Harvard Kennedy School, desarrollé varios temas de investigación que dieron lugar a ocho informes de 7

8

Repsol YPF-Harvard Kennedy School Fellows Research and Program Summary

political-economy analysis of the Mexican electricity reform, and natural gas regulation in Mexico. Most of these projects were of course motivated by specific public-policy problems the Mexican energy sector faces. “Pricing Electricity Transmission in Mexico” proposed an incentive regulatory framework to expand the large interregional links of the Mexican transmission network. I used a two-part tariff model within a combined merchant-regulatory approach and the “shadow” Mexican electricity system. Results suggest the best institutional structure for expanding the Mexican transmission grid would be a single transmission firm charging even tariffs along the national territory. “A Merchant Mechanism for Electricity Transmission Expansion,” Journal of Regulatory Economics (with Tarjei Kristiansen), proposed a merchant mechanism to expand electricity transmission based on long-term financial transmission rights (FTRs). Due to network loop flows, a change in network capacity might imply negative externalities on existing transmission property rights. The system operator thus needs a protocol for awarding incremental FTRs that maximizes investors’ preferences and preserves certain currently unallocated FTRs (or proxy awards) so as to maintain revenue adequacy. In this paper, we defined a proxy award as the best use of the current network along the same direction as the incremental awards. We then developed a bi-level programming model for allocation of long-term FTRs according to this rule and applied it to different network topologies. We found that simultaneous feasibility for a transmission expansion project crucially depends on the investor-preference and the proxy-preference parameters. Likewise, for a given amount of pre-existing FTRs the larger the current capacity the greater the need to reserve some FTRs for possible negative externalities generated by the expansion changes. “Different Approaches to Supply Adequacy in Electricity Markets” studies the long-run problem in electricity market design of ensuring enough generation capacity to meet future demand (resource adequacy). Worldwide reform processes have shown that it is difficult for the market alone to provide incentives to attract enough investment in capacity reserves, due to market and institutional failures. I studied several internationally

investigación y a dos proyectos futuros de investigación. Mis informes concluidos trataron de los mecanismos de expansión de la transmisión de energía eléctrica, la idoneidad de recursos, un análisis de política económica de la reforma eléctrica mexicana, y la reglamentación del gas natural en México. La mayoría de estos proyectos fueron motivados, naturalmente, por los problemas de política pública específicos que enfrenta el sector energético mexicano. “Determinación de precios de la transmisión eléctrica en México” (“Pricing Electricity Transmission in Mexico”), propuso un esquema reglamentario de incentivos para expandir los grandes enlaces interregionales de la red de transmisión mexicana. Utilicé un modelo de tarifas de dos partes dentro de un enfoque combinado negociante-reglamentario y el sistema de electricidad mexicano “sombra”. Los resultados sugieren que la mejor estructura institucional para expandir la red mexicana de transmisión sería una sola firma de transmisión que cargue tarifas uniformes por todo el territorio nacional. “Un mecanismo negociante para la expansión de la transmisión de la energía eléctrica” (“A Merchant Mechanism for Electricity Transmission Expansion”), Journal of Regulatory Economics (con Tarjei Kristiansen), propuso un mecanismo de negociante para expandir la transmisión de electricidad basado en derechos financieros de transmisión a largo plazo (FTR, por sus siglas en inglés). Debido a los flujos de lazos de las redes, un cambio en la capacidad de la red podría significar factores incidentales negativos sobre los derechos de propiedad de transmisión actuales. El operador de sistemas necesita por lo tanto, un protocolo para otorgar derechos financieros de transmisión (FTR) crecientes que maximicen las preferencias de los inversionistas y conserven ciertos derechos financieros de transmisión actualmente no asignados (u otorgamientos por delegación) de manera que se mantenga la aceptabilidad de los ingresos. En este informe, definimos un otorgamiento por delegación como el mejor uso de la red actual en la misma dirección que los otorgamientos graduales. Entonces desarrollamos un modelo de programación de dos niveles para la asignación de FTR a largo plazo conforme a esta regla, y la aplicamos a diversas topologías de red. Hallamos que la factibilidad simultánea para un proyecto de expansión de

Fellows 2003–2004

proposed measures to cope with this problem, including strategic reserves, capacity payments, capacity requirements, and call options. I also discussed the analytical and practical strengths and weaknesses of each approach. “The Mexican Electricity Sector: Economic, Legal and Political Issues,” Center for Environmental Science and Policy, Stanford University , Working Paper 05 (with Víctor Carreón and Armando Jiménez,), aimed to explain the motivations and strategies for reform in the Mexican electricity sector. We focused on the effects of politically organized interests, such as unions and parties, on the process of reform. Our paper showed how particular forms of institutions— notably state-owned enterprises (SOEs) within the power sector as well as the state firm that supplies most fuels for electricity generation—shape the possibilities and pace of reform. The tight integration of these SOEs with the political elite, opaque systems for cost accounting, and schemes for siphoning state resources explain why these institutions have survived and why the progress of reform has been so slow. Private investors have only been allowed into the market at the margin through the “independent power producer (IPP)” scheme—an oxymoron, since the purchase agreements and dispatch rules determining payment to these IPPs are state dominated. “Strategic Behavior and the Pricing of Gas” (with D.L. Brito) looked at various models that address strategic behavior in the supply of gas under a netback pricing regulatory regime. We obtained three very strong technical results. First, the netback pricing rule used in Mexico might lead to discontinuities in Pemex’s revenue function. Second, when Pemex must pay for the gas it uses and flares, the value of the Lagrange multiplier associated with the gas processing constraint increases. Third, if the gas processing constraint is binding, forcing Pemex pay for the gas it uses and flares does not change the optimal short-run solution to the optimization problem, so it will have no impact on short-run behavior. “Price Regulation in a Vertically Integrated Natural Gas Industry: The Case of Mexico,” (with Dagobert L. Brito). The Comisión Reguladora de Energía of Mexico has implemented a netback rule for linking the Mexican natural gas price to the Texas natural gas benchmark price in an industry

9

transmisión depende, esencialmente, de las preferencias del inversionista y de los parámetros de preferencia de delegación. De igual manera, para una cantidad dada de FTR preexistentes, mientras mayor es la capacidad de corriente, tanto mayor es la necesidad de reservar algunos FTR para los posibles factores incidentales negativos generados por los cambios de expansión. “Diversos enfoques de la suficiencia del suministro en los mercados de energía eléctrica” (“Different Approaches to Supply Adequacy in Electricity Markets”), estudia el problema de largo alcance en el diseño de mercado energético de asegurar suficiente capacidad de generación para satisfacer la demanda futura (idoneidad de recursos). Los procesos de reforma en todo el mundo han mostrado que es difícil para el mercado, por sí solo, el proveer incentivos para atraer suficiente inversión en reservas de capacidad debido a las fallas institucionales y del mercado. Estudié varias medidas internacionales propuestas para enfrentar este problema, incluyendo reservas estratégicas, pagos de capacidad, requisitos de capacidad y opciones de compra. También discutí las fortalezas y debilidades analíticas y prácticas de cada enfoque. “El sector eléctrico mexicano: problemas económicos, legales y políticos (“The Mexican Electricity Sector: Economic, Legal and Political Issues”), Centro para la Ciencia y Política Ambiental, Stanford University, Informe de Trabajo 05 (con Víctor Carreón y Armando Jiménez), dirigido a explicar las motivaciones y estrategias para la reforma en el sector eléctrico mexicano. Nos enfocamos en los efectos de los intereses políticamente organizados, tales como los sindicatos y partidos políticos, sobre el proceso de reforma. Nuestro informe muestra cómo las formas particulares de instituciones, notablemente las empresas propiedad del estado (SOE, por sus siglas en ingles) dentro del sector de energía eléctrica, así como la firma estatal que suministra la mayoría de los combustibles para la generación de electricidad, les dan forma a las posibilidades y avance de la reforma. La estrecha integración de estas SOE con la élite política, los sistemas opacos de contabilidad de costos y los esquemas para drenar recursos estatales explican por qué estas instituciones han sobrevivido y por qué el progreso de reforma ha sido tan lento. A los inversionistas privados sólo se les ha permitido entrar al mercado al

10

Repsol YPF-Harvard Kennedy School Fellows Research and Program Summary

structure characterized by a vertically integrated state-owned monopoly. This paper shows that in an open economy where agents can choose between gas and alternative fuels, and where the density function describing the distribution of agents along the pipeline can have mass points, the netback rule is Pareto optimal. “Implications of the Elasticity of Natural Gas in Mexico on Investment in Gas Pipelines and in Setting the Arbitrage Point” (with D.L. Brito) studies the optimal timing of investment in gas pipelines when the demand for gas is stochastic. While this problem can be solved in theory, we showed that the practical solution depends on functions and parameters that are either subjective or cannot be estimated. We then reformulated the problem in a manner that can Pareto rank investment strategies. These strategies could be implemented with reasonably straightforward policies. The demand for gas is inelastic, and thus the welfare losses associated with small deviations from a first best optimum are minimal. This implies that the gas pipeline system can be regulated with a relatively simple set of rules without any significant loss of welfare. Regulation of the gas pipeline system can then be transparent and as a result, some institutional arrangement might be found to ensure a substantial private investment in gas pipelines. The Repsol YPF-Harvard Kennedy School Fellows program made possible the development of the described research through generous financial support, as well as through access to the impressive resources of the Kennedy School, including research information, conferences and seminars, and feedback from innumerable brilliant colleagues. My research under the program also motivates my research agenda for the next few years. Two of the topics that I plan to develop— and that I have started working on with Bill Hogan—include the welfare implications of longterm FTRs and the development of a combined merchant-regulatory mechanism for transmission expansion.

margen mediante el esquema de “productor independiente de energía eléctrica (IPP, por sus siglas en inglés), una contradicción de términos, puesto que los contratos de compra y las reglas de despacho que determinan el pago a estas IPP están dominadas por el estado. “Comportamiento estratégico y la determinación de los precios del gas” (“Strategic Behavior and the Pricing of Gas”), (con D.L. Brito), examinó diversos modelos que tratan el comportamiento estratégico en el suministro de gas bajo un régimen reglamentario de determinación de precios “netback”. Obtuvimos tres resultados técnicos muy fuertes. Primero, la regla de determinación de precios netback utilizada en México podría llevar a discontinuidades en la función de ingresos de PEMEX. Segundo, cuando PEMEX debe pagar por el gas que usa y quema, el valor del multiplicador de Lagrange asociado con la restricción del procesamiento de gas aumenta. Tercero, si la restricción de procesamiento de gas es vinculante, y obliga a PEMEX a pagar por el gas que usa y quema, no cambia la solución óptima de corto alcance al problema de optimización, de manera que no tenga consecuencias sobre el comportamiento a corto alcance. “Regulación de precios en una industria de gas natural integrada verticalmente: el caso de México” (“Price Regulation in a Vertically Integrated Natural Gas Industry: The Case of Mexico”), (con Dagobert L. Brito). La Comisión Reguladora de Energía de México ha implementado una regla netback para vincular el precio del gas natural mexicano con el precio de referencia del gas natural de Texas en una estructura industrial caracterizada por un monopolio estatal integrado verticalmente. Este informe muestra que en una economía abierta donde los agentes pueden elegir entre gas y combustibles alternos, y donde la función de densidad que describe la distribución de los agentes a lo largo de las tuberías puede tener puntos de amontonamiento, la regla de netback es óptima según el procedimiento de Pareto. “Implicaciones de la elasticidad del gas natural en México en la inversión en gasoductos y en la fijación del Punto de Arbitraje” (“Implications of the Elasticity of Natural Gas in Mexico on Investment in Gas Pipelines and in Setting the Arbitrage Point”), (con D.L. Brito), estudia la oportunidad optima de la inversión en gasoductos cuando la

Fellows 2003–2004

11

demanda de gas es estocástica. Aunque este problema puede, en teoría, resolverse, demostramos que la solución práctica depende de funciones y parámetros que son subjetivos o no pueden estimarse. Entonces reformulamos el problema de un modo que puede clasificar las estrategias de inversión usando el procedimiento de Pareto. Estas estrategias pueden ponerse en práctica con políticas razonablemente sencillas. La demanda de gas es inelástica, de manera que las pérdidas de bienestar asociadas con pequeñas desviaciones de un mejor óptimo posible son mínimas. Esto implica que el sistema de gasoducto puede ser reglamentado con un conjunto relativamente sencillo de reglas sin una pérdida de bienestar. La reglamentación del sistema de gasoductos puede ser entonces transparente y en consecuencia, podrían hallarse algunos arreglos institucionales para asegurar una inversión privada importante en los gasoductos. El programa Becarios de Repsol YPF-Harvard Kennedy School ha hecho posible el desarrollo de la investigación descrita mediante un apoyo financiero generoso, así como también a través del acceso a los extraordinarios recursos de la Kennedy School, que incluyen información de investigaciones, conferencias y seminarios, y las reacciones de un sin número de colegas brillantes. Mi investigación bajo el programa también motiva mi programa de investigación por los próximos años. Dos de los temas que pienso desarrollar y sobre los cuales he comenzado a trabajar con Bill Hogan, incluyen las implicaciones de bienestar de las FTR a largo plazo y cómo desarrollar un mecanismo combinado negociante-reglamentario para la expansión de la transmisión.

Joseph Aldy Ph.D. Candidate, Department of Economics, Harvard University

Joe Aldy is a doctoral candidate in the Department of Economics at Harvard University whose academic interests include environmental and public economics. His current research addresses the value of reducing mortality risk as related to age and the relationship between economic development and carbon dioxide emissions. Before coming to Harvard, Joe served on the staff of the President’s Council of Economic Advisers from 1997 to 2000 where his portfolio included a wide range of environmental and natural resource issues, including climate change policy, air quality regulations, world oil and refined petroleum markets, electricity restructuring, environmental issues in China, and sustainable development. He was the lead author for the 1998 report “The Kyoto Protocol and the President’s Policies to Address Climate Change: Administration Economic Analysis” and participated in bilateral and multilateral workshops and meetings on climate change policy in Argentina, Bolivia, China, France, Germany, Kazakhstan, Korea, Israel, Mexico, and Uzbekistan as well as at COP-4, COP-5, and the OECD. He was a Presidential Management Intern from 1996 to 1998. Joe received a Master of Environmental Management degree from the Nicholas School of the Environment in 1995 and a Bachelor of Arts degree from Duke University in 1993.

Joe Aldy es un candidato al doctorado en el Departamento de Economía en la Harvard University y cuyos intereses académicos incluyen la economía pública y del ambiente. Su investigación actual considera el valor de reducir los riesgos de mortalidad según se relacionan con la edad y la relación entre el desarrollo económico y las emisiones de bióxido de carbono. Antes de venir a Harvard, Joe sirvió en el estado mayor del Consejo de Asesores Económicos del Presidente desde 1997 hasta el 2000, donde su cartera incluyó una amplia gama de asuntos sobre recursos naturales y el medio ambiente, incluyendo la política relativa al cambio de clima, reglamentación de calidad del aire, mercados de petróleo y petróleo refinado mundiales, reestructuración de electricidad, asuntos del medio ambiente en China y desarrollo sostenible. Él fue el principal autor del informe de 1998 “El protocolo de Kyoto y las políticas del Presidente para responder al cambio climático: análisis económico de la administración” (“The Kyoto Protocol and the President’s Policies to Address Climate Change: Administration Economic Analysis”), participó en talleres bilaterales y multilaterales, y en reuniones en política relativa al cambio de clima en Argentina, Bolivia, China, Francia, Alemania, Kazajstán, Corea, Israel, México y Uzbekistán, así como en COP-4, COP-5 y la OCDE. Fue un Presidential Management Intern desde 1996 hasta 1998. Joe recibió una Maestría en Gerencia del Ambiente de la Nicholas School of Environment en 1995, y una licenciatura en Filosofía de la Duke University en 1993.

Mr. Aldy Reviews His Year as a Repsol YPF-Harvard Kennedy School Fellow Over the past year, I have completed drafts of two papers based on my research on the relationship between carbon dioxide emissions and economic development: 1. 2.

El Sr. Aldy Refiere Su Año Como Becario de Repsol YPF-Harvard Kennedy School

Divergence in Per Capita Carbon Dioxide Emissions An Environmental Kuznets Curve Analysis of U.S. State-Level Carbon Dioxide Emissions

Durante el año pasado, he completado los borradores de dos informes de investigación basados en mi investigación sobre la relación entre las emisiones de bióxido de carbono y el desarrollo económico: 13

14

Repsol YPF-Harvard Kennedy School Fellows Research and Program Summary

The Divergence paper addressed two key questions: Have per capita carbon dioxide emissions been converging in the past? and, Should we expect them to converge in the future? Understanding the distribution of emissions is important in determining countries’ incentives to participate in multilateral climate change policies. Developing countries may not adopt emissions obligations that they perceive as inequitable given that developed countries have per capita emissions that are one to two orders of magnitude greater. Developed countries may not agree to allocation rules, such as per capita allocation schemes, designed to appeal to developing countries if the distribution of emissions differs significantly from the distribution of population. The Divergence paper uses publicly available data for carbon dioxide emissions from 88 countries and carbon dioxide emissions data constructed by the author for the 50 U.S. states. The analysis shows the lack of convergence in per capita carbon dioxide emissions among nations (between 1960 and 2000) and the divergence in emissions among the U.S. states (between 1960 and 1999). A special analysis of U.S. state-level emissions focuses on the difference between production-based and consumption-based emissions by accounting for the net effects of interstate electricity trade. The increasing dispersion in statelevel per capita emissions over time is much larger with the production-based measure than with the consumption-based measure. This suggests that some states are exporting emissions-intensive industrial activity and importing the finished goods and services from these industries. The paper forecasts future distributions of per capita carbon dioxide emissions for the 88-country data set and the U.S. states data set, and shows very little convergence over the next 50 to 100 years. It concludes with an analysis of the shortcomings of reduced-form parametric models (environmental Kuznets curve analyses) and structural models for forecasting future distributions of emissions. This paper should address the gap in the existing literature by providing the first characterization of the distribution of carbon dioxide emissions. In addition, it should help inform policymakers who have advocated for per capita emissions allocations without a solid understanding of past, current, or future distributions of emissions.

1. 2.

Divergencia en las emisiones de bióxido de carbono per cápita. Un análisis de curva de Kuznets ambiental del nivel de emisiones de bióxido de carbono por estado en los EE.UU.

El informe de Divergencia trató dos asuntos claves: ¿Han estado convergiendo en el pasado las emisiones de bióxido de carbono per cápita? Y, ¿debemos esperar que converjan en el futuro? Comprender la distribución de las emisiones es importante para determinar los incentivos de los países a participar en las políticas multilaterales relativas al cambio de clima. Los países en desarrollo pueden no adoptar obligaciones de emisiones que perciban como injustas dado que los países desarrollados tienen emisiones per cápita que son una a dos veces mayores. Los países desarrollados pueden no convenir en las reglas de asignación, tales como los esquemas de asignación per cápita diseñados para atraer a los países en desarrollo si la distribución de emisiones difiere considerablemente de la distribución de la población. El informe de Divergencia utiliza datos accesibles al público para las emisiones de bióxido de carbono de 88 países y datos de emisiones de bióxido de carbono construidos por el autor para los 50 estados de los EE.UU. El análisis muestra la falta de convergencia en las emisiones de bióxido de carbono per cápita entre las naciones (entre 1960 y el 2000) y la divergencia en las emisiones entre los estados de los EE.UU. (entre 1960 y 1999). Un análisis especial de las emisiones en el ámbito estatal en los EE.UU. enfoca la diferencia entre las emisiones basadas en la producción y las basadas en el consumo mediante una explicación de los efectos netos del comercio eléctrico interestatal. La creciente dispersión en las emisiones per cápita en el ámbito estatal con el tiempo es mucho mayor con la medida basada en la producción que la medida basada en el consumo. Esto sugiere que algunos estados están exportando actividad industrial que es emisión-intensiva e importando los bienes y servicios terminados de estas industrias. El informe predice distribuciones futuras de las emisiones per cápita de bióxido de carbono para el conjunto de datos de 88 países y para el conjunto de datos de los estados en los EE.UU., y muestra poca convergencia por los próximos 50 a 100 años. Concluye con un análisis de las carencias de modelos paramétricos de forma reducida (análisis de curva Kuznets del

Fellows 2003–2004

The EKC paper uses the same U.S. state-level carbon dioxide emissions data to evaluate whether per capita CO2 emissions follow an inverted-U shape with respect to per capita income. The standard story about the environmental Kuznets curve is that emissions are low at low levels of economic development, but increase as an economy develops (e.g., transitions from agriculture to manufacturing). At some point, as the economy grows further and becomes based on services, emissions begin to decrease. This paper takes aim at three issues: (1) What is the role of trade in emissions-intensive goods? (2) What effect do climate and historic energy endowments have on an economy’s per capita emissions? and, (3) How robust are the estimated environmental Kuznets curves to various econometric specifications? The EKC paper shows that trade in emissionsintensive goods (in this case, electricity) impacts estimated EKC curves significantly. EKC curves peak at higher incomes for consumption-based EKC curves (post-trade) than for productionbased EKC curves (pre-trade), and in some specifications the consumption-based models do not have in-sample peak incomes (which suggests that the income-emissions relationship may not take an inverted-U shape). Models that account for winter heating and summer cooling demand as well as historic endowments of coal, crude oil, and hydropower show that these all impact per capita emissions, but usually modestly (for most states, they account for less than ten percent of per capita emissions). The states carbon dioxide EKC models are not especially robust to various specifications, which suggests that future work should aim to expand the analysis to structural models that can better explain the relationship between development and emissions. The EKC paper has been accepted by the Environmental Protection Agency for inclusion in a special issue of the Journal of Environment and Development, arranged by the EPA on cover applications of the environmental Kuznets curve to greenhouse gas emissions. This special issue will aim to provide additional guidance to the EPA and other related policymakers on the relationship between greenhouse gas emissions and economic development. I presented preliminary versions of these papers to the Energy Policy Fellowship seminar in

15

ambiente) y modelos estructurales para predecir distribuciones futuras de las emisiones. Este informe debe cubrir la brecha en la literatura actual al dar la primera caracterización de la distribución de las emisiones de bióxido de carbono. Además, debe ayudar a informar a los responsables de formular políticas que han abogado por asignaciones per cápita de las emisiones sin una comprensión sólida de las distribuciones pasadas, actuales o futuras de las emisiones. El informe de Curva de Kuznets Ambiental (EKC, por sus siglas en inglés), utiliza los mismos datos de emisiones de bióxido de carbono en el ámbito estatal de los EE.UU. para evaluar si las emisiones per cápita de emisiones de CO2 siguen una forma U invertida con respecto al ingreso per cápita. La historia estándar acerca de la curva Kuznets ambiental es que las emisiones son bajas a bajos niveles de desarrollo económico, pero aumentan a medida que la economía se desarrolla (por ejemplo, cambio de agricultura a manufactura). En algún punto, a medida que la economía crece más y se convierte en una economía de servicios, las emisiones comienzan a disminuir. Este informe tiene como objetivos tres asuntos: (1) ¿cuál es el papel del comercio en los bienes emisión-intensivos? (2) ¿Qué efecto tienen el clima y las dotaciones históricas de energía en las emisiones per cápita de una economía? Y, (3) ¿qué tan robustas son las curvas de Kuznets ambientales aproximadas ante distintas especificaciones econométricas? El informe EKC muestra que el comercio en bienes emisión-intensivos (en este caso, la electricidad) tiene consecuencias importantes sobre las curvas EKC estimadas. Las curves EKC llegan a su ápice en ingresos más altos para curvas EKC basadas en el consumo (posteriores al comercio) que para las curvas EKC basadas en la producción (pre comercio), y en algunas especificaciones los modelos basados en el consuno no tienen ingresos máximos en la muestra (lo que sugiere que la relación de ingresos a emisiones pudiese no tomar una forma de U invertida). Los modelos que toman en cuenta la demanda de calefacción en el invierno y el acondicionamiento de los ambientes en verano, así como la dotación histórica de carbón, petróleo crudo, y energía hidroeléctrica muestran que todas éstas afectan las emisiones per cápita, pero por lo general muy poco (para la mayoría de los estados, representan menos del diez por ciento

16

Repsol YPF-Harvard Kennedy School Fellows Research and Program Summary

the fall of 2003 and to the Environmental Economics Program at Harvard University lunch seminar in the Spring of 2004. I plan to propose a presentation of this work at Camp Resources, an annual environmental economics workshop sponsored by North Carolina State University (Kerry Smith’s workshop). The fellowship provided two primary benefits. First, the stipend allowed me to focus exclusively on my research this year. The financial support freed me of teaching responsibilities that can divert time and attention from research. Second, the opportunity to present early stages of my research to the other fellows and faculty associated with the program yielded valuable insights that will guide the next steps in the research. Follow-up conversations with the other fellows also assisted my research.

de las emisiones per cápita). Los modelos estatales de EKC de bióxido de carbono no son particularmente fuertes para varias especificaciones, lo que sugiere que el trabajo futuro debe tener como objetivo ampliar el análisis a modelos estructurales que puedan explicar mejor la relación entre el desarrollo y las emisiones. El informe EKC ha sido aceptado por la Agencia de Protección Ambiental (EPA, por sus siglas en ingles) para ser incluido en una edición especial del Journal of Environment and Development, en aplicaciones de cobertura de la curva ambiental de Kuznets a las emisiones de gases de invernadero. Esta publicación especial tendrá el objetivo de proveer orientación adicional a la EPA y otras personas y entidades responsables de formular políticas sobre la relación entre la emisión de los gases de invernadero y el desarrollo económico. Presenté versiones preliminares de estos informes al seminario de la Sociedad de Académicos de Políticas de Energía en el otoño de 2003 y al Programa de Economía del Ambiente en el seminario-almuerzo de la Harvard University en la primavera de 2004. Pienso proponer una presentación de este trabajo en Camp Resources, un taller de economía del ambiente patrocinado por la Universidad del Estado de Carolina del Norte (Taller de Kerry Smith). La beca ofreció dos beneficios fundamentales. Primero, el estipendio me permitió enfocarme exclusivamente en mi investigación este año. El apoyo financiero me liberó de las responsabilidades de enseñanza que pueden restar tiempo y atención a la investigación. Segundo, la oportunidad de presentar etapas tempranas de mi investigación a los otros académicos y facultad asociados con el programa produjo valiosas ideas que guiarán los próximos pasos en la investigación. Conversaciones de seguimiento con los otros becarios también ayudaron a mi investigación.

Darby Jack Ph.D. Candidate in Public Policy, Kennedy School of Government, Harvard University

Darby Jack is a doctoral candidate in Public Policy at the Kennedy School. His research centers on the economic analysis of human-environment interactions in developing countries. Currently, he is analyzing the determinants of household energy technology choices by poor families in Latin America and the linkages between energy, indoor air pollution, and human health. Darby received a bachelor’s degree from Williams College in 1997. After college, he spent two years analyzing strategies to promote sustainable forest management in Guatemala, Chile, and Bolivia as a Watson Foundation Fellow. Darby then worked for the Mountain Institute in Huaraz, Peru, and helped start a consultancy that advises landowners and conservationists in Latin America on matters related to climate change. Darby has received several awards and fellowships, including the Thomas Hardie Prize from Williams College, a Joseph Crump Fellowship from Harvard University, and the Watson Fellowship. At Harvard he is affiliated with the Center for International Development, the Belfer Center for Science and International Affairs and the Environmental Economics Program.

Darby Jack es un candidato al doctorado en Política Pública en la Kennedy School. Su investigación se centra en el análisis económico de las interacciones humanas con el ambiente en los países en desarrollo. Actualmente, está analizando lo que determina las elecciones de tecnología energética del hogar hechas por las familias pobres en Latinoamérica y las relaciones entre la energía, la contaminación en el interior de las edificaciones, y la salud. Darby recibió una licenciatura de Williams College en 1997. Después de la universidad, pasó dos años analizando las estrategias para promover la administración sostenible de los bosques en Guatemala, Chile y Bolivia como un becario de la Watson Foundation. Darby trabajó después para el Instituto de Montaña en Huaraz, Perú, y ayudó a iniciar una consultoría que asesora a los propietarios de tierras y a los conservacionistas en Latinoamérica sobre asuntos relativos a los cambios del clima. Darby ha recibido varios premios y becas, que incluyen el Thomas Hardie Prize de Williams College, una beca Joseph Crump de la Harvard University, y la beca Watson. En Harvard está afiliado al Centro para el Desarrollo Internacional, el Centro Belfer para la Ciencia y Asuntos Internacionales y el Programa de Economía del Ambiente.

Mr. Jack Reviews His Year as a Repsol YPF-Harvard Kennedy School Fellow

El Sr. Jack Refiere Su Año Como Becario de Repsol YPF-Harvard Kennedy School

My research analyzes economic factors that shape how people in developing countries choose how to meet their household energy needs. The primary motivation for my work is to illuminate the economic antecedents of indoor air pollution in developing countries, a problem that kills over 1.6 million people each year according to World Health Organization (WHO) estimates. To put that figure in context, WHO estimates that HIV/AIDS kills about 2.8 million people each year, dirty drinking water about 1.7 million, and urban air pollution about 0.8 million. Virtually all indoor airrelated deaths in developing countries occur from

Mi investigación analiza los factores económicos detrás de las elecciones que hace la gente en los países en desarrollo sobre cómo satisfacer sus necesidades hogareñas de energía. La motivación principal para mi trabajo es aclarar las causas económicas de la contaminación dentro de las edificaciones en los países en desarrollo, un problema que mata a más de 1.6 millones de personas al año según la Organización Mundial de la Salud (OMS). Para poner esa cifra en su contexto, la OMS estima que el VIH y el SIDA matan cerca de 2.8 millones 17

18

Repsol YPF-Harvard Kennedy School Fellows Research and Program Summary

cooking with wood and other biomass fuels in poorly ventilated rooms. My empirical approach uses longitudinal data from Peru to explain household energy technology choices and to provide a reliable basis for formulating policies that will influence these decisions. I am concerned with a simple question: Why do people with alternatives use household energy technologies that kill them? This research also improves our understanding of the transition from traditional to modern energy carriers and thus improves our ability to predict future household energy demand in India, China, and other biomass-dependent regions. Finally, insight into energy use by poor households sheds empirical light on persistent questions about economic behavior in developing countries. The rational choice perspective presumes that households invest in clean household energy technologies until the benefits of cleaner air would be more than offset by the loss of utility due to the additional cost of cleaner technology. This behavioral postulate—the poor but efficient hypothesis— constitutes the theoretical foundation of the handful of previous papers on the determinants of energy technology choice in poor households. My research so far suggests that this hypothesis fails to explain the facts. The data show that income does not predict fuel choice, once we control for location, access to fuels, and other relevant household characteristics (using a random effects logit model). This surprising result contradicts findings from South Africa, Pakistan, and other countries where researchers have found significant correlations between income and fuel use. These other studies have suffered from a lack of longitudinal data, inadequate controls, and inappropriate statistical frameworks. Additional research will elucidate the cause and implications of this discrepancy, but at present I interpret my results as evidence against the poor but efficient hypothesis. My research also investigates two ways in which actual behavior may depart from a simple household demand model, and weighs the policy implications of these departures. Features of the decision context, including uncertainty regarding consequences and interactions between decisionmakers, suggest that the optimal decision hypothesis may miss important elements of the true decision process. These concerns lead to two addi-

de personas cada año, el agua de beber sucia cerca de 1.7 millones y la contaminación del aire en el medio urbano cerca de 0.8 millones. Virtualmente, todas las muertes relacionadas con el aire que ocurren dentro de las edificaciones en los países en desarrollo son causadas por cocinar con madera y otros combustibles de biomasa en habitaciones mal ventiladas. Mi enfoque empírico utiliza datos longitudinales del Perú para explicar las elecciones de tecnología energética hogareña y para dar una base confiable a la formulación de políticas que influirán esas decisiones. Me preocupa una pregunta sencilla: ¿por qué la gente con alternativas utiliza tecnologías energéticas hogareñas que la matan? Esta investigación también mejora nuestra comprensión de la transición de portadores de energía tradicionales a los modernos y mejora así nuestra capacidad para predecir la demanda de energía hogareña en India, China y otras regiones dependientes de la biomasa. Finalmente, la comprensión del uso de la energía por los hogares pobres aclara, empíricamente, las cuestiones acerca del comportamiento económico en los países en desarrollo. La perspectiva de la elección racional supone que los hogares invierten en tecnologías de energía hogareña limpia hasta que los beneficios de un aire más limpio compensen abundantemente la pérdida de utilidad debida al costo adicional de la tecnología más limpia. El postulado conductista, la hipótesis de pobre, pero eficiente, constituye la fundación teórica de los pocos informes anteriores sobre lo que determina la elección de tecnología energética en los hogares pobres. Mi investigación sugiere hasta ahora que esta hipótesis no explica bien los hechos. Los datos muestran que los ingresos no predicen la elección de combustible, una vez que controlamos la ubicación, el acceso a combustibles y otras características importantes del hogar (utilizando un modelo logit de efectos aleatorios). Este sorprendente resultado contradice las conclusiones extraídas de África del Sur, Pakistán y otros países donde los investigadores han hallado importantes correlaciones entre ingresos y utilización de combustible. Estos otros estudios han sufrido de una falta de datos longitudinales, controles inadecuados y una estructura estadística inapropiada. La investigación adicional aclarará la causa e implicaciones de esta discrepancia, pero ahora interpreto mis

Fellows 2003–2004

tional hypotheses regarding the basic economic structure of energy technology choices. The information hypothesis. People may be ignorant of the health consequences of dirty household energy. This is particularly plausible in areas where all or nearly all families use dirty technologies, and most woman and children suffer from chronic respiratory infections. In the absence of a “control group” of clean technology users, people may be slow to draw inferences about the consequences of indoor air pollution. I study the role of information by estimating a Bayesian social learning model in which poor households learn about the health consequences of available technologies by observing their neighbors’ technology choices and health outcomes. Preliminary results show that, ceteris paribus, households that observe that their neighbors experience health benefits after switching to cleaner fuels are more likely to switch than households that observe switching but see no health improvement. This suggests that information about health consequences is a significant factor, and that households lack complete information. I am currently developing a more refined model that accounts for the dynamics of the learning process and thus better controls for spurious correlation. The family structure hypothesis. The optimal decision hypothesis views households as unitary actors. In reality, households comprise individuals with different stakes and different abilities to affect joint decisions. Women and children, who spend more time indoors near the hearth, bear the brunt of the health impacts of indoor air pollution. Men, in contrast, are less affected by indoor air pollution but tend to control household budgets. This mismatch may lead to underinvestment in clean energy technologies. I test this hypothesis by studying how the probability of switching to a cleaner technology changes in response to changes in men’s and women’s incomes.A major goal of my research is to help lay the foundation for public policy that is genuinely welfare-enhancing and that constitutes an intelligent response to preferences and decision-processes of the rural poor. Thus the final stage of my research is to analyze my findings against a range of policy interventions that purport to reduce exposure to indoor air pollution. This analysis will feed into an outreach effort targeting government policymakers in Peru.

19

resultados como evidencia contra la hipótesis “pobre pero eficiente”. Mi investigación también sondea dos maneras en las cuales el comportamiento real puede desviarse de un modelo simple de demanda hogareña, y considera las implicaciones de política de estas desviaciones. Las características del contexto de decisión, incluyendo la incertidumbre concerniente a las consecuencias y las interacciones entre las personas responsables de las decisiones, sugieren que la hipótesis de la decisión óptima puede excluir erradamente elementos importantes del verdadero proceso de decisión. Estas preocupaciones llevan a dos hipótesis adicionales concernientes a la estructura económica básica de las elecciones de tecnología energética. La hipótesis de la información. La gente puede ignorar las consecuencias para la salud de la energía hogareña sucia. Esto es particularmente posible en las áreas en las cuales todas o casi todas las familias utilizan tecnologías sucias y la mayoría de las mujeres y niños sufren de infecciones respiratorias crónicas. En ausencia de un “grupo de control” de usuarios de tecnología limpia, la gente puede ser lenta para sacar conclusiones acerca de las consecuencias de la contaminación del aire dentro de las edificaciones. Yo estudio el rol de la información estimando un modelo bayesiano de aprendizaje social en el cual los hogares pobres aprenden las consecuencias sobre la salud de las tecnologías disponibles observando las elecciones de tecnología de sus vecinos y los resultados sobre la salud. Los resultados preliminares muestran que, ceteris paribus, los hogares que observan que sus vecinos experimentan beneficios de salud después de cambiarse a combustibles más limpios tienen una mayor tendencia a cambiar que los hogares que observan el cambio, pero no ven una mejora en la salud. Esto sugiere que la información acerca de las consecuencias sobre la salud es un factor importante, y que los hogares carecen de información completa. Actualmente estoy desarrollando un modelo más refinado que explica la dinámica del proceso de aprendizaje y de este modo controla mejor la correlación espuria. La hipótesis de la estructura familiar. La hipótesis de decisión óptima contempla a los hogares como actores unitarios. Realmente, los hogares comprenden individuos con diferentes intereses y

20

Repsol YPF-Harvard Kennedy School Fellows Research and Program Summary

The Repsol YPF-Harvard Kennedy School Fellows Program gave me the opportunity to present my research in our seminar; I enjoyed and learned from the interchanges with the other Fellows as well as with Bill Hogan and Adrian Lajous on these occasions. Further, I very much appreciated the funding for travel in support of my research.

capacidades para afectar decisiones conjuntas. Las mujeres y los niños, que pasan más tiempo en el hogar cerca del fuego, son los más afectados por las consecuencias de la contaminación dentro de las edificaciones. Los hombres, en contraste, se ven menos afectados por la contaminación dentro de las edificaciones, pero tienden a controlar el presupuesto del hogar. Esta disparidad tiende a llevar a una inversión inadecuada en las tecnologías de energía limpia. Pruebo esta hipótesis estudiando cómo la probabilidad de cambiar a tecnologías más limpias cambia en respuesta a los cambios en los ingresos de los hombres y las mujeres. Una meta importante de mi investigación es ayudar a establecer las bases para la política pública que genuinamente mejore el bienestar y constituya una respuesta inteligente a las preferencias y procesos de toma de decisiones de la gente pobre en el medio rural. Así, la etapa final de mi investigación fue analizar mis conclusiones contra una variedad de intervenciones de política que pretenden reducir la exposición a la contaminación dentro de las edificaciones. Este análisis contribuirá a un esfuerzo de promoción de servicios sociales y de salud dirigido a los responsables de formular las políticas en el Perú. El Programa Becarios de Repsol YPF-Harvard Kennedy School me dio la oportunidad de presentar mi investigación en nuestro seminario; disfruté y aprendí de los intercambios con los otros académicos así como con Bill Hogan y Adrian Lajous en estas ocasiones. Además, aprecio mucho el financiamiento para viajes en apoyo a mi investigación.

Cynthia Lin Ph.D. Candidate, Department of Economics, Harvard University

Cynthia Lin is a doctoral candidate in Economics at Harvard University. She is interested in applying the theories and methods used in industrial organization, microeconomic theory, public economics, and labor economics to environmental issues. For her Ph.D. dissertation, she is researching offshore petroleum production. Cynthia received a bachelor’s degree, summa cum laude, in Environmental Science and Public Policy from Harvard College in 2000. Her undergraduate atmospheric chemistry thesis on trends in ozone smog was awarded a Thomas Temple Hoopes Prize. She was elected to Phi Beta Kappa in her junior year. In addition to the Repsol YPF–Harvard Kennedy School Pre-Doctoral Fellowship, Cynthia’s graduate honors include an EPA Science To Achieve Results (STAR) Fellowship, a National Science Foundation (NSF) Graduate Research Fellowship, a Rita RicardoCampbell Fellowship in Economics, a Jens Aubrey Westengard Scholarship, and a Harvard Committee on Undergraduate Education (CUE) Certificate of Distinction in Teaching. Cynthia is currently a Pre-Doctoral Fellow in the Environmental Economics Program at Harvard University.

Cynthia Lin es una candidata al doctorado en Economía en la Harvard University. Ella está interesada en aplicar las teorías y métodos utilizados en las organizaciones industriales, la teoría microeconómica, las economías públicas y la economía laboral a los problemas ambientales. Para su disertación de Ph.D, ella está investigando la producción petrolera mar afuera. Cynthia recibió una licenciatura, summa cum laude, en Ciencias del Ambiente y Política Pública de Harvard College en 2000. Su tesis de pregrado en química atmosférica sobre las tendencias de la contaminación por ozono fue premiada con el Thomas Temple Hoopes Prize. Fue elegida a Phi Beta Kappa en su tercer año de estudios. Además de la beca de pre-doctorado Repsol YPF–Harvard Kennedy School, los honores de Cynthia incluyen una beca EPA Science To Achieve Results (STAR), una beca para investigación de posgrado de la National Science Foundation (NSF), una beca Rita Ricardo-Campbell en Economía, una beca Jens Aubrey Westengard, y un Certificado de Distinción en la Docencia del Comité de Harvard en Educación de Pregrado (CUE, por sus siglas en inglés). Actualmente Cynthia es una Profesora con nivel de pre-doctorado en el Programa de Economía del Ambiente en la Harvard University.

Ms. Lin Reviews Her Year as a Repsol YPF-Harvard Kennedy School Fellow I am interested in applying the theories and methods used in industrial organization, microeconomic theory, and econometrics to analyze the oil industry. I am currently pursuing three main research projects related to the oil industry. The first research project is on offshore petroleum production. For this project, I developed a structural econometric model to analyze the investment timing game in offshore petroleum production that ensues on wildcat tracts in U.S. federal lands off the Gulf of Mexico. Offshore petroleum production requires two irreversible investment decisions. First, a firm decides whether and when to invest in the rigs

El Srta. Lin Refiere Su Año Como Becario de Repsol YPF-Harvard Kennedy School Estoy interesada en aplicar las teorías y métodos utilizados en las organizaciones industriales, la teoría microeconómica, y la econometría al análisis de la industria petrolera. Actualmente trabajo en tres proyectos de investigación relacionados con la industria del petróleo. El primer proyecto de investigación es sobre la producción petrolera mar afuera. Para este proyecto, desarrollé un modelo estructural econométrico para analizar el juego de oportunidad de inversiones en la producción petrolera 21

22

Repsol YPF-Harvard Kennedy School Fellows Research and Program Summary

needed to begin exploratory drilling. Second, after it has explored a tract, a firm decides whether and when to invest in the production platforms needed to develop and extract the reserve. When individual firms make their exploration and development investment timing decisions, there are two types of externalities that they do not internalize. The first type is an information externality. Because tracts within the same area may be located over a common pool, firms acquire information about their own tracts when other firms explore and develop neighboring tracts. This information externality may lead to a non-cooperative timing game, which may cause firms to inefficiently delay production. A second type of externality is an extraction externality. When firms have competing rights to a common-pool resource, extraction may occur at an inefficiently high rate. While information externalities induce firms to inefficiently delay production, extraction externalities induce firms to produce too quickly. Thus, owing to information and extraction externalities, offshore petroleum production on wildcat tracts may be inefficient. Having completed the theoretical work during the period of the fellowship, the next stage is estimating structural econometric model parameters for the investment timing game in offshore petroleum production. Estimates of the structural parameters of my model will enable me to answer the following questions, among others. First, do firms care about what their neighbors do? In other words, how important are the information and extraction externalities described above? Second, can the federal government increase ex ante tract value by changing the lease term or the tract size? The primary goal of this research will be to determine the socially efficient timing policy for oil exploration, development, and extraction on offshore wildcat tracts. Using the optimal policy as a benchmark, one can then quantify the inefficiencies of the current Outer Continental Shelf (OCS) leasing program as well as the potential inefficiencies of various counterfactual institutional mechanisms. A second goal of this research project will be to design an institutional mechanism that induces a more socially efficient policy. Possible modifications to the current OCS auction mechanism might include: changing the lease terms; encouraging unitization of exploration programs, perhaps by limiting the

mar afuera que tiene lugar en áreas de perforaciones exploratorias o de prueba en las tierras federales de los EE.UU. costa afuera en el Golfo de México. La producción petrolera costa afuera exige dos decisiones irreversibles de inversión. Primero, una firma decide si, y cuándo invertir en las plataformas necesarias para comenzar la perforación exploratoria. Segundo, después de que ha explorado un área, una firma decide si, y cuándo invertir en las plataformas de producción necesarias para desarrollar y extraer la reserva. Cuando las firmas individuales toman sus decisiones de oportunidad de inversiones de exploración y desarrollo, hay dos tipos de circunstancias incidentales que no interiorizan. El primer tipo es un factor incidental de información. Debido a que las áreas dentro de la misma área pueden estar ubicadas sobre un yacimiento común, las firmas adquieren información acerca de sus propias áreas cuando las otras firmas exploran y desarrollan áreas vecinas. Este factor incidental de información puede llevar a un juego de oportunidad no cooperativo, que puede causar que las firmas demoren la producción ineficientemente. Un segundo tipo de factor incidental es de extracción. Cuando las firmas tienen derechos competitivos a un recurso de un yacimiento común, la extracción puede tener lugar a una tasa ineficientemente alta. Mientras los factores incidentales de información inducen a unas firmas a demorar ineficientemente la producción, los factores incidentales de extracción inducen a otras firmas a producir muy rápidamente. Así, debido a los factores incidentales de información y extracción, la producción petrolera costa afuera sobre áreas no exploradas y de prueba puede ser ineficiente. Habiendo completado el trabajo teórico durante la beca, la siguiente etapa es estimar los parámetros estructurales del modelo econométrico para el juego de oportunidad de inversión en la producción petrolera costa afuera. Los estimados de los parámetros estructurales de mi modelo me permitirán responder las siguientes preguntas, entre otras. Primero, ¿les importa a las empresas lo que hacen sus vecinos? En otras palabras, ¿qué tan importantes son la información y los factores incidentales de extracción descritos arriba? Segundo, ¿puede el gobierno federal aumentar el valor existente del área cambiando el plazo del contrato de arrendamiento o el tamaño del área?

Fellows 2003–2004

amount of non-unitized acreage that a firm can possess; requiring firms to make their seismic reports publicly available; changing the quantity, size, or location of the tracts offered in each lease sale; using multi-unit auctions; or making the contractual environment more conducive to coordination. A mechanism that induces socially optimal offshore petroleum production will not only increase efficiency and potentially enhance both firm and government revenues, but will also better account for environmental and strategic concerns as well. In addition to my work on offshore petroleum production, I pursued a second project, developing an econometric model of the world oil market. As with any other commodity, one of the fundamental questions economists would want to address about oil is: How do we model the world market for oil? In particular, what determines the supply for oil, what determines the demand for oil, and by what equilibrium process are oil prices and quantities determined? Although there have been countless empirical studies of the world oil market, none has produced a satisfactory model that adequately explains historical data, much less accurately predicts future developments. Moreover, the preponderance of these studies were conducted over two decades ago. For this research project, I re-examined the timeless issue of oil supply and demand estimation using instrumental variables techniques under a static framework. I am currently developing an econometric model of the world oil market that accounts for the dynamic nature of oil supply. In addition to my empirical work on offshore petroleum production and world oil markets, I am pursuing a third research project on the development of a theoretical model of oil extraction that is consistent with historical data on world oil prices and production. The basic Hotelling model of nonrenewable resource extraction predicts that the shadow price of the resource stock, which is an economic measure of the scarcity of the resource, should grow at the rate of interest. This prediction is now known as the “Hotelling rule.” If the natural resource market is perfectly competitive, the Hotelling rule implies that the market price minus marginal costs must grow at the rate of interest, and therefore the natural resource price should increase over time if marginal costs are constant. In

23

La principal meta de esta investigación será determinar la política de oportunidad socialmente eficiente para la exploración, desarrollo y extracción en áreas de prueba o exploración costa afuera. Utilizando como punto de referencia la política óptima, uno puede entonces cuantificar las ineficiencias del programa de arrendamiento actual de la Plataforma Continental Exterior (OCS, por sus siglas en inglés), así como también las ineficiencias potenciales de diversos mecanismos institucionales contrarios a los hechos. Una segunda meta de este proyecto de investigación será diseñar un mecanismo institucional que induzca a una política social más eficiente Las modificaciones posibles al mecanismo de subastas actual de OCS podrían incluir: cambiar los términos del contrato de arrendamiento; alentar la unificación de los programas de exploración, limitando quizá el área no dividida que una firma pueda poseer; exigir a las firmas que hagan de acceso publico sus informes sísmicos; cambiar la cantidad, tamaño o ubicación de las áreas ofrecidas en cada venta de contratos de arrendamiento; utilizar subastas multi-unidades; o hacer el ambiente contractual más conducente a la coordinación. Un mecanismo que induce a la producción petrolera costa afuera socialmente óptima, no sólo aumentará la eficiencia y mejorará potencialmente los ingresos de tanto la empresa como del gobierno, sino que también tomará en cuenta inquietudes estratégicas y ambientales. Además de mi trabajo en la producción petrolera costa afuera, continuo con un segundo proyecto, la creación de un modelo econométrico del mercado mundial del petróleo. Al igual que con cualquier otro producto básico, una de las preguntas fundamentales que los economistas querrían responder con respecto al petróleo es: ¿Cómo hacemos un modelo del mercado mundial para el petróleo? En particular, ¿qué determina el suministro de petróleo, qué determina la demanda de petróleo, y mediante cuáles procesos de equilibrio se determinan los precios y las cantidades de petróleo? Aunque han habido innumerables estudios empíricos del mercado mundial del petróleo, ninguno ha producido un modelo satisfactorio que explique adecuadamente los datos históricos, mucho menos pronostique los desarrollos futuros. Por otra parte, la mayoría de estos estudios se

24

Repsol YPF-Harvard Kennedy School Fellows Research and Program Summary

contrast to Hotelling’s theoretical prediction, however, empirical studies have shown that mineral prices have been roughly trendless over time. For my third research project, I am attempting to reconcile Hotelling’s theoretical model with empirical evidence on world mineral prices. I have benefited from the Repsol YPF-Harvard Kennedy School Pre-Doctoral Fellowship in numerous ways. The Repsol fellowship put me in touch with many people who have been invaluable to my research. Though the contacts I have been fortunate enough to make are too numerous to list, several deserve special mention. First, Adrian Lajous, who regularly attended the Repsol seminars, not only provided me with valuable feedback on my work, but also provided helpful information about the OCS auctions. Second, Juan Badosa Pagés took the time to talk with me about the role of risk in oil production. Third, William Hogan has been extremely supportive of my research; he not only takes the time to give detailed comments on my work, but has also been extremely helpful in helping me acquire the data, connections, and information I need. Fourth, and perhaps most important, it was through the Repsol YPF Fellowship that I met Bijan Mossavar-Rahmani, who arranged for my January visit to Apache Corporation’s headquarters in Houston and a drilling rig and production platform offshore of Louisiana. The trip was an invaluable experience for me; the knowledge I gained and connections I made during the visit have not only helped me with my current research projects, but have also given me ideas for several other projects I hope to pursue in the near future. The Repsol YPF Fellowship program has provided me with a wealth of resources, contacts, and information that have and will continue to enhance my efforts to explore and develop research on the economics of oil. As an aspiring academic economist, I have benefited tremendously from the opportunity to meet and learn from people directly involved in the oil industry, and my experiences as a Repsol YPF Fellow have better equipped me to apply my knowledge of academic economics to real-world policy-relevant issues relating to energy and petroleum.

realizaron hace más de dos décadas. Para este proyecto de investigación, he reexaminado el eterno asunto del suministro de petróleo y la estimación de la demanda mediante el uso de técnicas variables instrumentales bajo una estructura estática. Actualmente estoy desarrollando un modelo econométrico del mercado petrolero mundial que explique la naturaleza dinámica del suministro de petróleo. Además de mi trabajo empírico sobre la producción petrolera mar afuera y los mercados mundiales de petróleo, estoy ejecutando un tercer proyecto de investigación sobre el desarrollo de un modelo teórico de extracción de petróleo que sea consistente con los datos históricos sobre los precios y la producción mundial del petróleo. El modelo básico Hotelling de extracción de recursos no renovables predice que el precio sombra del recurso de materia prima, que es una medida económica de la escasez del recurso, debe crecer a la tasa de interés. Esta predicción es conocida como la “Regla Hotelling”. Si el mercado de recursos naturales es perfectamente competitivo, la regla Hotelling implica que el precio de mercado menos los costos marginales debe crecer a la tasa de interés, y por lo tanto el precio del recurso natural debe aumentar con el tiempo si los costos marginales son constantes. En contraste con la predicción teórica de Hotelling, sin embargo, los estudios empíricos han mostrado que los precios de los minerales casi no han tenido tendencias con el tiempo. Para mi tercer proyecto de investigación, estoy intentando reconciliar el modelo teórico de Hotelling con la evidencia empírica sobre los precios mundiales de los minerales. Me he beneficiado de la Beca de Pre-doctorado de Repsol YPF-Harvard Kennedy School de muchas maneras. La beca Repsol me puso en contacto con mucha gente que ha sido inapreciable para mi investigación. Aunque los contactos que he tenido la fortuna de hacer son demasiados para listar, varios merecen mención especial. Primero, Adrian Lajous, quien asistió regularmente a los seminarios de Repsol, no sólo me dio opiniones y reacciones sobre mi trabajo, sino que también me suministró información útil acerca de las subastas OCS. Segundo, Juan Badosa Pagés se tomó el tiempo para hablarme acerca del papel del riesgo en la producción petrolera. Tercero, William Hogan ha apoyado mi investigación de un modo extremo; no sólo se toma el tiempo para darme comentarios

Fellows 2003–2004

25

detallados de mi trabajo, sino que también ha sido extremadamente útil para ayudarme a adquirir los datos, las conexiones y la información que necesito. Cuarto, y quizás más importante, fue a través de la Beca Repsol YPF que conocí a Bijan Mossavar-Rahmani, quien hizo arreglos para mi visita en enero a la oficina matriz de Apache Corporation en Houston, y a una torre de perforación y una plataforma de producción costa afuera de Louisiana. El viaje fue una experiencia inapreciable para mí; el conocimiento que obtuve y los contactos que hice durante la visita no sólo me ayudaron con mis proyectos de investigación actuales, sino que también me han dado ideas para otros diversos proyectos que espero realizar en el futuro cercano. El programa de Becas Repsol YPF me ha dado una abundancia de recursos, contactos e información que han mejorado y mejorarán mis esfuerzos para explorar y desarrollar investigaciones en la economía petrolera. Como alguien que aspira a ser una economista académica, me he beneficiado fantásticamente de la oportunidad de conocer y aprender de personas que participan directamente en la industria petrolera, y mi experiencia como una becaria Repsol me ha preparado mejor para aplicar mis conocimientos de la economía académica a los asuntos importantes de política del mundo real concernientes a la energía y al petróleo.

Repsol YPF-Harvard Kennedy School Fellows Research and Program Summary

26

COLLECTED RESEARCH PAPERS Research Papers available on the internet at: http://www.ksg.harvard.edu/cbg/repsol_ypf-ksg_fellows/2003–2004_fellows.htm The collected papers appear in the volume: Repsol YPF-Harvard Kennedy School Fellows: Research Papers, Center for Business and Government, John F. Kennedy School of Government, Harvard University, April 2005. Juan Rosellón 1. 2. 3. 4.

5. 6. 7.

“Pricing Electricity Transmission In Mexico” “A Merchant Mechanism for Electricity Transmission Expansion,” with Tarjei Kristiansen, Journal of Regulatory Economics, forthcoming “Different Approaches to Supply Adequacy in Electricity Markets” “The Mexican Electricity Sector: Economic, Legal and Political Issues,” with Victor G. CarreónRodriguez, Armando Jiménez SanVicente, under revision to be published in a book edited by Stanford University “Strategic Behavior and the Pricing of Gas in Mexico,” with Dagobert L. Brito “Price Regulation in a Vertically Integrated Natural Gas Industry: The Case of Mexico,” Dagobert L. Brito, The Review of Network Economics, vol. 4, no. 1, March 2005 “Implications Of The Elasticity Of Natural Gas In Mexico On Investment In Gas Pipelines And In Setting The Arbitrage Point,” with Dagobert L. Brito

Joseph Aldy 1. 2.

“An Environmental Kuznets Curve Analysis of U.S. State-Level Carbon Dioxide Emissions” “Divergence in Per Capita Carbon Dioxide Emissions”

Darby Jack 1.

“Income, Household Energy And Health”

Cynthia Lin 1. 2. 3.

“The Multi-Stage Investment Timing Game In Offshore Petroleum Production: A Framework for an Econometric Model.” “Estimating Annual and Monthly Supply and Demand For World Oil: A Dry Hole?” “Optimal World Oil Extraction: Calibrating and Simulating The Hotelling Model”

Fellows 2003–2004

27

INFORMES DE INVESTIGACIÓN RECOPILADOS Los Informes de Investigación están disponibles en Internet en: http://www.ksg.harvard.edu/cbg/repsol_ypf-ksg_fellows/2003–2004_fellows.htm Los informes recopilados aparecen en el volumen: Repsol YPF-Harvard Kennedy School Fellows: Research Papers, Centro para Negocios y el Gobierno, John F. Kennedy School of Government, Harvard University, Abril de 2005. Juan Rosellon 1. 2. 3. 4.

5. 6. 7.

“Pricing Electricity Transmission In Mexico” “A Merchant Mechanism for Electricity Transmission Expansion,” con Tarjei Kristiansen, Journal of Regulatory Economics, de próxima aparición. “Different Approaches to Supply Adequacy in Electricity Markets” “The Mexican Electricity Sector: Economic, Legal and Political Issues,” con Víctor G. CarreónRodríguez, Armando Jiménez SanVicente, en revisión para su publicación en un libro dirigido por Stanford University “Strategic Behavior and the Pricing of Gas in Mexico,” con Dagobert L. Brito. “Price Regulation in a Vertically Integrated Natural Gas Industry: The Case of Mexico,” Dagobert L. Brito, The Review of Network Economics, vol. 4, no. 1, Marzo, 2005. “Implications Of The Elasticity Of Natural Gas In Mexico On Investment In Gas Pipelines And In Setting The Arbitrage Point,” con Dagobert L. Brito

Joseph Aldy 1. 2.

“An Environmental Kuznets Curve Analysis of U.S. State-Level Carbon Dioxide Emissions” “Divergence in Per Capita Carbon Dioxide Emissions”

Darby Jack 1.

“Income, Household Energy And Health”

Cynthia Lin 1. 2. 3.

“The Multi-Stage Investment Timing Game In Offshore Petroleum Production: A Framework for an Econometric Model.” “Estimating Annual and Monthly Supply and Demand For World Oil: A Dry Hole?” “Optimal World Oil Extraction: Calibrating and Simulating The Hotelling Model”

1

Pricing Electricity Transmission in Mexico

Juan Rosellón Centro de Investigación y Docencia Económicas (CIDE) and Harvard University

Abstract

during 2001–2006, which also appears very high compared to the annual historic growth rate of less than 4%. The Mexican government plans to cope with these needs with a threefold strategy. The first one is to work on improving the implementation of the current legal framework that permits private investments in self-supply, cogeneration, and independent power producers but that keeps the monopsony power of the State monopoly Comisión Federal de Electricidad (CFE). The second strategy is to improve the economic efficiency of the existing public companies through the creation of a virtual “shadow” market which tries to emulate a competitive environment in the internal operation of the distinct CFE’s generation, transmission, distribution and system operation subsidiaries. The third strategy is to lobby in the Mexican congress so as to implement a regulatory reform that allows the creation of a (real) electricity market where CFE’s is no longer a monopsony, and that permits competition between public and private generators in order to meet the needs of electricity consumers. The Mexican transmission network presents several signs of congestion, especially in the southeast and north of the country that might enhance the market power of regional power generators. Since the prospects of a regulatory reform that could permit private participation in transmission projects is for now halted in the congress, the Mexican government is nowadays studying the use of transmission pricing within the shadow market in order to promote adequate economic expansion of the transmission network. A “benefit-factors” pricing structure was recently proposed for the Mexican transmission system.2 In this paper I make another pricing proposal based on price cap regulation that provides a relief for the opposing short run (congestion) and long-run (investment) incentives of a transmission grid.

I propose an incentive regulatory framework to expand the large interregional links of the Mexican transmission network. I make an implementation of a two-part tariff model within the context of a combined merchant-regulatory approach, and the “shadow” Mexican electricity system. Results suggest that the best institutional structure for expanding the Mexican transmission grid would be one of a single transmission firm that charges even tariffs along the national territory. Keywords: Electricity transmission; financial transmission rights; incentive regulation; congestion management; Mexico. JEL classification code: L51 1.

Introduction

According to most recent official statistics,1 electricity demand in Mexico will annually grow at 5.6% during 2002–2011. In order to meet such demand increase, 30,300 MW of additional generation capacity will have to be added to the electricity system in that period, a large amount compared to the total current generation capacity of 38,519 MW. Likewise, total transmission capacity has to grow at an annual growth rate of 21%

I would like to thank Bill Hogan, Ingo Vogelsang and Marcelino Madrigal for insightful comments, as well as the CRE for their support. I also thank Manuel Ruiz and Armando Nevárez for very able research assistance. This paper was completed in residence under the Repsol YPF-Harvard Kennedy School Fellows program. Additional support from the Fundación México en Harvard is also gratefully acknowledged. 29

30

The literature on incentive structures for longterm expansion of the transmission network has an incipient development. There is (more or less) a consensus regarding the way to resolve short-term transmission congestion through nodal price differences. However, there is an intense debate regarding the optimal regulatory scheme to attract investment for the long-term expansion of the transmission network. Electricity transmission presents special characteristics due to so called “loop flows” which make impossible the definition of “available transmission capacity” in a point of time without the existence of complete information about the use of the network. In fact, under loop flows the addition of new transmission capacity can sometimes paradoxically reduce the total capacity of the network, which complicates the analysis of the welfare effects of certain transmission expansion projects. Analytical incentive structures proposed to deal with transmission expansion go from the “merchant” one, based on long-term financial right (LTFTR) auctions, to regulatory measures that make the transmission company pay for the social cost of transmission congestion. In the international practice, regulation has been basically applied in England, Wales and Norway to guide the expansion of the transmission network, while a mixture of planning and auctions of long-term transmission rights has been applied in the northeast of the US. Such combination is also being considered in New Zealand, and Central America. A combination of regulatory mechanisms and merchant incentives is alternatively used in the Australian market. Argentina also relies on a combined regulatory merchant approach with nodal pricing. I propose a two-part pricing model within a combined merchant-regulatory structure. I believe this scheme makes sense for the Mexican transmission network characterized by the coexistence of many meshed network regions that are connected by relatively large radial links. In this paper, I concentrate on analyzing the way to implement incentive-compatible regulation for the latter links. I show how through rebalancing of a two-part tariff, adequate expansion of the large interregional transmission links could be reached while LTFTR auctions would be used inside every electricity region. I identify the best institutional framework associated to this scheme.

Repsol YPF-Harvard Kennedy School Fellows Research Papers

The plan of the paper is as follows. Section 2 analyzes the characteristics of the Mexican transmission network, and studies its congestion in the context of the electricity shadow market. Section 3 carries out an analytical review of the literature on incentive structures for transmission expansion. Section 4 analyzes the benefit-factors pricing approach and, subsequently, presents a price-cap model and make several simulations with real data. Section 5 concludes. Table 1 Length of transmission lines (Km) CFE

LFC Tension level (Kv)

Year 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999* 2000*

400

230

161

150

5,997 6,035 6,080 6,287 7,610 7,827 7,908 8,380 8,810 9,099 9,103 9,162 9,710 10,623 10,979 11,337 11,908 12,249 12,489 13,263

9,581 10,801 10,892 11,515 12,237 13,174 13,925 15,283 16,090 16,417 17,315 17,673 18,267 18,217 18,532 18,878 19,375 20,292 20,595 21,275

225 291 291 291 291 291 291 342 379 379 379 379 379 379 379 379 379 379 379 379

786 786 786 834 842 842 851 851 888 918 920 983 920 920 921 992 993 995 995 995

FUENTE: Banco de México Banco de Datos 1999 *Own calculations with CFE data

2.

The Mexican Transmission System

In 2001, the Mexican transmission and distribution network was 670,902 km long. 5.4% were 230–400 kV transmission lines, 6.2% were 69–161 kV subtransmission lines, and the remaining 88.4% were distribution lines between 2.4 and 60 kV. Table 1 presents the detailed length evolution of transmission lines between 150 and 400 Kv. It can be seen that 98.17% correspond to tension levels between 230 and 400 Kv.

Juan Rosellón

31

Figure 1 Electricity Transmission Network of Mexico, 2001

Around 115,000 MVA of the total 170,000 MVA of installed transmission and distribution capacity correspond to transmission lines. The Mexican transmission network has grown at an annual rate of 3.74% per year since 1981. Figure 1 shows the maximum capacity of the regional transmission

links for the 32 regions of the Mexican electricity system in 2001. There are also several international interconnection points in the system, whose capacity is shown in figure 2. The Energy of Ministry of Mexico (Sener) foresees a 20,357 km increase in the length of 69–400Kv

Figure 2 Capacity of Interconnecting Transmission Links

Repsol YPF-Harvard Kennedy School Fellows Research Papers

32

Figure 3 Expansion of Interregional Transmission Capacity (MW), 2002–2006

transmission lines for 2002–2006. Figure 3 presents the expected capacity growth for each regional link. In order to meet an annual growth rate of 5.6% in electricity demand during 2002–2011, Sener foresees that total transmission capacity will grow from 12,740 Mw in 2001 to 25,985 Mw in 2006, which represents an average annual growth rate of around 21%, a huge increase compared to the historic annual growth rate of transmission capacity. This will require annual investments of USD 1.3 billion that will be carried out through public budget in a direct way (46%), or through financed public projects, or Pidiregas (54%).3 3.1 The Shadow Electricity Market and Transmission Congestion There have recently been several initiatives to reform the Mexican electricity sector, such as the 1999 privatization proposal of the Zedillo administration or the more conservative regulatory reform proposal made by the Fox administration in 2002.4 However, the only concrete reform carried out so far is the one in 1992 that allowed for private investment in co-generation, self supply, and independent power producer (IPP) projects under a single-buyer (or monopsony) scheme.

According to this scheme, all private producers must sell their exceeding power to CFE relying on a government credit: IPPs operate under 25-year power purchase agreements with CFE. So far, this scheme has attracted some private investment. In 2001, private capacity generation represented 12.6% of total capacity generation. Additionally, even though there has not been any major reform process in the Mexican electricity sector, an internal (or shadow) market is being implemented by CFE in a nodal fashion since September 2000. This virtual market seeks to emulate a competitive market. It uses an optimization module that least-cost dispatch based on actual generation costs (merit order rule) in one-dayahead and real-time markets. The one-day-ahead market establishes production, consumption and price schedules for each of the hours of the following day. The differences between forecasted and actual schedules are cleared at real-time prices. Bids are actually submitted to the system operator (CENACE) by the different “programmable” thermal CFE’s generation plants, which are administratively separated so that they function as different power producers.5 Total generation capacity amounts 38,519 MW, with the following

Juan Rosellón

33

generation mix: hydro 26%, thermal 38%, combined cycle 9%, gas 7%, carbon 7%, dual 6%, nuclear 4%, geothermal 2%, and others (wind, solar) 1%. Payments to generators include a “capacity” payment intended to foster the development of generation capacity reserves. The distribution companies are also divided into several distribution units. A MW-Mile method (as in Shirmohammadi, et al, 1989) is used to set transmission tariffs. Through this method, charges for transmission services for tensions greater than or equal to 69 Kv are calculated as the maximum between “fixed costs plus variable costs” and “operation and maintenance costs.” Administrative fixed costs are added to this amount. Fixed costs are basically the long-run incremental cost of the transmission network. They are allocated among consumers of the current grid and consumers of the future expanded grid according to the impact both have over the complete network.6 Nodal prices are determined in the 1,400 nodes of the main transmission grid through use of a power flow model.7 Using this nodal price system, Madrigal (2000) and Madrigal et al (2002) estimate transmission congestion rents as well as detect main congestion transmission links. Table 2 presents their estimation for congestion rents for three scenarios of the load duration curve in 2000. Table 2 Annual Congestion Rents Load Duration Curve Scenario

Annual Duration (%)

Annual Duration (hours)

High Medium Low Total

00.15 70.00 29.85 100

13.14 6132.00 2614.86 8760

Revenues, Payments and Rents from Congestion

Scenario

Generators’ Income (pesos)

Demand Payments (pesos)

Congestion Rents (pesos)

High 9951916.98 11441067.31 1489150.33 1963514.83 6266757.05 4303243 Medium Low 2811096.18 4260572 1449475.83 Annual Est 33868877214 49718889151 15850011937

Table 3 Main Congestions During Peak Demand Line From Central Oriental Sureste Colima Sonora Laguna Coahuila Monterrey

To

Flow (MW)

Limit (MW)

–600 –440 100 1216 –410 200 1550 250

600 440 100 1216 410 200 1550 250

Balsas Veracruz Campeche Occidental Sonsur Chihuahua Monterrey Laguna

Source: CRE main congested transmission lines and import areas

Area

Capacity Generation Demand Exportation

Sonnorte laguna Coahuila Monterrey Bravo Central Veracruz Occidental Campeche

806.00 643.00 2734.00 1640.00 520.00 2632.00 1845.00 1912.00 150.00

718.42 335.95 2575.79 1367.00 520.00 2632.00 510.11 1594.49 0.00

1128.42 –410.00 745.95 –410.00 876.46 1699.30 3080.99 –1713.99 559.98 –39.98 6519.01 –3887.01 523.80 13.69 4023.81 –2429.32 146.26 –146.26

Annual estimated congestion rents arising from congestion in transmission amount USD 1.4 billion. Table 3 presents the main transmission lines subject to congestion. Instituto de Investigaciones Eléctricas (2003, p. 44) show a volatile structure of nodal price differences between the Río Bravo node (in the Northeast US-Mexico border) and the Querétaro node (in the center of the country. For example, during January 2000 there were some days where the price difference between these two nodes could increase from less than USD 10 per MWH to more than USD 60 per MWH. Among other factors—including low hydro production and high-cost generation—this is explained by transmission congestion. Madrigal (2000) and Madrigal et al (2002) further develop a power-flow linear model so as to study the effects of transmission congestion on local generation market power in a hypothetical future electricity market in Mexico. The Yucatan area (zones 21, 22, 23, and 28) is particularly isolated since its only link to the system is a 100 Mw congested line that links the Campeche zone to the

Repsol YPF-Harvard Kennedy School Fellows Research Papers

34

rest of the system. The “must-dispatch” and Lerner indexes are calculated for distinct technology generators, and it is shown that a 179 MW generator could make use of its market power in the area due to congestion in transmission links, actually charging a USD 2000 per MW electricity price. Similar results are reached for the central and the Northeast areas. In a similar effort, Comisión Federal de Electricidad (2000) confirms that the isolated Peninsular and Northwest regions are more susceptible for market power abuse than central regions. Hartley and Martinez-Chombo (2002) further analyze the impact of Sener not developing its proposed expansion of transmission links on additional investment in generation capacity. They particularly find that the upgrades between regions 22 and 20 (additional 1,000 MW of capacity), and regions 18 and 20 (additional 3,700 of capacity) are critical to transmit power from the hydro plants in the Grijalva river region (22) to the central part of the nation. If these projects are not built by the end of 2004, forecasted demand in 2005 will not be able to be met without the construction of additional new generation capacity 3.

The Literature

Different from other industries, electricity transmission presents special characteristics—beyond economies of scale and cost subadditivity8—that complicate the regulatory analysis of adequate incentives for network expansion. Externalities in electricity transmission are mainly due to “loop flows,” which arise from interactions in the transmission network.9 The effects of loop flows imply that transmission opportunity costs and pricing critically depend on the marginal costs of power at every location. Energy costs and transmission costs are not independent since they are determined simultaneously in the electricity dispatch and the spot market. Then, as explained by Bushnell and Stoft (1997), certain transmission investments in a particular link might have negative externalities on the capacity of other (maybe remote) transmission links. In fact, the addition of new transmission capacity can sometimes paradoxically reduce the total capacity of the network (Hogan, 2002b). This situation is further complicated by the fact that equilibrium in the forward electricity trans-

mission market has to be coordinated with equilibrium in three other markets: the energy spot market, the forward energy market (or bilateral contract market), and the generation capacity reserve market (see Wilson 2002). Furthermore, the effects of an increase in transmission capacity are uncertain. As shown by Léautier (2001), the net welfare outcome of an expansion in the transmission grid depends on the weight in the welfare preferences of the generators’ profits relative to the consumers’ weight. Thus, incumbent generators are not always the best economic agents to carry out transmission expansion projects. Although transmission expansions might allow established generators to enhance their revenues due to improved access to new markets and increased transmission charges and FTRs, such gains are overcome by the loss of their local market power. The literature on the long-term investment in the transmission network has an incipient development. As Joskow and Tirole (2003) argue, the economic analysis of electricity markets has focused on short-term issues (such as spot markets for energy, short-run congestion management, nodal pricing, and day ahead auction rules) and has typically considered the transmission network capacity as given, fixed, and of common knowledge. However, transmission capacity is stochastic and its development mutually depends on the evolution of generation investment. There is (more or less) a consensus in the economics literature regarding the way to resolve short-term transmission congestion. As shown in Hogan (2002a), the difference of electricity prices between two nodes in a power flow model defines the price of congestion. However, there is an intense debate regarding the way to attract investment for the long-term expansion of the transmission network, and to solve the dual opposite incentives to congest the network (in the short run) and to expand it (in the long run). There are (at least) three existing hypotheses on structures for long-run investment in an electr icity transmission network: the long-term financial-transmission-right hypothesis, the incentiveregulation hypothesis, and the market-power hypothesis. Each one relies on a distinct institutional set up. The first approach, the “merchant” option, relies on the auction of LTFTR by an independent system operator (ISO). This approach directly faces the problems implied by loop flows

Juan Rosellón

so that, in order to proceed with a line expansion, the investor pays for the negative externalities generated. To restore feasibility, the investor has to buy back sufficient transmission rights (proxy awards) from those who hold them initially, or an ISO would have to retain some transmission rights in an auction for long-term rights to make sure that the expansion project does not violate the property rights of the original transmission right holders. This is the basis of a LTFTR auction (see Hogan, 2002 b). Joskow and Tirole (2003) carry out criticisms to the merchant approach. They however concentrate their analysis on the short-run version of the FRT. model. They argue that the efficiency results of this model rely on perfect-competition assumptions that are not met in the reality of transmission networks.10 Additionally to technical difficulties in defining an operational FRT. auction,11 these authors think that the FRT. typical analysis is static, which contradicts the dynamic nature of transmission investment and the interdependency between generation and transmission investments. Joskow and Tirole carry out an extensive analysis on the implications of lifting these strong perfect competition assumptions: Market power: The existence of market power and vertical integration might jeopardize the success of FRS auctions. Due to market power in constrained regions, prices will not reflect the marginal cost of production. Generators in constrained regions will tend to withdraw capacity to bring up their prices and this will overestimate the cost-saving gains from transmission investments.12 Lumpiness: Lumpiness in transmission investment implies that the total value paid to investors through FRS understates the social surplus created by such an investment. The large and lumpy nature of major transmission upgrades then calls for the need of long-term contracts before making a transmission investment, or of property rights to exclusively use the incremental investment for a certain period of time. Contingencies: The difficulties associated with contingencies in long-run electricity transmission might question the real capacity of the MERCHANT/merchant approach to solve the loop flow problem. Additionally, existing transmission

35

capacity and incremental capacity are not well defined and are of a stochastic nature. Even in the two-node case, realized capacity could be less than expected capacity so that revenue-adequacy condition is violated. Loop Flows: As shown by Bushnell and Soft (1997) for a network with loop flow, an addition in transmission capacity might have a negative social value. Additionally, the initially feasible MERCHANT set can depend on random exogenous variables. Information Asymmetries: The separation of transmission ownership and system operation in the MERCHANT model creates a moral hazard “in teams” problem. For example, an outage can be claimed to result from poor line maintenance (by the transmission owner) or from imprudent dispatch (by the system operator). Moreover, since transmission investment is not static in reality, there is no perfect coordination of interdependent investments in generation and transmission. In fact, stochastic changes in supply and demand conditions imply uncertain nodal prices. In addition, equal access to investment opportunities is not a good assumption because deepening investments of the incumbent’s network can only be efficiently implemented by the incumbent. Regarding the loop-flow issue, Hogan (2002b) analyzes the implications of loop flows on transmission investment raised by Bushnell and Stoft (1997). Hogan makes a preliminary attempt to analytically provide some general axioms to properly define LT FTRs. Hogan’s model relies on an institutional structure where there are various established agents (generators, Gridcos, marketers, etc.) interested in the transmission grid expansion. Under an initial condition of non-fully allocation of FTRs in the grid, the awarding of incremental LTFTRs should satisfy the following basic criteria: 1. 2.

3. 4.

An LTFTR increment must keep being simultaneously feasible (feasibility rule).13 An LTFTR increment remains simultaneously feasible given that certain currently unallocated rights (or proxy awards) are preserved. Investors should maximize their objective function (maximum value). The LTFTR awarding process should apply both for decreases and increases in the grid capacity (symmetry).

36

As shown by Bushnell and Stoft (1996) and Bushnell and Stoft (1997), under these conditions allocation of new PTP-FTR obligations will not reduce social welfare. Hogan explains however that defining proxy awards is a difficult task. One possibility would be to define every possible use of the current grid as a proxy award. However, this would imply that any investment beyond a radial line would be precluded, and that incremental award of FTRs might require adding capacity to every link on every path of a meshed network. A better possibility would be then to define as a proxy award the best use of the current grid along the same direction that the (positive or negative) incremental FTR was awarded. “Best use” could be defined in terms of a preset proxy references so that proxy awards maximize the value of such references. Another possibility is to define “best” in terms of the maximum value of investors’ preferences. Then, given a proxy rule, an auction is carried out in order to attract investment for transmission expansion. In case the investors’ preference criterion is chosen, the auction model maximizes investors’ preferences to avoid LTFTRs in the direction of the expansion subject to the simultaneously feasibility conditions and the “best” rule. Notwithstanding, Hogan (2003) recognizes that LTFTRs only provide efficient results under assumptions of no existence of market power and non-lumpy marginal incremental expansions of the transmission network. He then believes that regulation plays an important role in the development of large and lumpy projects, and in mitigation of market power abuse. Hogan’s response to contingency concerns is twofold.14 On one hand, only contingency conditions that are outside the control of the system operator could lead to revenue inadequacy of FTRs, but such cases do not describe the most important contingency conditions. On the other hand, most of the remaining contingencies are foreseen in a security-constrained dispatch in a meshed network with loops and parallel paths. If one of “n” transmission facilities were lost, the remaining power flows would still be feasible in an “n – 1” contingency constrained dispatch. Hogan (2003) recognizes that information asymmetries and agency problems are present in a reformed electricity industry with an ISO, independent transmission providers and decentralized

Repsol YPF-Harvard Kennedy School Fellows Research Papers

market players. However he believes that the main issue on transmission investment is deciding the boundary between merchant and regulated transmission expansion projects. It is not clear to him how asymmetric information can affect such a boundary. The second alternative to electricity transmission expansion seeks to solve the transmission expansion problem through regulation only, and within a different institutional framework. System operation and ownership of the transmission company are carried out by a “Transco” that is regulated through benchmark or price regulation so as to provide it with incentives to invest in the development of the grid, while avoiding congestion. Léautier (2000), Grande and Wangesteen (2000), and Harvard Electricity Policy Group (2002b) propose mechanisms that compare the Transco performance with a measure of welfare loss due to its activities. Joskow and Tirole (2002) propose a surplus-based mechanism to reward the Transco according to the redispatch costs avoided by the expansion, so that the Transco faces the entire social cost of transmission congestion. Another regulatory alternative is a two-part tariff cap that solves the opposite incentives to congest the existing transmission grid in the short run, and to expand it in the long run (see Vogelsang 2001). Incentives for investment in network expansion are achieved through the rebalancing of the fixed part and the variable part of the tariff. This approach tries to deepen into the analysis of the cost and demand functions for transmission services, which are not very well understood in the literature. However, in order to carry out this task, it has to assume a monotonic increasing behavior of the transmission cost function. As argued by Hogan (2002b), this assumption is not in general valid since an expansion in a certain transmission link can derive in a total decrease of the network capacity. Additionally, in order to study the cost and production characteristics of a Transco, Vogelsang finds it useful to define the Transco’s output (or throughput). As argued in the FTR literature (Bushnell and Stoft, 1997, Hogan, 2002a, Hogan, 2002b), this problem is very difficult since the physical flow through a meshed transmission network cannot be traced due to its multidimensionality. The third alternative method for transmission expansion seeks to derive optimal transmission

Juan Rosellón

expansion from the power-market structure of electricity generation, and considers conjectures made by each generator on other generators’ marginal costs due to the expansion (see Sheffrin and Wolak 2001, Wolak 2000, London Economics International 2002, and California ISO and London Economics International, 2003). The basic idea is to estimate the generator’s bidding behavior before and after a transmission upgrade. This method also uses a real-option analysis to derive the net present value of both transmission and generation projects through the calculation of their joint probability. The results of this model show that benefits of transmission expansion are small until added capacity surpasses a certain upper limit that, in turn, is determined by the possibility of induced congestion by the strategic behavior of generators with market power. Transmission expansion will only yield benefits until it is large enough with respect to a given generation market structure. The addition of cost uncertainty (due to environmental factors and local opposition to transmission projects) implies that many small upgrades are preferable to large greenfield projects.15 The main contribution of this approach is that it explicitly models the existing interdependence of generation investment and transmission investment. However, this approach relies on a transportation model with no network loop flows. As argued by Hogan (2002a), the use of a transportation model in the electricity sector is not adequate because it does not consider discontinuities in transmission capacity due to the multi-dimensional character of a meshed network. 4.

An Incentive Mechanism for the Mexican Network

In last section I analyzed different models for long-term investment in a transmission network. With the exception of the Vogelsang’s two-part tariff mechanism (Vogelsang, 2001), the three alternatives propose different general approaches that do not provide implementation details as to the specific type of pricing for transmission services. For example, the LTFTR option is a way to hedge consumers from nodal price fluctuations in the long run. However, as argued by PérezArriaga et al (1995), revenues from nodal prices only permit to recover 25% of total costs. Therefore, LTFTRs should be complemented with a cer-

37

tain fix pricing structure or, as in Rubio-Odériz and Pérez-Arriaga, 2000, a complementary charge that permits recuperation of fixed costs.16 This fact is recognized by Hogan (1999) who believes that complete reliance on market incentives for transmission investment is undesirable. Rather, Hogan (2003) argues that merchant and regulated transmission investments might be combined so that regulated transmission investment is limited to projects where investment is “large” relative to market size, and “lumpy” so that it only makes sense as a single project as opposed as many incremental small projects.17 Pricing for electricity transmission should satisfy certain desirable properties that are consistent with the regulatory scheme that supports it. A set of relevant principles is provided in Transpower (2002). A first principle is that pricing for the different cost components of transmission should not conflict with each other (integrated approach). So, sunk and fixed costs should be allocated in such a manner that they do not preclude the variable charge from reflecting nodal prices.18 Likewise, new investment costs should be allocated to the beneficiaries of such new investment.19 A second principle is that the allocation of sunk and fixed costs should not distort production and consumption schedules or investment decisions. In other words, fixed cost allocation should not reduce social welfare. A third property is that variable costs are based on marginal costs. In a system where a power flow model is used to determine nodal prices, variable charges should be determined by nodal price differences that reflect (short run) transmission congestion. A fourth principle is that transmission charges must preclude cross subsidies. In the remaining of the present section I concentrate on transmission pricing methods applied to the Mexican transmission network within the context of these guiding principles. In 4.1 I review a proposal based on the “benefit factors” approach. In 4.2 I propose my own model that is further applied to the Mexican transmission grid. Such a model derives from the regulatory alternative (second alternative) described in section 3. 4.1 The Benefit-Factors Approach Instituto de Investigaciones Eléctricas (2003) proposes a method for calculating CFE’s transmission

38

charges based on the benefit-factors method of Rubio-Odériz and Pérez-Arriaga (2000).20 Assuming the existence of a nodal pricing system, RubioOdériz and Pérez-Arriaga (2000) propose to base the complementary charge according to the economic benefit that each transmission network facility causes to each agent. For a consumer, the benefit is measured as the reduction in its total electricity charges before and after a new transmission corridor is added, while for a generator it is the increment in its profits. Rubio-Odériz and Pérez-Arriaga only consider positive benefits (no agent can receive a payment for a negative benefit). Under this approach, the total allowed transmission revenue is annually determined and regulated, typically by the regulatory agency.21 The calculation of this amount considers both the capital, operation and maintenance costs of the existing transmission infrastructure, as well as the cash flow necessary to cover the expansion costs to meet demand from future generators and consumers (demand is also forecasted during the planning process).22 Regulation of the annual revenue is accompanied by regulation of the quality of transmission services. The authorized annual revenue is monthly disaggregated by transmission corridor. The variable part of the authorized revenue is recovered through a variable charge that is directly determined by the difference in nodal prices, and it is merely a congestion charge. The “complementary” part of the authorized revenue is equal to the total authorized revenue less the variable-charge revenue. The complementary amount is recovered through the complementary charge according to the particular benefit that each consumer obtains through network expansion. Some characteristics of this model might be criticized. In first place, the authors of the benefit approach seem to develop an ad-hoc mechanism where there seems not to exist an objective way to determine the complementary charge, especially for long-term transmission projects. The problem of how to determine the beneficiaries of a certain transmission expansion project is as subjective as the “joint-cost” allocation dilemma.23 Likewise, an annual determination of the transmission revenue could in practice be at odds with the determination of the net present value of long-term transmission projects. Additionally, the authors consider the “nonnegative-benefits” assumption as valid because

Repsol YPF-Harvard Kennedy School Fellows Research Papers

“. . . normally, an agent with negative benefit would have market power (due to network constraints) without the line being considered for complementary charge.”24 However, Hogan (2002b) shows that negative benefits could occur during a transmission expansion project due to the power flow nature only. In this sense, the benefit-factors approach seems not to comply with the Transpower’s second principle for electricity transmission. Namely, that transmission pricing should not reduce social welfare. Examples can be shown where a certain transmission expansion project (such as the building of a parallel line), and its subsequent complementary charge, could diminish total transmission capacity implying negative externalities (and benefits) on certain transmission property rights holders and, hence, reducing social welfare (see Bushnell and Stoft, 1997, and Hogan, 2002b).25 4.2 The Two-Part Tariff Mechanism I now propose an alternative pricing scheme for the Mexican transmission network. Building on Hogan (2002b) and Vogelsang (2001) I propose a pricing regulatory method in the context of a combined merchant-regulatory mechanism for electricity transmission expansion. As discussed in section 3, there is not yet in theory or practice a single mechanism that guarantees an optimal expansion of the electricity transmission network. However, the distinct study efforts suggest a second-best standard that combines the merchant and the regulated transmission models, so that “small” transmission expansion projects rely on the merchant approach while “large and lumpy” projects are developed through incentive regulation. Figure 1 suggests that an LTFTR method could be used “inside” the 32 transmission regions of the country, while a price-cap rule could be applied to develop the large lumpy links among such regions.26 This approach could be used since the current CFE’s shadow market is already based on nodal pricing, the systems inside each of the 32 regions are relatively meshed, and the large links joining transmission regions are approximately node-to-node radial lines.27 The application of this method could later be adapted to a possible reform of the Mexican electricity market that allowed more competition from private players. In this paper I concentrate my analysis on incentive regulation of the large interregional links

Juan Rosellón

in figure 1, and develop in another research a merchant mechanism for the 32 meshed transmission regions.28 Given a “large” potential transmission expansion project (such as the one between regions 17 and 18 in figure 1)—that has proven not to be suitable for development through merchant mechanisms—the ISO carries out a feasibility test to check out the redispatch impact of the project.29 Next, a competitive bidding process (or “solicitation” process, as in Rotger and Felder, 2001) is carried out to select the party that will build such a link. The bid winner would carry out the project given a price-cap constraint,30 and a “payback” constraint (as in Bushnell and Stoft, 1997) that would internalize the negative externalities generated for the expansion project. My modeling strategy in this paper is to abstract from loop-flow effects (and, thus, from the payback condition), so as to study two scenarios.31 One scenario is a hypothetical situation where there is a single two-node radial line that provides the transmission service in all the country. In this first scenario, a single firm would own the transmission network and would apply a uniform twopart tariff along the country. The second abstraction would study a hypothetical situation where there are several radial transmission lines serving each of the five electricity regions of the country. Each one of these systems would be physically separated from the other systems. In this second scenario, I analyze two subcases. In the first, different firms that charge distinct variable and fixed fees with respect to the other regions would own the lines. In the second sub-case, a single firm would own each of the regional systems, and would charge the same variable fee across regions but with different fixed fees. I carry out simulations for the Mexican transmission network using a price-cap regulatory method similar to Vogelsang (2001). Vogelsang shows that price structure regulation can be used to solve congestion problems of transmission lines, in the short run, as well as capital costs and investment issues, in the long run. He proposes a twopart tariff regulatory model with variable (or usage) charges, and fixed (or capacity) charges. The variable charge can be actually defined in terms of nodal prices in an institutional structure where an ISO coordinates a competitive market and runs short-run capacity utilization, while a Transco owns and operates the network.32 The transmission

39

firm is a profit-maximizing monopolist that makes investment and pricing decisions subject to a regulation of its two-part tariff. The solution to this problem takes care of congestion problems through the variable charges. Recuperation of long-term capital costs is achieved through the fixed charge, while incentives for investment in expansion of the network are reached by a rebalancing of the fixed charge and the variable charge. Transmitted volumes for each type of service are used as weights for the corresponding different prices so that Transco’s profits increase as capacity utilization and network expansion increases. In equilibrium, the rebalancing of fixed and variable charges depends on the ratio between the output weight and the number of consumers.33 I now concentrate my analysis on the firm’s rebalancing of the fixed fee and the variable fee, and its subsequent impact on the firm’s profits. My simulations are performed using the Newton method through progressive derivatives.34 Common assumptions in the distinct simulations are: • • • •

A radial line links the production node with the consumption node. The inflation rate and the x efficiency adjustment factors are equal to zero. Operation costs are equal to zero. Previous periods transmission flows are used as Laspeyres weights

The period of analysis is 2001–2006. Year 2001 is the base year. The initial data for prices, electricity flows and necessary costs are obtained from this year. The analysis of rebalancing of fees is carried out for the 2000–2004 period. I also assume the following demand function:35 q t+1 = (1 + α)q – β p t. This equation establishes a demand increase at rate α and it presents an inverse relationship between demand and prices. Another important assumption is that the firm is myopic with respect to profit maximization so that in each period it maximizes profits separately from the other periods. Case 1: Monopolist with “postage stamp” fees I analyze first a single firm that covers the whole country applying uniform fees. In the two cases that I present in this section, the transmission firm solves the following optimization problem:

Repsol YPF-Harvard Kennedy School Fellows Research Papers

40

subject to

(1)

where: Ft = fixed fee in period t. pt = variable fee in period t. qt = real oriented energy flow in period t (in kWh). Kt = available transmission capacity in period t. w = type of weight.36 N = number of consumers CF = fixed costs The transmission cost function c(q,K) reflects the sunk cost nature of transmission investment and has the following form:

Figure 4 presents the optimal values of the simulation for fixed fees and variable fees. The fixed fee represents the amount charged to each consumer. The restriction on the fixed fee is binding in all periods. As shown in table 4, transmission capacity increases 42,927 MWh, and the length of transmission lines increases 10,165 kilometers. Figure 4 Rebalancing of fixed fees and variable fees (fixed number of consumers). p=variable fee; F= fixed fee (2001 pesos per KWH) 0.0380 0.0375 0.0370 0.0365 0.0360 0.0355 0.0350 0.0345 0.0340 0.0335 0.0330

where

p F

With respect to Vogelsang (2001), the following restrictions have been added: • • •

The fixed fee and the variable fee must be nonnegative. Income from the fixed fee must be greater than or equal to fixed costs at each period. The transmission firm must make the needed investments in each period so as to cover the difference between transmission capacity of the previous period, and transmission demand in the current period.

Additionally, I assume in this section that there is a single radial line in all the country. Its cost function is composed by fixed costs (that depend of the transmission capacity in period t – 1,) and by investment costs. When q t ≤ K t–1, investment costs are zero. Values corresponding to fixed costs, and investment costs were taken from Comisión Federal de Electricidad (2000).37 Fixed number of consumers In this subsection I assume a fixed number of consumers over the study period. Consumers might increase their demand for transmission services. I model, for example, the case of several established distribution companies whose number does not grow but their demand for transmission service does.

125 120 115 110 105 100 95 2002

2003

2004

2005

2006

0.0377 102.94

0.0370 107.69

0.0363 112.73

0.0357 117.13

0.0349 123.07

90

Table 4 Monopolist with “postage stamp” fees and fixed no. of consumers: Profits, capacity and expansion of the tranmsission network

Year

Profits (1999 pesos)

Transmission capacity (MWh)

2002 2003 2004 2005 2006

3,942,895,867 3,971,421,490 3,995,992,170 4,018,819,838 4,033,832,898

164,774 172,824 181,381 190,470 200,131

Length of transmission lines (kilometers) 39,020.9 40,927.1 42,953.6 45,106.1 47,393.8

Variable number of consumers I now assume that the number of consumers increases 3.1% every year.38 Figure 5 presents the results for the rebalancing of the fixed fee and the variable fee. As shown in table 5, transmission capacity increases 42,816 MWh, and the length of the transmission network increases 10,139 kilometers. It can also be noted that profits increase year by year since, as the number of consumers grows, the firm will have more possibilities for rebalancing the fixed and variable fees.

Juan Rosellón

41

Table 5 Monopolist with “postage stamp” fees and variable no. of consumers: Profits, capacity and expansion of the transmission network

Year

Profits (1999 pesos)

Transmission capacity (MWh)

2002 2003 2004 2005 2006

4,029,084,977 4,149,449,558 4,277,271,353 4,405,660,628 4,471,645,358

164,754 172,768 181,261 190,275 200,020

Length of transmission lines (kilometers) 39,016.1 40,913.8 42,925.2 45,059.9 47,367.6

Figure 5 Rebalancing of fixed fees and variable fees (fixed number of consumers). p=variable fee; F=fixed fee (2001 pesos per KWH) 0.0400 0.0380 0.0360 0.0340 0.0320 0.0300

2002

2003

2004

2005

2006

p

0.0382

0.0378

0.0376

0.0372

0.0328

F

100.12

102.59

103.69

106.55

136.39

160 140 120 100 80 60 40 20 0

national electricity system.40 Thus, there are 32 distribution companies, and this number remains constant from one regulatory period to the other. Table 7 presents the existing distribution companies for each of the electricity areas, as well as their respective demand growth rates. Figure 6 presents the distinct variable fees charged by the regional transmission companies. Such fees show a decreasing linear tendency. Table 8 presents the different fixed fees applied to each one of the distribution companies in each area. Fixed fees are different among distinct areas because the number of consumers and their demand also differs. The fixed charge is inversely proportional to the number of consumers; the less the number of consumers the greater the fixed charge. Likewise, constraints on the fixed charge remain binding in each case. Profits for each firm increase as well as capacity. Annual average growth rate of profits for all regional companies is greater than the one obtained for a monopolist with postage stamp prices (see table 9). Case 3: Monopolist with discriminatory tariffs In this subsection I analyze the behavior of a single independent transmission company that operates in all the areas of the national electricity Figure 6 Variable fee per area (2002–2006)

Case 2: A single transmission firm for each electricity area

0.0395 0.0390 0.0385 0.0380 Variable Fee

In this section I assume there exists one regional transmission company for each of the five areas in the national electricity system. Each company is a regional monopoly and does not have any relationship with the rest of the transmission companies. That is, I assume there are no interconnections among the different electricity areas, and that each area has its own power generators that satisfy demand increases within the area. I also assume that transmission networks within each area are radial lines. Each transmission company solves program (1) within its area. There are now different increments in demand given by the Secretaría de Energía forecasts.39 Transmission capacity per area in for the 2002–2006 period is presented in table 6. I assume that there is a conglomerate of distribution companies per each of the areas of the

0.0400

0.0375 0.0370 0.0365 0.0360 0.0355 0.0350

Noroeste Centro-occidente Noreste Centro Sur-sureste

2002 0.0387 0.0387 0.0387 0.0387 0.0387

2003 0.0381 0.0381 0.0379 0.0383 0.0383

2004 0.0374 0.0374 0.0371 0.0378 0.0379

2005 0.0367 0.0365 0.0363 0.0374 0.0375

2006 0.0360 0.0357 0.0354 0.0369 0.0371

Repsol YPF-Harvard Kennedy School Fellows Research Papers

42

Table 6 Maximum Transmission Capacity per area MW, 2002–2006 2002 NO NE CO CENTRO SS total

NO

NE

CO

1156 140 55 0 0 1351

140 3161 400 0 357 4058

55 400 3418 1966 170 6009

CENTRO

SS

total

0 0 1966 0 4945 6911

0 357 170 4945 6432 11904

1351 4058 6009 6911 11904

0 0 2018 0 95 2113

0 1000 174 0 0 1269

1777 5665 7424 1269 1269

0 0 2035 0 5148 7453

0 1000 174 5418 9364 15956

2412 1725 7424 7453 15959

0 0 1982 0 5689 7671

0 1000 190 5689 8719 15598

2431 1926 7728 7671 15598

0 0 2309 0 5570 10506

0 1000 194 5570 8264 7879

3255 10047 10506 7879 15028 15028

2003 NO NE CO CENTRO SS total

1528 140 109 0 0 1777

140 3662 863 0 1000 5665

109 863 3059 2018 174 6223 2004

NO NE CO CENTRO SS total

1808 322 282 0 0 2412

322 3971 1832 0 1000 7125

282 1832 3101 2035 174 7424 2005

NO NE CO CENTRO SS total

1836 365 230 0 0 2431

365 4382 2179 0 1000 7926

230 2179 3417 1982 190 7728 2006

NO NE CO CENTRO SS total

1806 414 1035 0 0 325

414 4460 4173 0 1000

1035 4173 2795 2309 194 10047

system, but that can discriminate the prices it applies in each of the areas. The Baja California and South Baja California regions are not included because these regions are not usually physically interconnected with the rest of the country’s transmission network.41 The transmission firm now solves the problem:

subject to (2)

In the new restriction of this problem, the subindex of the fixed charge runs over consumer groups in each area. In each different transmission region the same variable charge is applied but

Juan Rosellón

43

Table 7 Distribution Companies and Demand Growth Region

No. of Distribution Companies

Northwest Center-West Northeast Center South-Southeast

10 5 7 1 9

Demand Growth Rate 5.6 5.6 6.7 4.1 3.7

Source: *Secretaria de energía (2002)

with different fixed charges. That is, the firm uses a discriminatory two-part tariff for differentiated goods where product differentiation arises from different consumer groups in each area. Figure 7 presents the uniform variable fee for all the transmission areas. As in previous cases, I observe that the variable charge has a decreasing tendency. The difference is that such a tendency is no longer linear. Additionally, the variable charge in this case is lower than in the previous cases. Another relevant result is that, in all the analyzed areas, investment obtained through a single monopoly that permits pricing dis-

crimination is lower than in the other cases. Figure 8 presents the data for fixed charges in each area. The constraint over the fixed charge is binding in each of the areas that have a larger increase in demand an, hence, face greater investment challenges. Fixed charges remained at lower levels than those corresponding to discrimination in both parts of the tariff. Table 10 makes a comparison of all the cases of the simulation in terms of profits, capacity increase, and network expansion. The results critically depend on two effects. The first is an economies-of-scale effect due to the partition of the transmission grid from a single network (case 1) to several smaller networks (cases 2 and 3). The second effect is a discriminatory-pricing effect that due to the possibility of price discrimination in cases 2 and 3. The comparison of case 1 and case 3 shows that profits, capacity and expansion are greater under case 1 than under case 3. The economies-of-scale effect then dominates the discriminatory-pricing effect. The comparison of case 1 and case 2 shows that capacity and expansion are greater for case 1, but profits are greater for case 2. So, the discriminatory-pricing effect counteracts the economies-of-

Table 8 Fixed Fee per Distribution Companies (2002–2006) (2001 pesos) Region/Year Northwest Center-West Northeast Center South-Southeast

2002

2003

2004

2005

2006

31,296,289.9 106,691,620.1 87,298,224.1 626,430,086.1 35,221,909.7

32,700,628.6 111,479,125.7 92,175,782.61 645,143,036.1 36,133,183.9

34,189,512.9 116,880,510.5 97,400,425.42 664,703,017.4 37,082,076.5

35,767,760.4 123,594,822.2 102,995,685.2 685,145,915.9 38,070,036.5

37,440,454.6 130,783,986.7 108,986,655.6 706,509,060.8 39,098,565.3

2005

2006

Source: *Secretaria de Energía (2002)

Table 9 Profits per area (2001 pesos) Region/Year Northwest Center-West Northeast Center South-Southeast

2002 547,671,236 948,616,026 951,968,458 1,199,421,100 659,145,496

Source: *Secretaria de Energía (2002)

2003 553,803,306 958,408,358 959,962,574 1,216,103,487 665,640,655

2004 559,664,969 967,599,919 966,567,559 1,232,706,342 672,189,658

565,189,518 975,618,425 971,510,034 1,249,183,936 678,781,749

570,303,318 982,632,539 974,485,500 1,265,486,848 685,405,164

Repsol YPF-Harvard Kennedy School Fellows Research Papers

44

Figure 7 Variable fee for a monopolist with discriminatory tariffs (2002–2006) (2001 pesos)

results when a single firm owns the transmission network and charges an even two-part tariff across the country.42 A policy planner seeking to maximize welfare might then prefer this solution to the case of making a partition of the network into five

0.016 0.0145 0.014

Table 10 Comparisons in Profits, Capacity, and Network Expansion for the Four Cases

0.0126 0.012

Variable Fee

0.010

0.0097

0.008

Total Profits (pesos)

0.0080

0.006 0.004

0.0050

0.002 0.000 2000

2001

2002

2003

2004

scale effect and cause higher profits for case 2. Finally, comparison of case 2 and cased 3 reveals that profits, capacity and expansion are greater under case 2 than under case 3. This is completely explained by the discriminatory-pricing effect since for this last comparison the economies-ofscale effect is equal to zero. Case 1 then clearly provides the best results in terms of transmission capacity increase and transmission expansion, while case 2 results in larger profits. This means that the price-cap regulatory method provides the best network expansion

0.025

0.020

Pesos/KWh

0.015

0.010

0.005

0.000

2002

2003

0.00390683 0.00405370 Noroeste Centro-occidente 0.013893 0.013622 0.0024684 0.0024487 Noreste Centro 0.0181640 0.0190486 0.0016445 0.0016434 Sur-sureste

2004

2005

2006

0.00447672 0.013568 0.0024903 0.0188444 0.0016418

0.00418971 0.013658 0.0023544 0.0224148 0.0016409

0.00519390 0.012802 0.0024025 0.0221432 0.0016395

Network expansion (kilometers)

Case 1 fix no. 26,049,682,377 consumers

42,927,045,660.01 10,165.7231

Case 1 var. no 27,419,831,988 consumers

42,816,298,038.15 10,139.4966

Case 2

28,061,340,345

37,427,028,092.75

8,863.2423

Case 3

5,920,638,442

28,644,674,371.59

7,236.4919

firms, given that the profits obtained under this last case are only relatively slightly higher than in case 1. 5.

Figure 8 Fixed charge per area for a monopolist with discriminatory tariffs (2002–2001) (2001 pesos)

Capacity increase (MWh)

Concluding Remarks

I proposed a pricing method for incentives to expand the large interregional transmission links of the Mexican transmission network. This method must be understood within a combined merchant-regulatory framework where long term financial transmission rights are used within the 32 regions with meshed networks. The relatively radial nature of the interregional links makes sensible the use of the Vogelsang (2001) price-cap model. In this paper, I make an initial implementation assessment of this model under the restrictive assumption that the whole tranmsission network does not have any mesh. Results suggest that the best institutional structure for expanding the Mexican transmission grid would be one of a single transmission firm that charges even tariffs along the Mexican territory. I also discussed another pricing methodology based on the benefit-factors approach by RubioOdériz and Pérez-Arriaga (2000). Both this

Juan Rosellón

45

approach and ours seek different procedures to allocate the fixed cost of electricity transmission. The benefit factors approach defines a complementary charge in terms of the benefits to consumers from an expansion project, while my method relies in the rebalancing of the fixed and variable fees in order to provide adequate incentives for transmission investment. I believe that the former approach presents serious implementation hurdles due to the subjectivity in the allocation of “benefits.” It must also be recognized that Vogelsang’s method is difficult to implement due to the difficulty in defining the electricity transmission output (or throughput). However, the results of my simulations suggest that combined with a merchant structure, based on long term financial transmission rights, a two-part tariff regulatory model could promote expansion of the large interregional links of the Mexican transmission network. This assertion must of course be demonstrated in a more real model that actually uses LTFTRs for the meshed networks, and incentive regulation for the real large radial interregional links. In such a model, the issue of loop flows would be handled by an independent system operator that requires a payback constraint from the builders of the large regulated projects. Simulations would then be carried out for the actual links of the electricity regions together with a detailed definition of the importing and exporting areas. Another possibility could be to redefine the Vogelsang’s model in the context of an economic dispatch model so that the output of transmission is specified according to the implied incremental FTRs of new transmission links. The variable charges would depend on nodal price differences, and the fix (access) charge would be defined according to new transmission consumers.

A. Fixed cost is the sum of + cost for the use of the transmission infrastructure + transmission cost associated to power losses in transmission + generation costs implied by power losses in transmission. These last terms are given by the following equations: Cost for the use of the transmission infrastructure (CTser) CTser = CT*rser

where: CT = long-run total incremental cost of the network wj = cost per unit of capacity of transmission link j fsinj = the maximum power flow in link j (in absolute value) between the peak demand and the minimum demand scenarios, when the demanded transmission service is NOT considered fconj = the maximum power flow in link j (in absolute value) between the peak demand and the minimum demand scenarios, when the demanded transmission service is considered J = set of transmission elements that operate at tensions greater than or equal to 69 Kv (calculation of power flows is carried out in an AC model) Transmission cost associated to power losses in transmission

ANNEX 1 where: According to the MW-Mile method used by CFE, the Transmission price for tensions greater than or equal to 69 Kv is given by: Maximum {(Fixed cost + variable cost), minimum O & M cost} + Administrative costs

(A.1.1)

is the monthly cost of transmission capacity for tension level “v”, in region “a”

∆Pserva is the increment (or decrease) in losses at tension level “v”, in region “a”, due to the demanded transmission service. It is given by:

46

Repsol YPF-Harvard Kennedy School Fellows Research Papers

EP is the energy transmitted to all the load nodes of the demanded transmission service, during the billing month where: Pconj is the maximum power loss in the transmission element “j”, between the peak demand and the minimum demand scenarios, when the demanded transmission service is considered Psinj is the maximum power loss in the transmission element “j”, between the peak demand and the minimum demand scenarios, when the demanded transmission service is NOT considered Jva set of transmission elements at tension level “v”, in region “a” Generation costs implied by power losses in transmission

CMCgen is the monthly capacity cost of generation

Ω serva is the increment (or decrease) in losses at tension level “v”, in region “a”, due to the demanded transmission service, estimated at peak demand. It is given by:

Pmconj is the power loss in the transmission element “j”, which results from the peak demand scenario, when the demanded transmission service is considered Pmsinj is the power loss in the transmission element “j”, which results from the peak demand scenario, when the demanded transmission service is NOT considered Jva set of transmission elements at tension level “v”, in region “a” B. The variable cost (CVUR) in A.1.1 is given by:

nd is the number of days during the billing month PC is the transmission capacity for all the load nodes ENERatv is the energy cost in period “t”, in region “a”, at the tension level “v”

∆ESatv is the increment in transmission losses, due to the demanded transmission service, in period “t”, in region “a”, at the tension level “v”. It is given by:

Tat is the number of hours in period “t”, in region “a” Pconjet is the power loss in transmission element “j”, in the demand scenario “e t”, of the period “t”, when the demanded transmission service is considered Psinjet is the power loss in transmission element “j”, in the demand scenario “e t”, of the period “t”, when the demanded transmission service is NOT considered Na is the number of periods in region “a” C. The minimum operation costs (CMIN) in A.1.1 are given by:

where: CMIN is the minimum cost for the demanded transmission services ETPR is the energy transmitted measured at the load nodes at tension levels greater than or equal to 69 Kv M = mba*fad

FC is the transmission load factor observed during the billing month; it is given by:

Mba is the base charge in pesos per KWH. It is calculated as the quotient of annual O&M transmission costs of the previous year and the transmitted KWH through the network in the previous year fad is an distance adjustment factor and it is given by:

Juan Rosellón

47

Endnotes

where:

is the equivalent distance of the transmission service (kilometers)

is the equivalent distance of the system (kilometers) f j is the flow in the transmission element “j”, calculated without including the transmission service lj is the length of element “j” (1 for transformers)

∆ fj is the value of the change in the flow in the transmission element “j”, to the demanded transmission service.

1. Secretaría de Energía (2002). 2. As in Rubio-Odériz and Pérez-Arriaga (2000). 3. Pidiregas are contracts for public projects that the Mexican government bids to private firms, and that are intertemporally paid with public funds. Final ownership is public, and private investors are supposed to independently fund such projects. During June 2003, CFE signed contracts for USD $147 million with the Spanish-French consortia (Grupo Isolux and Alstom SA) for construction of 831 km of 400 kv, 230 kv and 115 kv lines, and eight substations in the northern states of Durango, Chihuahua, and Sonora, and 142 km of 115 kv lines and 10 substations in Aguascalientes, Guanajuato, Jalisco and Nayarit. 4. See Carreón and Rosellón (2002) 5. “Non-programmable” generators are small producers that only supply power according to a previously set energy delivery schedule. Hydro generators also make available all their generation capacity, and face production constraints in the oneday-ahead market. Both types of generators then have zero variable costs. 6. The implementation details of this method are shown in annex 1. 7. As described by Schweppe et al (1988) 8. Rotger and Felder (2001) analyze the naturalmonopoly status of electricity transmission. They argue that there are new technologies that could make transmission a contestable activity. This new technologies include: flexible AC transmission systems (FACTS), next-generation controllable highvoltage DC lines, undeground and submarine cables and low-impact cable installation techniques, as well as applications of superconducting materials. 9. See Joskow and Tirole (2000), and Léautier (2001). 10. These assumptions include: no increasing returns to scales, no sunk costs, nodal prices are able to fully reflect consumers’ willingness to pay, the network externalities are internalized by nodal prices, there is no uncertainty over congestion rents, there is no market power so that markets are always cleared by prices, there exists a full set of future markets, and the ISO has no internal intertemporal preferences regarding effective transmission capacity. 11. No restructured electricity sector in the world has adopted a pure merchant approach towards transmission expansion. The closest case is Australia where a mixture of regulated and merchant approaches has been recently implemented. Pope (2002), and Harvey (2002), recently propose LTFTR auctions for the New York ISO that provide a hedge against congestion cots. Gribik et al (2002) propose

Repsol YPF-Harvard Kennedy School Fellows Research Papers

48

12.

13.

14. 15. 16.

17.

18.

19.

20. 21. 22. 23.

24. 25.

26.

27.

an auction method based on the physical characteristics (capacity and admittance) of the transmission network. As shown in an extensive body of literature, generators can better exert local power when the transmission network is congested. See Bushnell, 1999, Bushnell and Stoft, 1997, Joskow and Tirole, 2000, Oren, 1997, Joskow and Schmalensee, 1983, Chao and Peck, 1997, Gilbert, Neuhoff, and Newbury, 2002, Cardell, Hitt, and Hogan, 1997, Borenstein, Bushnell, and Stoft, 1998, Wolfram, 1998, and Bushnell and Wolak, 1999. A set of FTRs is simultaneously feasible if the associated net power flows are also simultaneously feasible. See Hogan (2002a), Hogan (2002b), and Hogan (2003) See London Economics International (2002), chapters 3 and 5. In the US, transmission fixed costs are recuperated through a regulated fixed charge, even in those systems that rely on nodal pricing and FTRs. Typically, this charge is regulated by a cost of service methodology. See also Rotger and Felder (2001) for a discussion on the use of planning and regulation as a backstop for market based transmission. For Transpower (2002) sunk costs are basically capital costs that are “unavoidable” in the long run, while fixed costs are operating and maintenance costs that are “avoidable” in the long term. Bushnell and Stoft (1997), and Hogan (2002b) show that the definition of the beneficiaries of a transmission expansion is not an easy task. This method is design for consumers above a tension of 69 KV. See Instituto de Investigaciones Eléctricas (2003) In 2000, the CFE’s monthly total transmission revenue was around USD 0.5 billion. There is not an objective way to determine a “fair” amount of benefits for each user of the transmission network, especially when market agents have a wide portfolio for their energy purchases and sales. See Rubio-Odériz and Pérez-Arriaga (200), p. 451. In such a case, reduction of social welfare could only be reverted by making the agent that generates the negative externality (the one that expands the transmission network) pay to the affected agents (those holding original FTRs). See Bushnell and Stoft (1997). Then, as in Rotger and Felder (2001), the price-cap regulatory mechanism would be used as a backstop for expansion projects that are not developed by the merchant scheme. Large links are modeled using a DC approximation.

28. Kristiansen and Rosellón (2003) propose a concrete merchant mechanism that is designed to be applied to small line increments in meshed transmission networks. 29. More generally, such a feasibility test also checks out the economic, security and reliability impact of the project. 30. The initial price cap would be the result of the competitive bidding. Therefore, avoiding most of the drawbacks of PBR regulation discussed in Rotger and Felder (2002). 31. Note also that through this modeling strategy I avoid (or delay) the difficult problem of defining the output of transmission under loop flows (as pointed out by Hogan, 2002b). 32. See Vogelsang 2001. The setting of my model works under such an assumption, and an application of a merchant-regulatory mechanism to the actual links of the Mexican transmission network should determine variable charges from nodal prices. However, the simulations in this paper do not use nodal prices to determine variable charges due to the level of abstraction of the models. 33. Professor Vogelsang pointed out to me that his pricing mechanism might be directly compared to the pricing mechanism in Rubio-Odériz and PérezArriaga (2000). The latter bases benefits on last period’s quantities, and calculates benefits as cost savings for consumers and revenue increases for generators. This is similar to the Slutzky approximations to welfare increases used in Vogelsang (2001). Also, the total amount to be distributed in the complementary charge could be based on last period’s numbers, so that the Rubio-Odériz and PérezArriaga mechanism would be a Vogelsang-Finsinger mechanism with weak cost-reducing and investment incentives (as in Vogelsang and Finsinger, 1979). However, if the complementary charge is based on current cost data it would produce zero profits and would therefore have no cost-reducing or investment incentives. 34. It is important to stress that I sometimes found local maximums that, even though they satisfied the constraints imposed to the model, they did not provide the optimal benefit for the regulated firm. I therefore review in detail all the data so as to make sure that a global maximum was reached. 35. For the estimation of the demand function, parameter β was obtained through a price-quantity regression. Parameter α was taken from estimations carried out by the Secretaría de Energía regarding the behavior of electricity demand, both at the national level as for each of the regions included in the national electricity system. 36. I use Laspeyres weights for my simulation.

Juan Rosellón

49

Annex 2 Sells per area (GWh) Area Baja California South Baja California Central Northeast Northwest North Occidental Oriental Peninsular TOTAL

1989

1991

1993

1995

1997

1998

1999

3,640 610 22,062 13,479 6,796 7,280 16,966 15,584 2,073 88,490

3,849 634 22,424 14,760 7,359 7,274 19,572 16,304 2,541 94,717

4,129 626 24,355 16,274 7,641 7,790 21,376 16,166 2,869 101,226

4,870 691 25,289 18,675 8,561 9,087 24,389 18,514 3,233 113,309

6,184 845 27,971 22,209 9,872 10,264 27,986 21,198 3,632 130,161

6,347 863 29,026 23,746 10,020 11,113 29,724 22,337 3,961 137,137

7,020 944 30,208 25,629 10,541 11,701 31,724 22,983 4,169 144,919

SOURCE: Secretaria de Energía, 2000.

Average unit cost per transmission line kilometer (2001 pesos) Type of line 400 kV two circuits, 3 phase conductors 400 kV one circuit, 3 phase conductors 400 kV two circuits, 2 phase conductors 400 kV two circuits, 2 phase conductors 230 kV two circuits, 1113 MCM 230 kV one circuit, 1113 MCM 230 kV two circuits, 900 MCM 203 kV one circuit, 900 MCM 115 kV two circuits, 795 MCM 115 kV one circuit, 795 MCM 115 kV two circuits, 477 MCM 115 kV one circuit, 477 MCM

Direct cost

Direct cost plus indirect cost

3,873,088 2,159,550 2,976,352 1,689,926 1,808,795 1,119,235 1,653,380 1,031,639 1,331,443 850,744 1,096,962 729,271

4,376,590 2,440,291 3,363,278 1,909,616 2,043,938 1,264,736 1,868,319 1,165,752 1,504,531 961,341 1,239,567 824,076

Source: Comisión Federal de Electricidad, 2000.

37. Annex 2 presents a summary of these data. 38. This consumer growth rate is equal to the country’s growth rate in dwellings. It has been recently used by the Secretaría de Energía in its forecast studies. See Secretaría de Energía (2002), p. 56. 39. See Secretaría de Energía (2002). 40. In reality, the Secretaría de Energía only defines 13 distribution zones. 41. The regional tariffs of last section would apply to these two regions. 42. This result might be in conflict with studies that foresee better welfare results for a firm that is allowed to charge discriminatory two-part tariffs (as in Bertoletti, P., and C. Poletti, 1997). However, the myopic profit-maximizing nature of my model explains the possibility of such a result. It then remains as a future research question the behavior of a non-myopic firm under the proposed price cap.

References Bertoletti, P., and C. Poletti (1997). Welfare Effects of Discriminatory Two-Part Tariffs Constrained by Price Caps. Economics Letters 56, pp. 292–298. Borenstein, S., J. Bushnell, and S. Stoft (1998) The Competitive Effects of Transmission Capacity in a Deregulated Electricity Industry. POWER Working Paper PWP-040R. University of California Energy Institute (http://www.ucei.berkely.edu/ucei). Bushnell, J. (1999) Transmission Rights and Market Power. The Electricity Journal, 77–85. Bushnell, J. B., and F. Wolak (1999) Regulation and the Leverage of Local Market Power in the California Electricity Market. POWER Working Paper PWP070R. University of California Energy Institute (http://www.ucei.berkely.edu/ucei). Bushnell, J. B., and S. E. Stoft (1997). Improving Private Incentives for Electric Grid Investment. Resource and Energy Economics 19, 85–108.

50

California ISO and London Economics International (2003),”A Proposed Methodology for Evaluating the Economic Benefits of Transmission Expansions in a Restructured Wholesale Electricty Market,” mimeo, (http://www.caiso.com/docs/2003/03/25/200303 2514285219307). Cardell, C., C. Hitt, and W. Hogan (1997) Market Power and Strategic Interaction in Electricity Networks. Resource and Energy Economics, 109–137. Carreón, V. y J. Rosellón (2002). La Reforma del Sector Eléctrico Mexicano: Recomendaciones de Política Pública. Gestión y Política Pública, Vol. XI, No. 2, 2002. Chao, H.-P., and S. Peck (1997) An Institutional Design for an Electricity Contract Market with Central Dispatch. The Energy Journal 18(1), 85–110. Comisión Federal de Electricidad (2000). A Market Power Analysis of Generation in Mexico. Mimeo. Gilbert, R., K. Neuhoff, and D. Newbery (2002) Mediating Market Power in Electricity Networks. Mimeo. Grande, O. S., and I. Wangensteen (2000). Alternative Models for Congestion Management and Pricing Impact on Network Planning and Physical Operation. CIGRE, Paris, aug/sept. Gribik, P. R., J. S. Graves, D. Shirmohammadi, and G. Kritikson (2002) Long Term Rights for Transmission Expansion. Mimeo. Hartley, P. and E. Martínez-Chombo (2002). Electricity Demand and Supply in Mexico. Mimeo, Rice University. Harvard Electricity Policy Group (2002a). Beyond Slicing and Dicing: Incentives for Transmission Owners. Rapporteur’s Summaries of HEPG Twenty-Eight Plenary Session, Session Three, May 30, 31. Harvard Electricity Policy Group (2002b). Transmission Expansion: Market Based and Regulated Approaches. Rapporteur’s Summaries of HEPG Twenty-Seventh Plenary Sessions, Session Two, January 24–25. Harvey, S. M. (2002) TCC Expansion Awards for Controllable Devices: Initial Discussion. Mimeo. Hogan, W. (1992) Contract Networks for Electric Power Transmission. Journal of Regulatory Economics, 4, 211–242. Hogan, W. (1999). Market-Based Transmission Investments and Competitive Electricity Markets, Mimeo, JFK School of Government, Harvard Electricity Policy Group, Harvard University, http://www.ksg .harvard.edu/people/whogan Hogan, W. (2002a). Financial Transmission Right Formulations. Mimeo, JFK School of Government, Harvard Electricity Policy Group, Harvard University, http://www.ksg.harvard.edu/people/whogan. Hogan, W. (2002b). Financial Transmission Right Incentives: Applications Beyond Hedging. Presentation

Repsol YPF-Harvard Kennedy School Fellows Research Papers

to HEPG Twenty-Eight Plenary Sessions, May 31, http://www.ksg.harvard.edu/people/whogan. Hogan, W. (2003). Transmission Market Design. Mimeo, JFK School of Government, Harvard Electricity Policy Group, Harvard University, http://www.ksg .harvard.edu/people/whogan. Instituto de Investigaciones Eléctricas (2003). Metodología de Asignación de los Costos de Transmisión con Base en los Costos Marginales de Corto Plazo de la Energía y en los Beneficios a los Usuarios. Mimeo. Joskow, P. and J. Tirole (2000). Transmission Rights and Market Power on Electric Power Networks. RAND Journal of Economics, vol. 31, no. 3, Autumn, pp. 450–487. Joskow, P. and J. Tirole (2002). Transmission Investment: Alternative Institutional Frameworks. Mimeo. Joskow, P. and J. Tirole (2003). Merchant Transmission Investment. Mimeo. Joskow, P. and R. Schmalensee (1983) Markets for Power: An analysis of Electric Utility Deregulation, MIT Press. Kristiansen, T. and J. Rosellón (2003). A Merchant Mechanism for Electricity Transmission Expansion. 23rd IAEE North American Conference Proceedings. Léautier, T.-O. (2000). Regulation of an Electric Power Transmission Company. The Energy Journal, vol. 21, no. 4, pp. 61–92. Léautier, T.-O. (2001). Transmission Constraints and Imperfect Markets for Power. Journal of Regulatory Economics, 19(1), 27–54. London Economics International (2002). Final Methodology: Proposed Approach for Evaluation of Transmission Investment. Prepared for the CAISO. Mimeo. Madrigal, M. (2000). Escenarios de Estructura Competitiva del Mercado Eléctrico Mexicano, Mimeo presentation. Madrigal, M., de Rosenzweig, F., G. Gutiérrez, and J.C. Femat (2002), Predicting Structural Market Power in the Mexican Electricity System: An Optimization Based Approach. Presentation at INFORMS Annual Meeting, San José California, November 19. Pérez-Arriaga, J. I., F. J. Rubio and J. F. Puerta Gutiérrez et al. (1995). Marginal Pricing of Transmission Services: An Analysis of Cost Recovery. IEEE Transactions on Power Systems, vol. 10, no. 1, February. Pope, S. (2002) TCC Awards for Transmission Expansions. Mimeo. Rotger, J., and F. A. Felder (2001). Reconciling MarketBased Transmission and Transmission Planning. The Electricity Journal, Volume 14, Number 9, November. Rubio-Odériz, J., and I. J. Pérez-Arriaga (2000). Marginal Pricing of Transmission Services: A Comparative Analysis of Network Cost Allocation Methods.

Juan Rosellón

IEEE Transactions on Power Systems, vol. 15, no. 1, February Schweppe, F., M. Caramanis, R. Tabors, and R. Bohn (1988). Spot Pricing of Electricity, Norwell, MA: Kluwer. Secretaría de Energía (2002), Prospectiva del Sector Eléctrico 2002–2011, México. Sheffrin, A., and F. A. Wolak (2001). Methodology for Analyzing Transmission Upgrades: Two Alternative Proposals. Mimeo Shirmohammadi, D. et al (1989). Evaluation of Transmission Network Capacity Use for Wheeling Transactions. IEEE Transactions on Power Systems, vol. 4, no. 4, November. Transpower (2002). Confirmed Pricing Methodology. Final Design Principles. Mimeo. Vogelsang, I. (2001). Price Regulation for Independent Transmission Companies. Journal of Regulatory Economics, vol. 20, no. 2, September.

51

Vogelsang, I., Finsinger, J. (1979). A Regulatory Adjustment Process for Optimal Pricing by Multiproduct Monopoly Firms. The Bell Journal of Economics. Vol. 10 (1). p 157–71. Spring. Wilson, R. (2002). Architecture of Power Markets. Econometrica, vol. 70, No. 4, pp. 1299–1340. Wolak, F. A. (2000). An Empirical Model of the Impact of Hedge Contract on Bidding Behavior in a Competitive Electricity Market. International Economic Journal, Summer, 1–40. Wolfram, C. (1998) Strategic Bidding in a Multi-Unit Auction: An Empirical Analysis of Bids to Supply Electricity in England and Wales. The RAND Journal of Economics 29: 703–725.

2

A Merchant Mechanism for Electricity Transmission Expansion Tarjei Kristiansen Norwegian University of Science and Technology Juan Rosellón Centro de Investigación y Docencia Económicas (CIDE) and Harvard University

Abstract

flows” that come up from complex network interactions.1 The effects of loop flows imply that transmission opportunity costs are a function of the marginal costs of energy at each location. Power costs and transmission costs depend on each other since they are simultaneously settled in electricity dispatch. Loop flows imply that certain transmission investments might have negative externalities on the capacity of other (perhaps distant) transmission links (see Bushnell and Stoft, 1997). Moreover, the addition of new transmission capacity can sometimes paradoxically decrease the total capacity of the network (Hogan, 2002a). The welfare effects of an increment in transmission capacity are analyzed by Léautier (2001). The welfare outcome of an expansion in the transmission grid depends on the weight in the welfare function of the generators’ profits relative to the consumers’ utility weight. Incumbent generators in load pockets are not in general the best agents to carry out transmission expansion projects. Even though an increase in transmission capacity might allow them to increase their revenues due to increased access to new markets and higher transmission charges, such gains are usually overcome by the loss of their local market power. The literature on incentives for long-term expansion of the transmission network is scarce. The economic analysis of electricity markets has been reduced to short-run issues, and has typically assumed that transmission capacity is fixed (see Joskow and Tirole, 2003). However, transmission capacity is random in nature, and it jointly depends on generation investment. The way to solve transmission congestion in the short run is well known. In a power flow model, the price of transmission congestion is determined by the difference in nodal prices (see Hogan, 1992, 2002b). Yet, there is no consensus with respect to the method to attract investment to

We propose a merchant mechanism to expand electricity transmission based on long-term financial transmission rights (FTRs). Due to network loop flows, a change in network capacity might imply negative externalities on existing transmission property rights. The system operator thus needs a protocol for awarding incremental FTRs that maximize investors’ preferences, and preserves certain currently unallocated FTRs (or proxy awards) so as to maintain revenue adequacy. In this paper we define a proxy award as the best use of the current network along the same direction as the incremental awards. We then develop a bi-level programming model for allocation of long-term FTRs according to this rule and apply it to different network topologies. We find that simultaneous feasibility for a transmission expansion project crucially depends on the investorpreference and the proxy-preference parameters. Likewise, for a given amount of pre-existing FTRs the larger the current capacity the greater the need to reserve some FTRs for possible negative externalities generated by the expansion changes. 1.

Introduction

The analysis of incentives for electricity transmission expansion is not easy. Beyond economies of scale and cost sub-additivity externalities in electricity transmission are mainly due to “loop

We are grateful to William Hogan for very useful suggestions and discussions. We also thank an anonymous referee for insightful comments. The second author prepared this paper in residence under the Repsol YPFHarvard Kennedy School Fellows program. Additional support provided by the CRE and the Fundación México en Harvard. 53

54

finance the long-term expansion of the transmission network, so as to reconcile the dual opposite incentives to congest the network in the short run, and to expand it in the long run. Incentive structures proposed to promote transmission investment range from a “merchant” mechanism, based on long-term financial transmission right (LTFTR) auctions (as in Hogan, 2002a), to regulatory mechanisms that charge the transmission firm the social cost of transmission congestion (see Léautier 2000, Vogelsang, 2001, and Joskow and Tirole, 2002). In practice, regulation has been used in England, Wales and Norway to promote transmission expansion, while a combination of planning and auctions of long-term transmission rights has been tried in the Northeast of the U.S. A mixture of regulatory mechanisms and merchant incentives is alternatively used in the Australian market. In this paper we develop a merchant model to attract investment to small-scale electricity transmission projects based on LTFTR auctions. Locational prices give market players incentives to initiate transmission investments. FTRs provide transmission property rights, since they hedge the market player against future price differences. Our model further develops basic conditions under which FTRs and locational pricing provide incentives for long-term investment in the transmission network. In meshed networks, a change in network capacity might imply negative externalities on transmission property rights. Then, in the process of allocation of incremental FTRs, the system operator must reserve certain unallocated FTRs so that the revenue adequacy of the transmission system is preserved. In order to deal with this issue, we develop a bi-level programming model for allocation of long-term FTRs and apply it to different network topologies. The structure of the paper is as follows. In section 2 we carry out an analytical review on the relevant literature on electricity transmission expansion. In section 3 we develop our model. We first introduce FTRs and the feasibility rule, and then address the rationale for FTR allocation and efficient investments. We develop general optimality conditions as well. In section 4, we carry out applications of our model to a radial line, and to a threenode network. Next, we describe the welfare implications in section 5. In section 6 we provide concluding comments.

Repsol YPF-Harvard Kennedy School Fellows Research Papers

2.

Literature review

There exist some hypotheses on structures for transmission investment: the market-power hypothesis, the incentive-regulation hypothesis, and the long-run financial-transmission-right hypothesis. The first approach seeks to derive optimal transmission expansion from the powermarket structure of power generators, and takes into account the conjectures of each generator regarding other generators’ marginal costs due to the expansion (Sheffrin and Wolak, 2001, Wolak, 2000, and The California ISO and London Economics International, 2003). The generators’ bidding behaviors are estimated before and after a transmission upgrade, and a real-option analysis is used to derive the net present value of transmission and generation projects together with the computation of their joint probability. The model shows that there are few benefits of transmission expansion until added capacity surpasses a certain threshold that, in turn, is determined by the possibility of induced congestion by the strategic behavior of generators with market power. The generation market structure then determines when transmission expansion yield benefits. Additionally, many small upgrades of the transmission grid are preferable to large greenfield projects when cost uncertainty is added to the model. The contribution of this method is that it models the existing interdependence of transmission investment and generation investment within a transportation model with no network loop flows. However, as pointed out by Hogan (2002b), the use of a transportation model in the electricity sector is inadequate since it does not deal with discontinuities in transmission capacity implied by the multidimensional character of a meshed network. The second method for transmission expansion is a regulatory alternative that relies on a “Transco” that simultaneously runs system operation and owns the transmission network. The Transco is regulated through benchmark regulation or price regulation so as to provide it with incentives to invest in the development of the grid, while avoiding congestion. Léautier (2000), Grande and Wangesteen (2000), and Harvard Electricity Policy Group (2002) discuss mechanisms that compare the Transco performance with a measure of welfare loss due to its activities.

Tarjei Kristiansen and Juan Rosellón

Joskow and Tirole (2002) propose a surplus-based mechanism to reward the Transco according to the redispatch costs avoided by the expansion, so that the Transco faces the complete social cost of transmission congestion. Another regulatory alternative is a two-part tariff cap proposed by Vogelsang (2001) that addresses the opposite incentives to congest the existing transmission grid in the short run, and to expand it in the long run. Incentives for investment in expansion of the network are achieved through the rebalancing of the fixed part and the variable part of the tariff. This method tries to deepen into the analysis of the cost and production functions for transmission services, which are not very well understood in the economics literature. Nonetheless, to achieve this goal Vogelsang needs to define an output (or throughput) for the Transco. As argued in the FTR literature (Bushnell and Stoft (1997), Hogan, (2002a), Hogan, (2002b)), this task is very difficult since the average physical flow through a meshed transmission network is not well defined. The third approach is a “merchant” one based on LTFTR auctions by an independent system operator (ISO). This method deals with loop-flow externalities in that, to proceed with line expansions, the investor pays for the negative externalities it generates. To restore feasibility, the investor has to buy back sufficient transmission rights from those who hold them initially, or the ISO retains some unallocated transmission rights (proxy awards) during the LTFTR auction to protect unassigned rights while simultaneous feasibility of the system protects the rights of the existing FTR holders. This is the core of an LTFTR auction (see Hogan, 2002 a). Joskow and Tirole (2003) criticize the LTFTR approach. They argue that the efficiency results of the short-run version of the FTR model rely on perfect-competition assumptions, which are not real for transmission networks. Moreover, defining an operational FTR auction is technically difficult2 and, according to these authors, the FTR analysis is static (a contradiction with the dynamics of transmission investment). Joskow and Tirole analyze the implications of eliminating the perfect competition assumptions of the FTR model. First, market power and vertical integration might impede the success of FTR auctions. Prices will not reflect the marginal cost of production in

55

regions with transmission constraints. Generators in constrained regions will then withdraw capacity in order to increase their prices, and will overestimate the cost-saving gains from investments in transmission.3 Second, lumpiness in transmission investment makes the total value paid to investors through FTRs less than the social surplus created. The large and lumpy nature of major transmission upgrades requires long-term contracts before making the investment, or temporal property rights for the incremental investment. Third, contingencies in electricity transmission impede the merchant approach to really solve the loop-flow problem. Moreover, existing transmission capacity and incremental capacity are stochastic. Even in a radial line, realized capacity could be less than expected capacity and the revenue-adequacy condition would not be met. Even more, the initial feasible FTR set can depend on random exogenous variables. Fourth, an expansion in transmission capacity might negatively affect social welfare (as shown by Bushnell and Stoft, 1997). Fifth, a moral hazard “in teams” problem arises due to the separation of transmission ownership and system operation in the FTR model. For instance, an outage can be claimed to be the consequence of poor maintenance (by the transmission owner) or of negligent dispatch (by the system operator).4 Additionally, there is no perfect coordination of interdependent investments in generation and transmission, and stochastic changes in supply and demand conditions imply uncertain nodal prices. Likewise, there is no equal access to investment opportunities since only the incumbent can efficiently carry out deepening transmission investments. Hogan (2003) responds to the above criticisms by arguing that LTFTRs only grant efficient outcomes under lack of market power, and nonlumpy marginal expansions of the transmission network. Furthermore, Hogan argues that regulation has an important role in fostering large and lumpy projects, and in mitigating market power abuses. As argued by Pérez-Arriaga et al (1995), revenues from nodal prices typically recover only 25% of total costs. LTFTRs should then be complemented with a fix-price structure or, as in RubioOdériz and Pérez-Arriaga (2000) a complementary

Repsol YPF-Harvard Kennedy School Fellows Research Papers

56

charge that allows the recovery of fixed costs.5 This fact is recognized by Hogan (1999) who states that complete reliance on market incentives for transmission investment is undesirable. Rather, Hogan (2003) claims that merchant and regulated transmission investments might be combined so that regulated transmission investment is limited to projects where investment is large relative to market size, and lumpy so that it only makes sense as a single project as opposed as to many incremental small projects. Hogan also responds to contingency concerns.6 On one hand, only those contingencies outside the control of the system operator could lead to revenue inadequacy of FTRs, but such cases are rare and do not represent the most important contingency conditions. On the other hand, most of remaining contingencies are foreseen in a securityconstrained dispatch of a meshed network with loops and parallel paths. If one of “n” transmission facilities is lost, the remaining power flows would still be feasible in an “n-1” contingency constrained dispatch. Hogan (2003) also assumes that agency problems and information asymmetries are part of an institutional structure of the electricity industry where the ISO is separated from transmission ownership and where market players are decentralized. However, he claims that the main issue on transmission investment is the decision of the boundary between merchant and regulated transmission expansion projects. He argues that asymmetric information should not necessarily affect such a boundary. Hogan (2002a) finally analyzes the implications of loop flows on transmission investment raised by Bushnell and Stoft (1997). He analytically provides some general axioms to properly define LTFTRs so as to deal with negative externalities implied by loop flows. We next present a model that develops the general analytical framework suggested by Hogan (2002a). 3.

ures.7 Also assume that transmission projects are incrementally small (relative to the total network) and non-lumpy so that the project does not imply a relatively large change in nodal-price differences. However, although projects are small, they might change or not the power transfer distribution factors (PTDFs) of the network.8 Under an initial condition of non-fully allocation of FTRs in the grid, the auctioning of incremental LTFTRs should satisfy the following basic criteria in order to deal with possible negative externalities associated with the expansion (1) An LTFTR increment must keep being simultaneously feasible (feasibility rule). (2) An LTFTR increment remains simultaneously feasible given that certain currently unallocated rights (or proxy awards) are preserved. (3) Investors should maximize their objective function (maximum value). (4) The LTFTR awarding process should apply both for decreases and increases in the grid capacity (symmetry). The need for proxy awards arises whenever there is less than full allocation of the capacity of the existing grid. This occurs prominently during a transition to an electricity market when there is reluctance to fully allocate the existing grid for all future periods. Hence FTRs for the existing grid are short term (this period), but investors in grid expansion seek long term rights (next period). Full allocation of the existing grid seems necessary but not sufficient for defining and measuring incremental capacity. Hogan explains though that defining proxy awards is a difficult task. We next address this issue in a formal way in the context of an auction model designed to attract investment for transmission expansion. 3.1 The power flow model and proxy awards Consider the following economic dispatch model:9

The model

Assume an institutional structure where there are various established agents (generators, Gridcos, marketers, etc.) interested in the transmission grid expansion. Agents do not have market power in their respective market or, at least, there are in place effective market-power mitigation meas-

(1)

Y = d− g

(2) (3)

K (Y, u) ≤ 0

(4)

Tarjei Kristiansen and Juan Rosellón

57

where d and g are load and generation at the different locations. The variable Y represents the real power bus net loads, including the swing bus S – (YT = (Ys,Y T ). B(d-g) is the net benefit function,10 and τ is a unity column vector, τ T = (1,1,...,1). All other parameters are represented in the control variable u. The objective Equation includes the maximization of benefit to loads and the minimization of generation costs. Equation denotes the net load as the difference between load and generation. Equation is a loss balance constraint where L(Y,u) is a vector which denotes the losses in the network. In equation K(Y,u), is a vector of power flows in the lines, which are subject to transmission capacity limits. The corresponding multipliers or shadow prices for the constraints are (P, λ ref ,λ tran) for net loads, reference bus energy (or loss balance) and transmission constraints, respectively.11 The locational prices P are the marginal generation cost or the marginal benefit of demand, which in turn equals the reference price of energy plus the marginal cost of losses and congestion. With the optimal solution (d*,g*,Y*,u*) and the associated shadow prices, we have the vector of locational prices as:

(5) If losses12 are ignored, only the energy price at the reference bus and the marginal cost of congestion contribute to set the locational price. FTR obligations13 hedge market players against differences in locational prices caused by transmission congestion.14 FTRs are provided by an ISO, and are assumed to redistribute the congestion rents. The pay-off from these rights is given by: (6) where Pj is the price at location j, Pi is the price at location i, and Qij is the directed quantity injected at point i and withdrawn at point j specified in the FTR. The FTR payoffs can take negative, positive or zero values. A set of FTRs is said to be simultaneously feasible if the associated set of net loads is simultaneously feasible, that is if the net loads satisfy the loss balance and transmission capacity constraints as well as the power flow equations given by:

(7) f

where Σt k is the sum over the set of point-to-point k obligations.15 If the set of FTRs is simultaneously feasible and the system constraints are convex,16 then the FTRs satisfy the revenue adequacy condition in the sense that equilibrium payments collected by the ISO through economic dispatch will be greater than or equal to payments required under the FTR forward obligations.17 Assume now investments in new transmission capacity. The associated set of new FTRs for transmission expansion has to satisfy the simultaneous feasibility rule too. That is, the new and old FTRs have to be simultaneously feasible after the system expansion. Assume that T is the current partial allocation of long-term FTRs, then by assumption it is feasible (K(T,u) ≤ 0). Suppose there is to be a total possible incremental award, and that a fraction of the possible awards is reserved as proxy awards for the existing grid with the remainder provided to the incremental investor as representing the proportion that could only be awarded as a result of the investment. Let a be the scalar amount of incremental FTR awards, and tˆ the scalar amount of proxy awards. Furthermore let δ be directional vector18 such that aδ is the MW amount of incremental FTR awards, and tˆδ is the MW amount of proxy awards between different locations. Any incremental FTR award aδ should comply with feasibility rule in the expanded grid. Hence we must have K +(T + aδ,u) ≤ 0, where K + corresponds to capacity of the expanded grid. When certain currently unallocated rights (proxy awards) tˆδ in the existing grid must be preserved, combined with existing rights they sum up to T + tˆδ.19 Then K + should also satisfy simultaneous feasibility so that K(T + tˆδ,u) ≤ 0, K +(T + aδ,u) ≤ 0, and K +(T + tˆδ + aδ,u) ≤ 0 for incremental awards aδ. A question then arises regarding the way to best define proxy awards. One possibility is to define them as the “best use” of the current network along the same direction as the incremental awards.20 This includes both positive and negative incremental FTR awards. The best use in a threenode network may be thought of as a single incre-

Repsol YPF-Harvard Kennedy School Fellows Research Papers

58

mental FTR in one direction or a combination of incremental FTRs defined by the directional vector δ, depending on the investor preference. Hogan (2002a) suggests two ways of defining “best use”:

(8) In the preset proxy formulation the objective is to maximize the value (defined by prices p) of the proxy awards given the pre-existing FTRs, and the power flow constraints in the pre-expansion network. In the investor preference formulation the objective is to maximize the investor’s value (defined by the bid functions for different directions, β (aδ )) of incremental FTR awards given the proxy and pre-existing FTRs and the power flow constraints in the expanded network, while simultaneously calculating the minimum proxy scalar amount that satisfies the power flow constraints in the pre-expansion network. We will use as a proxy protocol the first definition. We next analyze the way to use this protocol to carry out an allocation of LTFTRs that stimulates investment in transmission.

In this model, the investor’s preference is maximized subject to the simultaneous feasibility conditions, and the best use protocol. We add a constraint on the (two-)norm21 of the directional vector to preclude the trivial case δ = 0. We want to explore if such an auction model approach can produce acceptable proxy and incremental awards. We next analyze this issue within a framework that ignores losses, and utilizes a DC-load flow approximation. The auction model is a nonlinear optimization problem of “bi-level” nature.22 There are two optimization stages. Maximization is non-myopic since the result of the lower problem (first stage) depends on the direction chosen in the upper problem (second stage).23 Bi-level problems may be solved by first transforming the lower problem (i.e. the allocation of proxy awards) into to a set of Kuhn-Tucker equations that are subsequently substituted in the upper problem (i.e. the maximization of the investors’ preference). The model can then be understood as a Stackelberg problem although it is not intending to optimize the same type of objective function at each stage.24 The Lagrangian (L) for the lower problem is:

where λT is the Lagrange multiplier vector associated with transmission capacity on the respective transmission lines before the expansion. It is the Lagrange multiplier of the simultaneous feasibility restriction for proxy awards. The Kuhn-Tucker conditions are:

3.2 The auction model Assume the preset proxy rule is used to derive prices that maximize the investor preference β (aδ ) for an award of a MWs of FTRs in direction δ. We then have the following auction maximization problem:

(9)

The transformed problem is then written as:

Tarjei Kristiansen and Juan Rosellón

59

(10) where ω, γ, θ, ζ, ε, ϕ, κ and π are Lagrange multipliers associated with each constraint. More specifically, ω is the shadow price of the simultaneous feasibility restriction for existing and incremental FTRs; γ is the shadow price of the simultaneous feasibility restriction for existing FTRs, proxy awards and incremental FTRs; θ, ζ, ε are the shadow prices of the restriction on optimal proxy FTRs; ϕ, κ are the shadow prices of the non-negativity constraints for a and λ, respectively; and π is the shadow price of the unit restriction on δ. The Lagrangian of the auction problem is:

(14)

(15)

(16)

(17)

(18) (11) where Ω = (ω,γ,θ,ζ,ε,ϕ,κ,π) denotes the vector of Lagrange multipliers. Kuhn-Tucker conditions for the upper problem are:

(19)

(20)

(21) (12)

(12)

(22)

(23)

(13)

(13)

(24)

(25)

(26)

Repsol YPF-Harvard Kennedy School Fellows Research Papers

60

(27)

The corresponding optimization problem is:

(28) The constraint

is redundant when the preset proxy preference (p) is non-zero, since it is a sub-gradient of the constraint and ε is therefore zero when p is non-zero. We show in a later example that θ and ϕ are zero because the associated constraints are redundant. The binding constraint in the lower level problem is , since some transmission constraints are fully utilized by proxy awards. This is a nonlinear and non-convex problem, and its solution depends on the investor-preference parameters, the current partial allocation (T), and the topology of the network prior to and after the expansion.25 A general solution method utilizing Kuhn-Tucker conditions would be through checking which of the constraints are binding.26 One way to identify the active inequality constraints is the active set method.27 In this paper we solve the problem in detail for different network topologies, including a radial line and a three-node network. 4.

Simulations

4.1 Radial line Let us first analyze a radial transmission line that is expanded as in Figure 1. Figure 1 An expanded line and its feasible expansion 1-2 FT R s Feasible expansion 1

C12 C12+

2

(29) where C12 is the transmission capacity of the net+ work before the expansion, C12 is the transmission capacity of the network after the expansion, and b12 is the investor preference. The first order conditions of the lower maximization problem can then be added as constraints to the upper problem:

(30) Since the grid is being expanded, the constraint on simultaneous feasibility of incremental FTRs T12 + + aδ12 ≤ C 12 is non-binding. The solution to this problem provides the values for the decision variables, and shadow prices.28 First, δ 12 = 1, because the network is being expanded. Additionally γ = b12 which implies that the higher the value of the investor-preference parameter b12 the more the investor values post-expansion transmission capacity (its marginal valuation of transmission capacity increases with the bid value). Similarly, we get λ = p12 , which implies that the higher the value of the preset proxy preference parameter p12 the higher marginal valuation of pre-expansion transmission capacity. Other results are θ = 0, ζ = γ / p12 = b12 / p12 and ε = 0. This was expected since only one restriction for the lower problem is binding because the two other are redundant. The value of the binding Lagrange multiplier equals the ratio between the investor’s bid value and the preset proxy parameter.

Tarjei Kristiansen and Juan Rosellón

61

It also follows that ϕ = 0 which is to be expected because the directional vector δ is non-zero. Furthermore, tˆ = C12 – T12 , which means that for given existing rights the higher the current capacity the larger the need for reserving some proxy FTRs for possible negative externalities generated by the expansion. Proxy awards are auctioned as a hedge against externalities generated by the expanded network. + + We finally get a = C12 – T12 – tˆ = C12 – C12 , which shows that the optimal amount of additional MWs of FTRs in direction δ directly depends on the amount of capacity expansion. Transmission capacity is in fact fully utilized by proxy awards (in the pre-expansion network), and by incremental FTRs (in the expanded network). Likewise, the investor receives a reward equal to the MW amount of new transmission capacity that it creates.

The network expansion problem for identical links and FTRs between buses 1–3 and 2–3 is formulated as:

(31)

4.2 Three-node network with two links We now consider a three-node network example from Bushnell and Stoft (1997) where there is an expansion of line 1–2. The network is illustrated in Figure 2 and the feasible expansion in Figure 3. Figure 2 Three-node network with expansion of line 1–2 900 MW MAX

1

3 (31)

200 MW MAX

900 MW MAX 2

Figure 3 Feasible expansion of FTRs FTR: 2->3 2000 1500

tˆ δ aδ

Reduction 2-1 Feasible Expansion

2-3

1000 T12 = 100, T23 = 800 500

1-2 1-3

T+aδ

500

1000

FTR: 1->3 1500

2000

Annex 2 presents the calculations to obtain the power transfer distribution factors (PTDFs) for the post expansion network. In Figure 3 the pre-existing FTRs in the direction 2–3 do not use the full capacity of the pre-expansion network and become infeasible after inserting line 1–2. The preference is for FTRs in the direction 1–3 for transmission expansion. As seen from Figure 3 the maximum amount of proxy and incremental FTRs in the direction 1–3 that can be obtained is 1100, and corresponds to the point where the 1–3 and 1–2 transmission capacity constraints intersect.

Repsol YPF-Harvard Kennedy School Fellows Research Papers

62

In solving this problem, we get:29

with

where γ 1 and γ 2 are the Lagrange multipliers associated with transmission capacity on the lines 1–3 and 1–2, respectively, in the expanded network, and ζ is the multiplier associated with the KuhnTucker condition regarding transmission capacity in the pre-expansion network for the line 1–3. This line has the Lagrange multiplier λ associated with it before expansion. So as to characterize the solution to our model, we now calculate the Lagrange multipliers and decision variables for particular parameter values. In particular, we find the solution for the allocation presented in Figure 3. We assume the following bid values, preset proxy preferences and pre-existing amount of FTRs:

From these parameters we find that the marginal value of transmission capacity on line 1–3 and line 1–2 are γ 1 = 39.6 and γ 2 = 33.6, respectively. Thus the investor values transmission capacity

on line 1–3 more than on line 1–2. We find that the product of the Kuhn-Tucker multiplier and the transmission capacity multiplier for the line 1–3 is ζλ = 37. Likewise, the values of the decision variables are calculated as:

The MW amount of awarded proxy FTRs in the direction 1–3 is tˆδ 13 = 800, and the amount of awarded incremental FTRs is aδ 13 = 200. The amount of incremental 1–3 FTRs corresponds to the new transmission capacity on line 1–2 that the investor has created. There is also an allocation of proxy FTRs such that the full capacity of line 1–3 is utilized. Similarly the proxy awards in direction 2–3 is tˆδ 23 = –240, and the amount of awarded incremental FTRs is aδ 23 = –60. The amount of incremental 2–3 FTRs is minimized and corresponds to 20% of the reduction (300) in pre-existing FTRs. The incremental 2–3 awards are mitigating FTRs, and are necessary to restore feasibility. The investor is then responsible for additional counterflows so that it pays back for the negative externalities it creates. The solution is indicated by the black arrow in Figure 3 and consists of both pre-existing and incremental FTR awards amounting to T13 + aδ 13 = 300 and T23 + aδ 23 = 740. The allocation of incremental 2–3 FTRs is minimized because the model takes into account that one line is expanded, and some of the pre-existing FTRs become infeasible after the expansion. This illustrates that the amount of incremental FTRs in the preference direction must be greater than zero such that feasibility is restored. Both the proxy and incremental FTRs exhaust transmission capacity in the pre-expansion and expanded grid, respectively. The proxy FTRs help allocating incremental FTRs by preserving capacity in the preexpansion network, which results in an allocation of incremental FTRs amounting to the new transmission capacity created in 1–2 direction.30 The proxy awards are transmission congestion hedges that can be auctioned to electricity market players in the expanded network.31 In the example provided by Bushnell and Stoft (1997), the investor with pre-existing FTRs chooses the most profitable incremental FTR based on optimizing its final benefit. The investor is then awarded a mitigating incremental 1–2 FTR with

Tarjei Kristiansen and Juan Rosellón

associated power flows corresponding to the difference between the ex-ante and ex-post optimal dispatches. The pre-existing FTRs correspond to the actual dispatch of the system and become infeasible after expanding line 1–2, and therefore a mitigating 1–2 FTR32 is allocated so that feasibility is exactly restored (that is, the investor “pays back” for the negative externalities to other agents). There is no allocation of proxy awards because the pre-expansion network is fully allocated by FTRs before the expansion. The amount of incremental FTRs is minimized because they represent a negative value to the investor and decrease its revenues from the pre-existing FTRs. 5.

The auction model and welfare

Bushnell and Stoft (1997) demonstrate that the increase in social welfare will be at least as large as the ex-post value of new contracts, when the FTRs initially match dispatch in the aggregate and new FTRs are allocated according to the feasibility rule. In particular, if social welfare is decreased by transmission expansion, the investor will have to take FTRs with a negative value (If social welfare is increased there will be free riding). With only the aggregate match of FTRs and dispatch, some agents might still benefit from investments that reduce social welfare, whenever their own commercial interests improve to an extent that more than offsets the negative value of the new FTRs. Further, Bushnell and Stoft show that incentives for expansion that reduce social welfare would be removed if FTRs for each agent as a perfect hedge and match their individual net loads. In such a case, FTRs allocated under the feasibility rule ensure that no one will benefit from an expansion that reduces welfare. Although apparently similar, our mechanism and its implications on welfare are different from those in the Bushnell and Stoft (1997) model. Bushnell and Stoft analyze the welfare implications of transmission expansion given matching of dispatch both in the aggregate and individually. In our model, we assume unallocated FTRs both before and after the expansion, so that there is no match in dispatch.33 However, the proxy award mechanism developed in this paper implies nonnegative effects on welfare in the sense that future investments in the grid cannot reduce the welfare of aggregate use for FTR holders. The reason is that simultaneous feasi-

63

bility is guaranteed before and after the enhancement project so that revenue adequacy is also guaranteed after expansion. Only those non-hedged agents in the spot market might be exposed to rent transfers. For feasible long term transactions identified ex ante, the FTRs provide perfect congestion hedges for the existing grid or for any future grid that develops under the feasibility rule. However, FTRs cannot provide perfect hedges ex post for all possible transactions. A similar property carries over to any welfare analysis under FTRs. More formally, suppose we have a social welfare function B for dispatch in a single period. Also assume that there is no uncertainty, that all functions are known, and that agents are price takers in the electricity and FTR markets. A simple welfare model associated with transmission expansion ∆ is:34 (32)

where

Then Y* is the dispatch that maximizes social welfare without the expansion. Let ∆+ be the dispatch that would be provided as an increment due to transmission expansion. ∆+ solves program (32). Note that if P + = ∇B(Y* + ∆+), then under reasonable regularity conditions ∆+ is also a solution to: (33)

Formulation (33) is interpreted as the maximization of congestion rents for the incremental allocation ∆. In the context of Bushnell and Stoft assumptions,35 suppose now that the current allocation of FTRs T satisfies T = Y*, then (33) would award the maximum value of incremental FTRs. In this case, we need not know the full benefit function. We could rely on the expander to estimate P+, and provide this preference ranking function as part of the bid. Then solving (33) would give the maximum value incremental award for expansion K+, and this award would preserve the welfare maximizing property of the FTRs for the expanded grid.36 Now suppose that (for some reason) T ↑ Y*. To preserve simultaneous feasibility the constraint

Repsol YPF-Harvard Kennedy School Fellows Research Papers

64

K+(T + ∆) ≤ 0 should be imposed. A natural (second best) rule might be:

(34)

Hence, the existing users of the grid could continue to do as before the expansion, and the expander receives the incremental values arising from the expansion. Then the example in Hogan (2002a, p. 12; see also Annex 3) illustrates the case of a beneficial expansion where the only solution to (34) is ∆ = 0 so that the expansion project does not occur. In fact, K+(T + ∆) ≤ 0 cannot be relaxed without violating the critical property of simultaneous feasibility. We illustrate the argument in the following examples. Consider the example in Hogan (2002a, p. 12) illustrated in Figure 4. Here the dispatch in the preexisting network does not match the current allocation of FTRs. The limiting constraints for the dispatch are the 1–3 and 2–3 constraints. Likewise, the limiting constraints for the current allocation of FTRs are the 1–2 and 1–3 constraints. Assume that the incremental dispatch in the 1–3 and 2–3 directions may be caused by the increased capacity of line 1–3. The relevant constraints are K+(Y* + ∆) ≤ 0 for the current dispatch and K+(T + ∆) ≤ 0 for the current allocation of FTRs. The corresponding objective is Max(P +13 ∆ 13 + P +23 ∆ 23). Then the following constraints would apply for the specific network topology: Figure 4 Dispatch Y does not match the current allocation of FTRs T.

First assume that T13 = 1100, and T23 = 500 and Y13 = 900, Y23 = 900 and that the incremental benefit of expansion is greater in the 1–3 direction than in + the 2–3 direction. Also assume that C13 = 1000. We notice that the mismatch between the dispatch and existing FTRs is Y13 – T13 = 300 and Y23 – T23 = –400. Furthermore, the marginal expansion occurs from the current dispatch to where the 1–3+ and 2–3 transmission constraints intersect. This amounts to the incremental dispatch ∆13 = 200 and ∆23 = –100. If the above numbers are substituted in the constraint we find that the transmission capacity constraint for line 1–2 and existing FTRs are violated because

2 >3 2000

2-1

1-3 1500

1-3+

2-3

. 1-2

1000 Y 500

T 1>3 -

500

1000

1500

2000

Hence the expansion does not occur. Conversely, assume that the location of the current dispatch and existing FTRs are interchanged so the mismatch between the dispatch and existing FTRs is Y13 – T13 = –300 and Y23 – T23 = 400 and assume that the marginal benefit of the expansion is greater in the 2–3 direction than in the 1–3 direction. Then the incremental dispatch would be ∆13 = 0 and ∆23

Tarjei Kristiansen and Juan Rosellón

65

= 300. In this case the 2–3 transmission capacity constraint would be violated for the existing FTRs

welfare effects and regulatory implications would be of value. 6.

. Similar problems would arise with rules such as preserving proxy awards to allow for any possible dispatch on the existing grid, where the only expansions incented would be those that added to every constraint in the system, virtually foreclosing the possibility of investment under this rule. Given the complicated externalities of electric grid, a first best system based on decentralized property rights is not known. Traditionally, all investment decisions relied on central decisions by regulators under certification of need. This often produced regulatory gridlock precisely because of the grid externalities considered here (not to mention the siting and environmental issues). The FTR feasibility rule always preserves the property that the incidence of any welfare reductions falls to those whose transaction were not selected ex ante to be hedged by FTRs. In dealing with the aggregate welfare effects, the second best motivation is shown in (33) (without going all the way to (34)). In the absence of the known welfare function or the possibility of allocating all the existing grid, the total award is divided between proxy awards and incremental awards for the investor. The proportional part of the resulting total award that could be achieved with the existing grid is preserved as a proxy award. The remainder is assigned to the investor. Subject to this rule, the total award is chosen to maximize the market value of the incremental award to the investor. Presumably this would reinforce the incentive for the investor to provide an accurate estimate of the market value. Given the prices, the special case of FTRs matching dispatch or T = Y* (considered by Bushnell and Stoft, 1996, and Bushnell and Stoft, 1997) is consistent with this rule, and the welfare maximizing results apply. In the case where there is not a full allocation of the existing grid, the likely result is that there would be more scope for welfare reducing investments. The need for regulatory oversight would not be eliminated, but the intent is that the scope of the regulatory issues would be reduced. Since proxy award mechanisms are in use and more are under development, further investigation of the private incentives,

Concluding remarks

We proposed a merchant mechanism to expand electricity transmission. Proxy awards (or reserved FTRs) are a fundamental part of this mechanism. We defined them according to the best use of the current network along the same direction of the incremental expansion. The incremental FTR awards are allocated according to the investor preferences, and depend on the initial partial allocation of FTRs and network topology before and after expansion. Our examples showed that the internalization of possible negative externalities caused by potential expansion is possible according to the rule proposed by Hogan (2002a): allocation of FTRs before (proxy FTRs) and after (incremental FTRs) the expansion is in the same direction and according to the feasibility rule. Under these circumstances, the investor will have the proper incentives to invest in transmission expansion in its preference direction given by its bid parameters. Likewise the larger the existing current capacity the greater the number of FTRs that must be reserved in order to deal with potential negative externalities depending on post network topology. Our mechanism of long term FTRs is basically a way to hedge market players from long-run nodal price fluctuations by providing them with the necessary property transmission rights. The main purpose of the four basic criteria that support our model (feasibility rule, proxy awards, maximum value and symmetry) were to define property rights for increased transmission investment according to the preset proxy rule. However, the general implications on welfare, and incentives for gaming are still an open research question. Although our model is specifically designed to deal with loop flows, and the security-constrained version of our model can take care of contingency concerns, our proposed mechanism is to be applied to small line increments in meshed transmission networks. LTFTRs are efficient under non-lumpy marginal expansions of the transmission network, and lack of market power. Regulation has then an important complementary role in fostering large and lumpy projects where investment is large relative to market size, and in

Repsol YPF-Harvard Kennedy School Fellows Research Papers

66

mitigating market power. Since revenues from nodal prices only recover a small part of total costs, LTFTRs must be complemented with a regulated framework that allows the recovery of fixed costs. The challenge is to effectively combine merchant and regulated transmission investments or, as Hogan (2003) puts it, to establish a rule in practice for drawing a line between merchant and regulated investment. 7.

(44)

(45)

(46) (47)

Annexes

Equation (44) gives δ12 = 1. Equation (36) gives γ = b12. Equation (41) gives λ = p12, equation (38) ζ = γ / p12 = b12 / p12 (ε is zero because the constraint is redundant), and equation (39) θ = 0. From this it follows (equation 37) that ϕ = 0. Furthermore equation (42) gives tˆ = C12 – T12. Equation (40) + + implies that a = C12 – T12 – tˆ = C12 – C12.

7.1 Annex 1 7.1.1 Solution to program The Lagrangian of the problem is:

7.1.2 Solution to program (35)

The Lagrangian of the problem is:

where γ, θ, ζ, ε, ϕ, κ, and π are the multipliers associated with the respective constraints. At optimality the Kuhn-Tucker conditions are: (36)

(48)

(37) (48) (38)

(39) (40)

(41)

(42)

(43)

where γ 1 and γ 2 are the Lagrange multipliers associated with transmission capacity on the lines 1–3 and 1–2 in the expanded network, respectively. ζ is the multiplier associated with the Kuhn-Tucker condition of transmission capacity in the preexpansion network for line 1–3. This line has the Lagrange multipliers λ associated with it before expansion. ε is the investor’s marginal value of transmission capacity in the pre-expansion network when allocating incremental FTRs. The normalization condition has the multiplier ϕ and the non-negativity conditions have the associated multipliers κ and π. The first order conditions are:

(49) (49)

Tarjei Kristiansen and Juan Rosellón

67

(50)

(51)

(52)

(53)

,

with

(54) ,

7.2 Annex 2 (55) ,

,

,

,

,

,

(56)

(57)

(58)

This annex derives the power transfer distribution factors (PTDFs) for the three-node network with two parallel lines, and where all lines have identical reactance. The net injection (or net generation) of power at each bus is denoted Pi. We have the following relationship between the net injection, the power flows Pij and phase angles θ i :

where xij is the line inductive reactance in per unit. We can write the power flow equations as:

(59)

(60)

The solution for the first order conditions is given by:

The matrix is called the susceptance matrix. The matrix is singular, but by declaring one of the buses to have a phase angle of zero and eliminating its row and column from the matrix, the reactance matrix can be obtained by inversion. The resulting equation then gives the bus angles as a function of the bus injection:

Repsol YPF-Harvard Kennedy School Fellows Research Papers

68

The PTDF is the fraction of the amount of a transaction from one node to another node that flows over a given line. PTDFij,mn is the fraction of a transaction from node m to node n that flows over a transmission line connecting node i and node j. The equation for the PTDF is:

Figure 6 Feasible expansion FTR set FTR: 2 >3

tδ dδ

2000 Feasible expansion

1500 1000

where xij is the reactance of the transmission line connecting node i and node j and xim is the entry in the ith row and the mth column of the bus reactance matrix. Utilizing the formula for the specific example network gives:

T13 = 900 T23 = 500

500

FTR: 1->3

1000

500

1500

2000

7.3.2 Transmission investment that does change PTDFs 7.3 Annex 3 7.3.1 Transmission investment that does not change PTDFs An example on an investment that does not change the PTDFs of the network is shown in Figure 5 where there is an expansion of line 1–3 from 900 MW to 1000 MW transmission capacity. The associated feasible expansion FTR set is shown in Figure 6. We observe that whatever feasible FTRs that existed before expansion, none of these will become infeasible after the expansion. Figure 5 Three-node network with expansion in one line

1

900 MW MAX to 1000 MW

200 MW MAX

3

900 MW MAX

Figure 7 shows a three-node network where a line is inserted in parallel with an existing line between the nodes 2 and 3. Inserting a parallel line with identical reactance as the existing line halves the total reactance between nodes 2 and 3. As a result the PTDFs of the expanded network change, but not as substantial as in the example in section 4.2. For example PTDF12,13 = 1/3 and PTDF13,13 = 2/3 change to PTDF12,13 = .04 and PTDF13,13 = 0.6. Furthermore, the inserted line has identical transmission capacity to the existing one so that the total transmission capacity is doubled between the buses 2 and 3. However, the simultaneous interaction of the reactances and transmission capacities changes the feasible expansion FTR set as illustrated in Figure 8. Then some of the pre-existing FTRs may become infeasible. Figure 7 Three-node network where a line is inserted in parallel with an existing line

1

900 MW MAX

3

2 200 MW MAX

2

1800 MW MAX

Tarjei Kristiansen and Juan Rosellón

69

Figure 8 Feasible expansion FTR set t^δ aδ

FTR: 2->3 2000 1500

Feasible expansion T + aδ

2000

5.

6. 7.

500

Reduction 2000

1500

2000

FTR: 1->3 500 8.

Endnotes 1. See Joskow and Tirole (2000), and Léautier (2001). 2. No restructured electricity sector in the world has adopted a pure merchant approach towards transmission expansion. Australia has implemented a mixture of regulated and merchant approaches (see Littlechild, 2003). Pope (2002), and Harvey (2002) propose LTFTR auctions for the New York ISO to provide a hedge against congestion costs. Gribik et al (2002) propose an auction method based on the physical characteristics (capacity and admittance) of a transmission network. 3. Generators can exert local power when the transmission network is congested. (See Bushnell, 1999, Bushnell and Stoft, 1997, Joskow and Tirole, 2000, Oren,1997, Joskow and Schmalensee, 1983, Chao and Peck, 1997, Gilbert, Neuhoff, and Newbury, 2002, Cardell, Hitt, and Hogan, 1997, Borenstein, Bushnell, and Stoft, 1998, Wolfram, 1998, and Bushnell and Wolak, 1999). 4. An example is the power outage of August 14, 2003, in the Northeast of the US, which affected six control areas (Ontario, Quebec, Midwest, PJM, New England, and New York) and more than 20 million consumers. A 9-second transmission grid technical and operational problem caused a cascade effect, which shut down 61,000 MW generation capacity. After the event there were several “finger pointings” among system operators of different areas, and transmission providers. The US-Canada System Outage Task Force identified in detail the causes of the outage in its final report of April, 2004. It shows that the main causes of the black out were deficiencies in corporate policies, lack of adherence to industry policies, and inadequate management of reactive

9.

10.

11.

12.

power and voltage by First Energy (a firm that operates a control area in northern Ohio) and the Midwest Independent System Operator (MISO). See US-Canada Power System Outage Task Force (2004). In the US, transmission fixed costs are recovered through a regulated fixed charge, even in those systems that are based on nodal pricing and FTRs. This charge is usually regulated through cost of service. See Hogan (2002a), Hogan (2002b), and Hogan (2003). In fact, market power mitigation may be a major motive for transmission investment. A generator located outside a load pocket might want to access the high price region inside the pocket. Building a new line would mitigate market power if it creates new economic capacity (see Joskow and Tirole, 2000). Examples of projects that do not change PTDFs include proper maintenance and upgrades (e.g. low sag wires), and the capacity expansion of a radial line. Such investments could be rewarded with flowgate rights in the incremental capacity without affecting the existing FTR holders (we assume however that only FTRs are issued). In our three-node example in section 4.2, PTDFs change substantially. In certain cases, the change in PTDFs could not exist (see annex 3, 7.3.1) or be small if, for example, a line is inserted in parallel with an already existing line (see annex 3, 7.3.2). In a large-scale meshed network the change in PTDFs may not be as substantial as in a three-node network. However the auction problem is non-convex and nonlinear, and a global optimum might not be ensured. Only a local optimum might be found through methods such as sequential quadratic programming. Hogan (2002b) shows that the economic dispatch model can be extended to a market equilibrium model where the ISO produces transmission services, power dispatch, and spot-market coordination, while consumers have a concave utility function that depends on net loads, and on the level of consumption of other goods. Function B is typically a measure of welfare, such as the difference between consumer surplus and generation costs (see Hogan, 2002b) When security constraints are taken into account (n–1 criterion) this is a large-scale problem, and it prices anticipated contingencies through the security-constrained economic dispatch. In operations the n–1 criterion can be relaxed on radial paths, however, doing the same in the FTR auction of large-scale meshed networks may result in revenue inadequacy. We do not use the n-1 criterion in our paper. In the PJM (Pennsylvania, New Jersey and Maryland) market design, the locational prices are

Repsol YPF-Harvard Kennedy School Fellows Research Papers

70

13. 14. 15.

16.

17.

18.

19.

20.

21. 22. 23.

24.

defined without respect to losses (DC-load flow model), while in New York the locational prices are calculated based on an AC-network with marginal losses. FTRs could be options with a payoff equal to max ((Pj – Pi)Qij,0). See Hogan (1992). The set of point-to-point obligations can be decomposed into a set of balanced and unbalanced (injection or withdrawal of energy) obligations (see Hogan 2002b). This has been demonstrated for lossless networks by Hogan (1992), extended to quadratic losses by Bushnell and Stoft (1996), and further generalized to smooth nonlinear constraints by Hogan (2000). As shown by Philpott and Pritchard (2004) negative locational prices may cause revenue inadequacy. Moreover, in the general case of an AC or DC formulation to ensure revenue adequacy the transmission constraints must satisfy optimality conditions (in particular, if such constraints are convex they satisfy optimality). See O’Neill et al. (2002), and Philpott and Pritchard (2004). Revenue adequacy is the financial counterpart of the physical concept of availability of transmission capacity (see Hogan, 2002a). Each element in the directional vector represents an FTR between two locations and the directional vector may have many elements representing combinations of FTRs. Proxy awards are then currently unallocated FTRs in the pre-existing network that basically facilitate the allocation of incremental FTRs and help to preserve revenue adequacy by reserving capacity for hedges in the expanded network. Another possibility would be to define every possible use of the current grid as a proxy award. However, this would imply that any investment beyond a radial line would be precluded, and that incremental award of FTRs might require adding capacity to every link on every path of a meshed network. The idea of defining proxy awards along the same direction as incremental awards originates from a proposal developed for the New Zealand electricity market by Transpower. We use “two-norm” to guarantee differentiability. See Shimizu et al. (1997). The model could also be interpreted as having multiple periods. Although we do not explicitly include in our model a discount factor, we assume that it is included in the investor’s preference parameter b. Other examples in the economics literature where an upper level maximization takes the optimality conditions of another problem as constraints are given in Mirrlees (1971), Brito and Oakland (1977),

and Rosellón (2000). 25. According to Shimizu et al (1997), the necessary optimality conditions for this problem are satisfied. The objective function and the constraints are differentiable functions in the region bounded by the constraints. A local optimal solution and Kuhn-Tucker vectors then exist. 26. There are other methods available such as transformation methods (penalty and multiplier), and nontransformation methods (feasible and infeasible). See Shimizu et al. (1997). 27. This method considers a tentative list of constraints that are assumed to be binding. This is a working list, and consists of the indices of binding constraints at the current iteration. Because this list may not be the solution list, the list is modified either by adding another constraint to the list or by removing one from the list. Geometrically, the active set method tends to step around the boundary defined by the inequality constraints. (See Nash and Sofer, 1988). 28. The mathematical derivation of these values is presented in annex 1. 29. The detailed mathematical derivation of solutions to program is presented in annex 1. 30. Note that this result will depend on the network interactions. In some cases the amount of incremental FTRs in the preference direction will differ from the new capacity created on a specific line. However, it will always amount to the new capacity created as defined by the scalar amount of incremental FTRs times the directional vector. 31. Whenever there is an institutional restriction to issue LTFTRs there will be an additional (expected congestion) constraint to the model. A proxy for the shadow price of such a constraint would be reflected by the preferences of the investor that carries out the expansion project (assuming risk neutrality and a price taking behavior). The proxy award model takes the “linear” incremental and proxy FTR trajectories to the after-expansion equilibrium point in the ex-post FTR feasible set to ensure the minimum shadow value of the constraint. 32. The incremental 1-2 FTR can be decomposed into a 1-3 FTR and a 3-2 FTR. 33. Additionally, Bushnell and Stoft explicitly define loads, nodal prices, and generation costs so that the effects on welfare are measured as the change in net generation costs. In contrast, we do not define a net benefit function of the users of the grid in terms of prices, generation costs or income from loads. Alternatively, our model maximizes the investors’ objective function in terms of incremental FTRs. 34. We are grateful to William Hogan for the insights in the formulation of the following model.

Tarjei Kristiansen and Juan Rosellón

35. See Bushnell and Stoft (1997, pp. 100–106). 36. This is however a particular type of welfare maximization since, as opposed to Bushnell and Stoft, costs of expansion are not addressed.

8.

References

Brito, D. L. and W. H. Oakland (1977). Some Properties of the Optimal Income Tax. International Economic Review, 18, 407–423. Borenstein, S., J. Bushnell, and S. Stoft (1998). The Competitive Effects of Transmission Capacity in a Deregulated Electricity Industry. POWER Working Paper PWP-040R. University of California Energy Institute (http://www.ucei.berkely.edu/ucei). Bushnell, J. (1999). Transmission Rights and Market Power. The Electricity Journal, 77–85. Bushnell, J. B. and S. E. Stoft (1996). Electric Grid Investment Under a Contract Network Regime, Journal of Regulatory Economics 10, 61–79. Bushnell, J. B. and S. E. Stoft (1997). Improving Private Incentives for Electric Grid Investment. Resource and Energy Economics 19, 85–108. Bushnell, J. B. and F. Wolak (1999). Regulation and the Leverage of Local Market Power in the California Electricity Market. POWER Working Paper PWP070R. University of California Energy Institute (http://www.ucei.berkely.edu/ucei). Cardell, C., C. Hitt, and W. Hogan (1997). Market Power and Strategic Interaction in Electricity Networks. Resource and Energy Economics, 109–137. The California ISO and London Economics International LLC (2003). A Proposed Methodology for Evaluating the Economic Benefits of Transmission Expansions in a Restructured Wholesale Electricity Market, (http://www.caiso .com/docs/2003/03 /25/2003032514285219307 .pdf) Chao, H.-P. and S. Peck (1997). An Institutional Design for an Electricity Contract Market with Central Dispatch. The Energy Journal 18(1), 85–110. Gilbert, R., K. Neuhoff, and D. Newbery (2002). Mediating Market Power in Electricity Networks. Mimeo. Grande, O. S., and I. Wangensteen (2000). Alternative Models for Congestion Management and Pricing Impact on Network Planning and Physical Operation. CIGRE, Paris, aug/sept. Gribik, P. R., J. S. Graves, D. Shirmohammadi, and G. Kritikson (2002). Long Term Rights for Transmission Expansion. Mimeo. Harvard Electricity Policy Group (2002). Transmission Expansion: Market Based and Regulated Approaches. Rapporteur’s Summaries of HEPG

71

Twenty-Seventh Plenary Sessions, Session Two, January 24–25. Harvey, S. M. (2002). TCC Expansion Awards for Controllable Devices: Initial Discussion. Mimeo. Hogan, W. (1992). Contract Networks for Electric Power Transmission. Journal of Regulatory Economics, 4, 211–242. Hogan, W. (1999). Market-Based Transmission Investments and Competitive Electricity Markets, Mimeo, JFK School of Government, Harvard Electricity Policy Group, Harvard University, http://www.ksg .harvard.edu/people/whogan Hogan, W. (2000). Flowgate rights and wrongs, mimeo, JFK School of Government, Harvard Electricity Policy Group, Harvard University, http://www .ksg.harvard.edu/people/whogan. Hogan, W. (2002a). Financial Transmission Right Incentives: Applications Beyond Hedging. Presentation to HEPG Twenty-Eight Plenary Sessions, May 31, http://www.ksg.harvard.edu /people/whogan. Hogan, W. (2002b). Financial Transmission Right Formulations. Mimeo, JFK School of Government, Harvard Electricity Policy Group, Harvard University, http://www.ksg.harvard.edu/people /whogan. Hogan, W. (2003). Transmission Market Design. Mimeo, JFK School of Government, Harvard Electricity Policy Group, Harvard University, http://www .ksg.harvard.edu/people/whogan. Joskow, P. and J. Tirole (2000). Transmission Rights and Market Power on Electric Power Networks. RAND Journal of Economics, vol. 31, no. 3, Autumn, pp. 450–487. Joskow, P. and J. Tirole (2002). Transmission Investment: Alternative Institutional Frameworks. Mimeo. Joskow, P. and J. Tirole (2003). Merchant Transmission Investment. Mimeo. Joskow, P. and R. Schmalensee (1983). Markets for Power: An analysis of Electric Utility Deregulation, MIT Press. Léautier, T.-O. (2000). Regulation of an Electric Power Transmission Company. The Energy Journal, vol. 21, no. 4, pp. 61–92. Léautier, T.-O. (2001). Transmission Constraints and Imperfect Markets for Power. Journal of Regulatory Economics, 19(1), 27–54. Littlechild, S. (2003). “Transmission Regulation, Merchant Investment, and the Experience of SNI and Murraylink in the Australian National Electricity Market,” mimeo. Mirrlees, J. A. (1971). An Explanation in the Theory of optimum Income Taxation. The Review of Economic Studies, 38, 175–208 Nash, S. G. and A. Sofer (1988). Linear and Nonlinear Programming, John Wiley and Sons, New York.

72

O’Neill, R.P., U. Helman, B.F. Hobbs, W.R. Stewart, and M.H. Rothkopf (2002). A Joint Energy and Transmission Rights Auction: Proposal and Properties, IEEE Trans. Power Systems, 17(4), Nov., 1058–1067. Philpott, A. and G. Pritchard. (2004). Financial transmission rights in convex pool markets, Operations Research Letters, Volume 32, Issue 2, March, 109–113. Pérez-Arriaga, J. I., F. J. Rubio, J. F. Puerta Gutiérrez et al. (1995), Marginal Pricing of Transmission Services: An Analysis of Cost Recovery, IEEE Transactions on Power Systems, vol. 10, no. 1, February. Pope, S. (2002). TCC Awards for Transmission Expansions. Mimeo. Rosellón, J. (2000) The Economics of Rules of Origin, The Journal of International Trade and Economic Development, Vol. 9, No. 4 December 2000. Rubio-Oderiz, J. and I. J. Pérez-Arriaga (2000). Marginal Pricing of Transmission Services: A Comparative Analysis of Network Cost Allocation Methods. IEEE Transactions on Power Systems, 15. Sheffrin, A., and F. A. Wolak (2001). Methodology for Analyzing Transmission Upgrades: Two Alternative Proposals. Mimeo

Repsol YPF-Harvard Kennedy School Fellows Research Papers

Shimizu, K., Y. Ishizuka J. F. Bard (1997). Nondifferentiable and Two-Level Mathematical Programming, Kluwer Academic Publishers, Norwell MA. US-Canada Power System Outage Task Force (2004). “Final Report on the August 14, 2003 Blackout in the United States and Canada: Causes and Recommendations.” (available at: https://reports .energy.gov/BlackoutFinal-Web .pdf) Vogelsang, I. (2001). Price Regulation for Independent Transmission Companies. Journal of Regulatory Economics, vol. 20, no. 2, September. Wolak, F. A. (2000). An Empirical Model of the Impact of Hedge Contract on Bidding Behavior in a Competitive Electricity Market. International Economic Journal, Summer, 1–40. Wolfram, C. (1998). Strategic Bidding in a Multi-Unit Auction: An Empirical Analysis of Bids to Supply Electricity in England and Wales. The RAND Journal of Economics 29: 703–725.

3

Different Approaches to Supply Adequacy in Electricity Markets Juan Rosellón Centro de Investigación y Docencia Económicas (CIDE) and Harvard University

Abstract

so that current reserves generally remain above 16%, which seems acceptable for reliability purposes. Likewise, several of the examples of electricity crises have been in systems that heavily depend upon hydropower. However, there is a growing concern on whether liberalized markets will be able to provide adequate incentives for sufficient investment in generation capacity. This is particularly problematic due to some intrinsic characteristics of electricity markets such as: a) a short-term inelastic demand that implies that the (long-term) supply-demand balance cannot be achieved through a market-clearing price; b) a lack of forward electricity markets beyond one or two years; c) the favorable arena for strategic behavior due to the difficulty to get market clearing prices in tight situations, and d) final consumers do not feel the need to engage in long-term contracts because they are usually isolated from spot prices by regulated tariffs.3 Several measures have been proposed to ensure a sufficient amount of generation capacity reserves. As shown in figure 1, such measures might be analyzed in terms of their degree of centralization or decentralization with regards to the

This paper studies the electricity market design long run problem of ensuring enough generation capacity to meet future demand (resource adequacy). Reform processes worldwide have shown that it is difficult that the market alone provides incentives to attract enough investment in capacity reserves due to market and institutional failures. We study several measures that have been proposed internationally to cope with this problem including strategic reserves, capacity payments, capacity requirements, and call options. The analytical and practical strengths and weaknesses of each approach are discussed. 1.

Introduction

The recent electricity power crises in California, New York, Italy, Norway, Sweden, Brazil, Argentina, Chile and New Zealand have dramatically showed the importance of a reliable electricity supply.1 As of 2000, generation reserves have declined in most markets since liberalization.2 Average reserves have also decreased in most IEA markets except for the UK An extreme case is Australia where there was significant initial overcapacity but reserves drop significantly after the reform. In the cases of UK, Sweden and PJM reserves in 2000 kept similar to those observed at the time of the original reform, but in Norway there was a decrease of 2% from 1991 to 2000, and in California a decrease of 7.5% from 1990 to 1998. The change in reserve margins has occurred in most cases from a starting point of large reserves

Figure 1

amount of capacity DECENTRAL

I am grateful to William Hogan for insightful comments. The research reported in this paper was completed in residence under the Repsol YPF-Harvard Kennedy School Fellows program. Additional support provided by the Fundación México en Harvard, and the Comisión Reguladora de Energia.

CENTRAL

DECENTRAL

CENTRAL

73

price of capacity

Repsol YPF-Harvard Kennedy School Fellows Research Papers

74

amount of capacity and the price of capacity (see Knopff, 2002). In this paper we carry out an analysis of each one of these measures both studying their theoretical fundaments as well as their international application and assessment. 2.

Totally Centralized vs. Totally Decentralized Resource Adequacy

We start analyzing two extreme approaches to resource adequacy and investment in capacity reserves. One extreme is a fully centralized solution where a vertically integrated utility centrally deals with imbalances and manages congestion and ancillary services using its own generation resources. This is the “wheeling” model that is utilized in the United States in areas that have not gone into a competitive structure and that have no spot market (Hunt, 2002). The Mexican model is another example of centralized supply adequacy where private independent power producers sell energy to the state monopsony CFE under longterm power purchase agreements that are supported by government funds.4 Another centralized alternative is the creation of a “moth ball” (or strategic) reserve with government subsidy and centralized decisions regarding both amount and price of capacity (see figure 2). The moth ball reserve would imply a strategic reserve of generation capacity,5 with an operation centrally controlled by the government and that would only be used during emergencies. There is of course a social cost to this procedure since subsidies would be financed through public funds at large. Supply of capacity reserves would then be categorized as a public service obligation (Knopff, 2002). Figure 2

amount of capacity DECENTRAL

CENTRAL moth ball reserve

DECENTRAL

CENTRAL

price of capacity

An opposite extreme approach to resource adequacy is a fully decentralized solution where the market determines the amount and price of capacity resource that will grant resource adequacy. Under such a solution, the different energy markets would be separated and a sequential equilibrium would be reached in the spot market, the forward energy market, the market for capacity reserves, and the forward transmission market through the voluntary participation of agents, and a minimal supervision of an ISO (Wilson, 2002). Different decentralized models have been tried internationally as in Texas, California, Australian Victoria pool, and NETA in the United Kingdom.6 The aim has been in some cases (NETA) to get the system operator out of the spot markets, so that traders manage the spot market as well as manage congestion, and separate arrangements are set up for ancillary services. Typically, the primary income for recovery of capacity costs is the difference between the market clearing price and the generators’ marginal cost (scarcity payments). Hunt (2002) argues that the basic problem of a decentralized model is precisely that it ends up creating private markets not only for spot energy, but also markets for congestion energy, markets for imbalance energy, and markets for ancillary services. She states that all these markets deal with the same energy product, and in an efficient market all these products would end up being traded at the same price.7 In reality, these prices do not converge, and alternatively higher prices, shortages, bureaucracy and new transaction costs are created. This view is endorsed by Joskow (2003) who shows that wholesale market designs that separate energy and individual ancillary service markets have performed poorly and have made electricity markets subject to unilateral behavior that leads to price increases. California did an actual separation of five electricity markets (Hunt, 2002). Some theoretical studies try to find the optimality conditions for such an approach (e.g., Wilson, 2002, and Chao and Wilson, 2002). However elegant in theory, 8 the electricity industry practice has clearly shown the inconvenience of separating the different markets. Borenstein (2002) also agrees that electricity markets do not fulfill the conditions for full competition to work, so that decentralized sequential and efficient equilibrium of the different electricity

Juan Rosellón

75

markets is impossible. Market power and volatility are really inherent to electricity markets since demand is difficult to forecast and inelastic. Likewise, supply faces binding constraints at peak times, and it is inelastic and very costly to store. This implies that short-term prices are extremely volatile so that small changes in demand or supply conditions lead to price bursts, and even smallshare generators can exercise market power. Borenstein then claims that the best way that regulators can handle market power is through longterm forward contracts between power buyers and sellers together with real-time pricing. Forward contracts help to lower the average price paid in both spot and forward markets, while realtime pricing also makes the demand curve flatter.9 Knopff (2002) describes another market-based mechanism for resource adequacy based on subscription of capacity. The desired generation capacity would be decentrally determined (see figure 3). When demand approaches supply every consumer is restricted to the peak capacity contracted in advance from generators. Peak capacity can be sold by each generator in any amount, and the price for this capacity is left to the market. With this solution both the price and the quantity of peak capacity would also be decentrally determined.10 However, at this moment, such a solution is not technically feasible. Figure 3

amount of capacity DECENTRAL

CENTRAL moth ball reserve

completely market based mechanisms price of DECENTRAL capacity

CENTRAL

In the context of an integrated ISO that reaches a centralized equilibrium in all the electricity markets, De Vries and Neuhoff (2003) analyze the “energy-only” market solution. Such a solution relies on the spot market run by the ISO to take care of resource adequacy so that price spikes sig-

nal the need of investment in generation capacity. De Vries and Neuhoff argue that there are insufficient incentives for generators in an energy-only market to invest in capacity whenever there exists economic uncertainty, or fluctuations in demand. Moreover, they show that when generators and consumers are risk averse, the optimal level of investment from the perspective of generators is below the level consumers wish to finance with long-term contracts. The main reason is that market designs do not have the institutions that permit long-term contracts to develop sufficiently, and generators are restricted in the amount of risk that they can transfer to consumers. Likewise, complete reliance on price spikes is not advisable because they are usually not politically acceptable,11 and they can also be manipulated by the generation companies. Even more, electricity markets that rely on short-term energy revenues might lead to shortfalls in capacity over time that might originate investment cycles where investment lags the demand in the market. Regulators worldwide are then very concerned that energy prices are not enough to cover generators’ capacity costs. Most markets have implemented some type of resource adequacy measure. Texas has recently changed to generation adequacy assurances, and FERC’s Standard Market Design (SMD) also recognized the adequate contracted provision of capacity reserves (FERC, 2002).12 California in 2001 also changed its market approach to capacity supply and prompted a proposal for an available capacity requirement (ACAP) to be imposed on load serving entities (LSEs). It is therefore not surprising that several methods have been formally studied in the literature on incentives for investment in reserve capacity such as capacity payments, capacity requirements, and capacity options. The literature on resource adequacy analyzes these mechanisms in the context of an integrated ISO. We next study such mechanisms. 3.

Capacity Payments

Capacity payments provide remuneration to generators for making available their generation capacity (whether they get dispatched or not). The price of capacity is set while the market determines the amount of capacity available. That is, prices are centrally determined while capacity decisions are decentralized (see figure 4). Capacity

Repsol YPF-Harvard Kennedy School Fellows Research Papers

76

payments are collected from consumers through an uplift charge and determine the cost behavior of the firm but leave the amount of reserves uncertain. Oren (2003) explains that capacity payments are rooted in the theory of peak-load pricing so that energy is priced at marginal cost and a capacity payment is used to recover the fixed capacity cost imposed on peak-period energy users. The optimality condition is such that the shadow price of the capacity constraint is equal to the incremental cost of capacity. Figure 4

amount of capacity

capacity payments

DECENTRAL

CENTRAL moth ball reserve

completely market based mechanisms DECENTRAL

price of capacity

CENTRAL

Capacity payments have been used in Argentina,13 Chile, Colombia, Peru, Spain (together with bilateral capacity contracts), and the United Kingdom.14 Two different kinds of capacity payments have been applied in the international practice: fixed payments and fluctuating payments. Fixed per MW payments have been implemented in Spain, where the compensation depends on the availability and the technology of the power plant, and in Argentina, where the Secretaría de Energía set a $10 MWH ($5 for base capacity and $5 for reliability) payment paid during peak demand blocks (6am–11pm during workdays). Fluctuating payments vary with the need for reserve capacity. Although later rescinded under NETA, they were implemented in the early UK (England and Wales) electricity market. The market merit-order pricing rule is modified during periods of high demand when reserve capacity margins are low. In such circumstance, the market price is defined as the weighted average of two factors: the price of the last accepted offer to generate (LAO) and the value of lost load (VOLL). The weight is the LOLP. The formula for the market

price is then market price = LAO * (1 – LOLP) + VOLL * LOLP, where: 0 ″ LOLP ″ 1. The greater (lower) the surplus reserve capacity the smaller (higher) is LOLP. Generators would ideally add capacity when the expected sum of all these payments over all hours of the year is greater than the cost of installing new capacity. This formula also implies a price cap for VOLL when the system is short of power. A main assessment of capacity payments is that they do not favor very much competition because they create artificial rents that might lead to increased market power in generation. In a simple Cournot model, Carreón-Rodríguez and Rosellón (2004) find the conditions under which a fluctuating capacity payment (as the one put in practice in the UK) might lead to worse results in terms of consumer surplus, profits and net social benefits compared to a system where the market price is not artificially increased and excess demand is satisfied in a regulated reserve (or standby) market.15 They show that implementation of a bypass reserve market makes social sense in terms of prices only if there is a large efficiency gap between old and new generation plants. In such a case, the implementation of the capacitypayment solution would only create artificially high rents that could provide incentives for a development of oligopolistic generation markets (the mathematical derivation of these results is presented in annexes 7.2 through 7.3). In a similar effort, Joskow and Tirole (2004) analyze the effects of an uplift charge of an ISO to recover the costs of resources. They do so in the context of a general model that studies the effects on the theorems of welfare economics of market failures as those existing in electricity markets. They find that capacity payments grant inefficient results: •



When the uplift charge is applied both to peak and off-peak periods, large ISO purchases discourage the build up of base load capacity and push down the peak price. For small purchases, off-peak capacity decreases when the uplift is applied in both peak and off peak periods, and the peak capacity decreases when the uplift is only applied during the peak period.

In a model of imperfect information, Oren and Sioshansi (2003) analyze payments for

Juan Rosellón

reserve capacity in a joint day-ahead energy and reserves auction. Reserves are procured through the energy market using energy only bids, and capacity payments are made based on the generator’s opportunity cost. Oren applies the revelation principle to show that generators have an incentive to understate their costs so as to capture higher capacity rents.16 Such theoretical assessments are confirmed in practice by the case of Argentina that substituted its fixed capacity payment mechanism for a hybrid system of payments and contracts because fixed payments were found to distort the merit order dispatch and negatively affected the long-term financial situation of thermal generators. In the UK, the LOLP system was manipulated by large players at the end of the pre-NETA period.17 In several other countries, capacity payments have also led to construction of inefficient peaking units, promote the use of one fuel over others, and eliminated the incentive for availability during crisis of deficit supply. Likewise, Singh (2002) asserts that, as in any price-cap procedure, setting the optimal level of capacity payments is very difficult, and Knopff (2002) points out that a practical problem of fluctuating capacity payments is that variations in such mechanism happen in the short run, whereas the relevant time for investment in capacity reserves is the long term. Additionally, Gülen considers that the LOLP method is not adequate for largely hydro-based systems (as Brazil) as the LOLP would be very small during wet seasons, which would lead to disproportionate low revenues for thermal generators. Hunt (2002) then claims that any capacity adder should be designed to reflect the value of the plant to the system, which is in turn affected by the technology plant composition in such a system. Capacity payments might be combined with price caps to protect consumers (International Energy Agency, 2002) because when capacity is paid separately there is no need that price spikes remunerate reserve capacity. Hobbs, B. F., Iñón, J. and S. E. Stoft (2002) show that the result of such combination could be a reduction in price volatility without affecting average prices and reserves. However, price caps can also have a locational influence on generators that would seek high price-cap areas.

77

4.

Capacity Requirements

Capacity requirements are set as an obligation to maintain a certain amount of reserve capacity. Such an amount is centrally determined through an administratively forecast of demand, and is usually imposed by the ISO (or the regulator) to LSEs. Conversely to capacity payments, the price is decentrally determined by the market once the amount of reserve capacity is set (see figure 5). LSEs must buy enough “capacity tickets” to meet the expected peak load of their customers multiplied by (1 + X), where X is the expected reserve margin that will cover an estimated level of reliability to cope with random outages. The tickets are sold by generators who are usually allowed to export their reserve capacity to other markets. With a capacity requirement, the regulator is able to control the reserve level but the cost remains uncertain (IEA, 2002) Figure 5

amount of capacity

capacity payments

DECENTRAL

CENTRAL moth ball reserve

completely market based mechanisms DECENTRAL

CENTRAL

price of capacity

capacity requirements

Capacity requirements are used in the Pennsylvania, New Jersey, Maryland (PJM), New York and New England markets where an obligation is imposed on LSEs to arrange for Installed Capacity (ICAP). In particular, PJM put into practice a bidbased, day-ahead and month-ahead ICAP markets.18 LSEs are required to buy ICAP in order to be able to serve loads, and they can trade their ICAP with other LSEs. The ICAP requirements can be met by LSEs through self supply, bilateral transactions with suppliers, capability period auctions (several-month strip), monthly auctions, deficiency-spot market auctions, and so forth. Capacity resources can be exported from (or imported to) the PJM area. Generators sell a recall right that

Repsol YPF-Harvard Kennedy School Fellows Research Papers

78

enables PJM to recall energy exports from capacity resources when required. When capacity is recalled, the supplier is paid the market price for energy. The system operator determines demand through the choice of obligations of LSEs, which must own or purchase capacity resources greater than or equal to their expected peak-load plus a reserve margin. If an LSE is short of capacity, it pays a penalty that equals the daily amount of deficiency in capacity times the number of days. When the system itself is short of capacity, the deficiency charge is the double of the capacity deficiency rate (equal in 2003 to USD 174.73 per MW-day).19 Oren (2003) then proposes to view long-term reserves as a price insurance and be treated as a private good but within the framework of a centralized provision of the ISO that imposes mandatory levels of such insurance on LSEs. These mandatory rules would compensate for several obstacles that consumers face when choosing an adequate level of protection, such as technological barriers on metering control, politically barriers to set electricity tariffs efficiently, and so forth. For a market based on operating reserves backed by high prices Stoft (2002) shows that optimal investment in generation capacity depends on the inverse relationship between capacity requirements and the purchase price limit on the system operator: the higher the reserve requirement the lower the optimal price limit.20 Creti and Fabra (2004) make a theoretical analysis of the PJM ICAP market. They build a two-stage game theory model. In the first stage, prior to the realization of demand, generators compete in the capacity market and receive their payments for the capacity amounts they commit. In the second stage, once demand is realized, generators compete in the domestic and foreign markets. When there is excess demand, the regulator recalls the suppliers’ committed capacity resources, which are paid at market prices. Finally, suppliers get their payments for the energy sold. Creti and Fabra analyze this game for the monopoly and the perfect competition cases, and also study the role of the regulator in choosing the capacity requirement as well as in setting a capacity price cap. Creti and Fabra derive several results from their model on: 1.

The opportunity costs of committing capacity resources.

2. 3.

The firm’s optimal behavior in the capacity market. The regulator optimal decisions regarding capacity price caps and the optimal reserve requirement.

In their first result, Creti and Fabra show the trade-off that a generator faces between committing more resources to the capacity market against the foregone revenues from exports (in the case of being recalled). The difference between the foreign and domestic prices then determines the opportunity cost of committing capacity resources.21 The second result shows that two types of equilibria are possible for the firm’s optimal behavior given the value of the capacity price cap, and the reserve requirement set by the regulator. When the price cap is too “low,” the generator’s opportunity costs will not be covered and a capacity deficit would arise (capacity deficit equilibrium). When the price cap is “high” enough capacity resources are able to cover the needed capacity requirement (market clearing equilibrium).22 Finally, Creti and Fabra show that the regulator should always set the capacity requirement equal to peak demand so as to fully avoid the risk of shortage, and to set the capacity price cap equal to the firm’s opportunity costs of providing full capacity commitment. Creti and Fabra’s results show the fragility of the ICAP system, which crucially depends on the capacity price cap and the capacity requirement. The administrative calculation of the latter variable is a subjective one,23 while the optimality of the former variable depends on the market structure of financial transmission rights (FTRs) since the opportunity cost of the generator is given by the price difference between the domestic and foreign markets: if the FTR is subject to market power that will be reflected in the ICAP market. In practice, ICAP mechanisms have failed to provide investment signals when they are most needed. ICAP markets were subject to market manipulation24 that caused price spikes in 2000 in PJM. The pool was deficient some days in June, July and August 2000 since owners of capacity increased their exports for periods when external prices surpassed the PJM market price. In January 2001, there were price spikes of more than $300 MW-day with a deficiency in system capacity. Furthermore, high market concentration in capacity ownership has also been observed.

Juan Rosellón

In New England, Joskow (2003) has showed that the scarcity rents generated are far below from what would be necessary to attract reserve “peaking” capacity to invest (or continue operation) so as to supply the needed operating reserves and energy during scarcity conditions.25 This means that the combination of an ISO spot market with ICAP markets has not been capable to provide enough incentives to attract generating capacity to maintain adequate reliability levels. Similar results have been obtained for the New York ISO (Patton, 2002). The ICAP system is flawed in part because it derives from short-term adequacy concerns rather than long-term, and since it depends on a subjective estimation of a “right” capacity level which depends on generation stocks, fuel prices, load shapes, and elasticity of demand for reserves. Also, since ICAP is combined with the possibility of exportation of capacity, the value of the ICAP depends on the price differences across the adjacent markets. Furthermore, ICAPs have not provided incentives to build new generation facilities and, conversely, have contributed to keep old inefficient plants in place (Harvard Electricity Policy Group, 2003).26 PJM has then been looking to modify its ICAP system by developing a new methodology for peak load obligation, and by changing the monthahead and day-ahead markets to a price-taker auction while retaining mandatory participation in the day-ahead market. Likewise, the ISO New England proposed a new locational installed capacity (LICAP) market since the capacity markets in New England were registering at certain times prices of zero while generation in constrained areas needed to be valued more highly (Davis, 2004).27 The LICAP proposal includes basing prices in demand curves for Maine, Connecticut, metropolitan Boston, and the rest of New England. New prices are to be phased-in through capped increments in a five-year period. These proposals have been widely opposed by LSEs and other consumers since—in their opinion—it will only produce huge transfers from LSEs to generators, without providing long-term incentives to increase new generation (Davis, 2004). FERC’s original SMD (FERC, 2002) also criticized ICAP requirements and proposed instead the use of resource adequacy requirements with targeted curtailments, penalties for undercontracting, and long-term contracting mandatory measures (FERC, 2002). Chandley and Hogan (2002) argue

79

that this is a further flawed policy because there is no objective way to solve the resource-adequacy problem in accordance with SMD without incurring the many difficult issues faced in ICAP design. They think that the best solution would be to allow prices to clear the energy and reserve markets (so that scarcity costs are properly signaled)28 while allowing financial hedging contracts and demandside measures. According to Chandley and Hogan, FERC should not mandate the replacement of ICAP mechanism while totally discouraging a marketclearing alternative for reserve capacity markets. 5.

Call Options

As seen in the previous section, capacity requirements have the problem of artificially setting a capacity requirement and the value of maintaining such a capacity. Call options are proposed as an alternative system that would represent a more real value of capacity (Vázquez et al, 2001), and that bundles generation adequacy with price insurance. The desired capacity is centrally determined, while price is decentrally determined but consumers are hedged against huge price spikes (see figure 6). Typically, the system operator would purchase call options from the generators in a competitive bidding process that would cover the desired capacity.29 The buyer exercises the option if the spot price is greater than the strike price (and receives a premium equal to the difference between the spot price and the strike price).30 The strike price of options is used as a price-cap in case of emergencies, and high penalties are imposed for failure to deliver when the option is called. This assures that the promised capacity is really made available, especially during the peak periods. Figure 6

amount of capacity

capacity payments

DECENTRAL

CENTRAL moth ball reserve

completely market based mechanisms DECENTRAL

call options CENTRAL

capacity requirements

price of capacity

80

The price cap of a call options system works as a protection to consumers, which will assure that prices stay within a socially acceptable range so that the regulatory intervention becomes a form of insurance against price volatility. Compared to the ICAP system, the risk is now changed to the system operator that now bears the uncertainty of whether the options are used or not. Risk is removed from generators that now face a more stable revenue horizon compared to an uncertain and volatile income for peak generation. The expected generators income for prices above the strike price equals the price of the call options, and the generators now receive a fixed payment for the option. Prices and corresponding capacity payments are then derived as market based premia from the market players’ strategies for risk management. Oren (2003) claims that the provision of supply adequacy through LSE’s hedging obligations captures several important features. If the LSE obligations are adjusted (say) monthly to reflect fluctuations in forecasted peak demand, a secondary market for call options should emerge that would permit the trading of call options among LSEs. However, while secondary markets permit the LSEs to adjust their positions each month, price volatility in such markets increases the LSEs risk. Oren proposes then to treat hedging as another ancillary service, allowing LSEs self provision through bilateral contracts with the ISO acting as a provider of last resort. The danger is of course that this may interfere with incentives in the contract market, and be perceived by LSEs as an alternative to prudent risk management. Oren (2003) further alerts that in countries where there is not a well-developed infrastructure of financial markets, LSEs or generators may assume more risk than they might reliably handle.31 In particular, LSEs might not be able to manage risk in a socially optimal way, so that the regulator should need to set a minimum contracting or hedging level on LSEs. Then again, this would lead to non-market arbitrariness. Vázquez et al (2002) analyze a call-option mechanism for the electricity market in Colombia. The regulator requires the system operator to buy a prescribed volume of reliability contracts that allow consumers to get a market compatible price cap in exchange for a fixed capacity remuneration for generators. This entitles consumers to enough available generation capacity. Reliability contracts

Repsol YPF-Harvard Kennedy School Fellows Research Papers

then consist of a combination of a financial call option with a high strike price, and an explicit penalty for generators in case of non-delivery.32 The regulator carries out a yearly auction of option contracts and sets the strike price (at least 25% above the variable cost of the most expensive generator) and the volume of capacity to be auctioned (in terms of the expected peak demand and the available installed capacity). However, generators decide how to divide their total capacity into different blocks (firm, less-firm, new entrants, and least-firm) and how to price each block, so that capacity assigned to each generator is a market result and not the outcome of an administrative process. The Vázquez et al proposal is very sensitive to market power. Therefore, they propose for implementation in the Colombian electricity market that: a) the maximum amount that a generator can bid is limited to its nominal capacity; b) portfolio bidding is not allowed; and c) the winning bids cannot transfer their obligations of physical delivery to other generators. 6.

Concluding remarks

This paper has surveyed the contributions made to the literature on supply adequacy in electricity markets. We studied the different existing approaches and described their analytical properties and implementation characteristics. In assessing the different alternatives, the trend in the literature is to look for some kind of transitory regulatory intervention that grants resource adequacy. However, Hunt (2002) claims that capacity obligations or capacity payments can only be useful if hourly metering, hourly pricing, and demand bidding are “woefully inadequate” and cannot be implemented expeditiously. Otherwise, the energy and the reserve markets should not be separated. The ideal would be an ISO that runs day-ahead markets and spot markets that takes care of imbalances and reaches equilibrium of all electricity markets in an integrated way. Market players would meet their long run expectations for the demand-supply balance in well-developed forward and futures markets. Energy and reserve pricing would take care of supply adequacy. However, in practice electricity markets are usually implemented together with transitory resource-adequacy measures. Capacity payments and requirements alone have been found to be

Juan Rosellón

inadequate both in theory and practice. The most advanced developments in the literature point to the use of some type of hedging instruments such as call options. Oren (2003) even argues that capacity payments or requirements might work efficiently if combined with risk management approaches and hedging instruments that promote demand side participation. Regulatory intervention would then be focused on promoting rules that facilitate liquid markets for energy futures and risk management.

81

Notwithstanding, capacity mechanisms designed to be applied in the context of an electricity market seem to be missing a fundamental central issue. If regulators set the type, level and location of capacity levels and payments there will not be much left for markets to do. All that would be left is competitive procurement, very much like what is done through traditional regulation. So a fundamental dilemma is that electricity markets with generation capacity mechanisms might exist in separation but their combination seem to be condemned to failure.33

Repsol YPF-Harvard Kennedy School Fellows Research Papers

82

7

Annex

7.1 Reserve Margins in IEA Countries

Juan Rosellón

83

Repsol YPF-Harvard Kennedy School Fellows Research Papers

84

increases the price of electricity) and the marginal costs of each firm.

7.2 The Capacity Payment Model Let us first study a simple stylized version a capacity-payment model. Assume that the inverse demand function at a peak period has the form: (1) where P(Q) is the inverse demand function, Q is the amount of electricity generated, a > 0 and b > 0 are positive constants, and k > 0 is a factor added to the price of electricity during peak periods.34 We assume there are only two firms, firm 1 and firm 2. We then have Q = q1 + q2 (where q1 and q2 are the amounts of electricity generated by firm 1 and firm 2, respectively). The cost functions are (2)

7.3 The Regulated Standby Model Let us now formally analyze the regulated standby model in which excess demand is satisfied in a reserve or standby market. Now firm 1 is a monopoly in the pool market, while firm 2 is also a monopoly operating in the reserve market. Firm 2 only takes care of excess demand. Firm 1’s inverse demand function is given by (8) and its cost function is (9) The profit maximization problem of firm 1 is then:

where c1 is the marginal cost of power generation for firm i = 1,2. Suppose that c1 < c2. The profit maximization problem for firm i = 1,2 is then

(10) In this case, the equilibrium quantity and price are

(3)

(11)

The optimal quantities of a Cournot duopoly and the market price that solve problem (3) are

(12) Then, profits are

(4) (13) (5) Given these optimal values, profits for firm i = 1,2 are

Firm 2 only operates to satisfy excess demand at peak periods. This firm faces an inverse demand function of the form: (14)

(6)

and its cost function is (15)

Therefore, the net social benefit, equal to the sum of total profits plus total consumer surplus is Firm 2’s profit maximization problem is

(16) (7) Note that that this expression is mainly determined by the value of k (the term that artificially

In this case, the equilibrium quantity and price are (17)

Juan Rosellón

85

(18) Then, profits are (23) (19) Hence, the net social benefit in the standby model is

while consumer surplus in the standby model is greater than consumer surplus under the capacitypayment model if

(20) Now, given that c1 c2 (since the firms that operate in the pool are typically more efficient than the firms that operate in the reserve market), we get

(21) and

(22)

7.4 Comparison of the Capacity Payment and Standby Models Once we have obtained the equilibrium values for quantities, prices, profits, consumer surplus and net social benefits in both models, it is possible to compare under what conditions one policy is superior to the other. For this purpose we will assume that generators in the capacity-payment and the standby models face the same cost and demand functions, that is

(24) Given that c1 c2, it is evident from these equations that profits, consumer surplus and net social benefits are greater under the standby model than under the English model the greater is the value of (c2 – c1). That is, the standby model provides better social and private outcomes for economies where the marginal cost difference between modern and old plants is large enough. Moreover, both models can also be compared in terms of implied electricity prices. According to (22), the equilibrium reserve-market price in the standby model is greater than the corresponding spot price. However, what is the relation between the former price and the equilibrium price of the capacity-payment model? It can be shown that (25) whenever the difference (c2 – c1) is sufficiently large. That is, implementation of a bypass reserve market makes social sense in terms of prices only if there is a large efficiency gap between old and new generation plants. In such a case, the implementation of the capacity-payment solution would only create an artificially high rent that could provide incentives for a development of oligopoly generation markets. Endnotes

We carry out the comparison both at the firm level and at the social level. Total profits under the standby model are greater than total profits under the capacity-payment model if

1. Reliability in electricity markets is usually understood as the sum of adequacy and security standards. Adequacy (security) is generally associated with the long run (short run). Security describes the ability of the system to deal with contingencies,

Repsol YPF-Harvard Kennedy School Fellows Research Papers

86

2. 3.

4. 5. 6.

7.

8.

9.

10.

while adequacy refers to the ability of the system to meet the aggregate consumer energy requirements at all times. Security includes the so called ancillary services (voltage support, regulation capacity), spinning reserves, black start capability, etc.). See Singh (2002), and Oren (2003). Annex 7.1 presents data on generating reserves for IEA countries See Bouttes (2004), and Vázquez et al (2002). De Vries and Neuhoff (2003) carry out an extensive analysis of the market and institutional failures in the electricity industry that impedes the development of long-term contracts including: lack of generators’ counter-parties to sign long-term contracts, producers’ imperfect information of the demand function, regulatory uncertainty on whether the regulator will impose price caps in periods of price spikes, investment cycles due to long-lead times for new generation facilities, generators’ market power, and so forth. See Carreón-Rodríguez, et al (2003), and Madrigal and de Rosenzweig (2003). In Norway and Sweden there is direct ownership of some peaking plants (Güllen, 2000). In England and Wales the existing integrated system was substituted with an extreme version of a decentralized model that discourages the use of imbalances and trading in markets remote from the system operator. According to Hunt (2002) this implies a reduction in the transparency of energy markets because imbalance prices do not reflect efficient contract prices. This is theoretically confirmed by CarreónRodríguez and Rosellón (2004) which show that prices in the capacity reserves, peak capacity and non-peak capacity markets converge to the same price in a model that separates these three markets. For example, Chao and Wilson (2002) analyze the two-part Californian procurement auction for the market of spinning reserves. One part of the auction was designed for making capacity available, while the other part was for supplying incremental energy. A scoring rule is meant for comparing bids, while a settlement rule for paying accepted bids. The revelation principle applied to this model makes that each suppliers’ optimal energy bid reveals their true marginal cost. Additionally, the ISO and the generators are not required to agre on the probability distribution of dispatched energy Most of the recent electricity reform proposals also promote the use of demand side bidding measures (see for example Commonwealth of Australia, 2002) Carreón-Rodríguez and Rosellón (2004) develop a two-stage oligopolistic model where generators decide first if they should enter to the long-term

11.

12.

13. 14.

15.

16. 17. 18.

19.

20.

21.

22.

reserves market or the spot market. If they go into the spot market, they decide in the second stage to supply either peak or non-peak capacity. Therefore, both amount and price of long-run capacity reserves and peak capacity are set in the market. Also in a theoretical framework, Murphy and Smeers (2002) build a closed-loop Cournot two stage game that describes a situation where investments in capacity reserves are decided in a first stage while sales in the spot market occur in a second stage. Both stages take place in oligopolistic markets. Their framework does not include forward contracting. They find non-convexities in the first stage of the problem (a fact common of bi-level programs) but are able to conclude that a model with a spot market has lower prices and higher quantities than a model without a spot market. Gülen (2002) shows that if the probability of lost load in the PJM market is 1 day in 10 years, price spikes in the range of $12,000–$30,000 per Mwh are needed in an energy-only market. Energy-only markets work however in Australia and New Zealand with maximum prices between $2,500 and $5,000. However FERC has recently backed of and recognized the State’s jurisdiction over resource adequacy measures. Argentina changed to a capacity market in 2000. With the adoption of “NETA” in October 2000, the UK abandoned capacity payments based on the loss of load probability (LOLP) method along with the pool system. A similar approach to a standby market was applied in Victoria, Australia, with obligations to ensure capacity in an energy-only market. See also Newbery (1995). See Green (2004). On October 1, 1998, PJM initiated monthly and multi-monthly capacity markets, while daily capacity markets initiated their operation in 1999. The capacity deficiency rate indicates the annual fixed cost of a combustion turbine in PJM plus transmission costs (PJM, 2003). Stoft (2002) also shows that in a perfectly competitive market a price cap equal to the average value of lost load results in an optimal level of investment in generation capacity. Ford, 1999; Hobbs et al. 2001 also discuss the need for price caps when markets do not clear. More specifically, the opportunity cost is also a function of the probability of recall, the amount of resources needed by the system to assure resource adequacy, and the intensity of price competition in the energy market. Joskow and Tirole (2004) also build a model that shows how a combination of capacity requirements

Juan Rosellón

23.

24.

25.

26.

27.

28.

29.

30.

with capacity price caps might potentially restore investment incentives. Even in the presence of market power, a (Ramsey) optimum can be achieved when: (i) LSE capacity requirements can be met both by peak and base load generators, (ii) capacity requirements are determined using the demand from all consumers, and the capacity prices reflect the prices paid by all retail consumers, and (iii) the market for peaking capacity is contestable. However, this result is not true when there are more than there states of nature (where two state of nature are “off-peak” and “peak”). In such a case strict pricecap regulation might be used to alleviate market power off-peak and allow peakers to recover their investment (Joskow and Tirole, 2004, pp. 45-46). There have been efforts to improve the calculation of the capacity requirement. For example, in the New York ISO a demand curve is proposed to be constructed as an alternative to an ICAP market that intends to increase resource reliability by valuing additional ICAP above the fixed capacity requirement (Harvard Electricity Policy Group, 2003). ICAP gives incentives in the short run for manipulating the availability of plants to increase revenue. Anticompetitive behavior is potentially higher when capacity and system constraints are binding. Another practical problem of ICAP is the interaction among systems with and without capacity requirements, which might lead to inefficient distortions. (IEA, 2002). The average scarcity rents in New England of $10,000 Mw-Year are very low compared to the fixed cost of a new combustion turbine built to provide reserve capacity estimated in between $60,000– $80,000 Mw-year (Joskow, 2003). Joskow and Tirole (2004) theoretically show that the inefficient dispatch of resources procured by the ISO in order to be used during reserve scarcity conditions will lead in the long run to substitution of base load units by peak units. Creti and Fabra (2004) deduce from their theoretical model the possibility that capacity markets clear at zero prices if there is no spread between national and foreign prices. This is of course confronted with the political motivation to keep prices low. However, from a strictly economic point of view, the experience in industries different from the electricity industry is that “the best cure for high prices is high prices” (Harvard Electricity Policy Group, 2003, p.18). Alternatively, LSEs could be the buyers of options through self-provision from their own controlled resources or bilateral contracts with generators. The buyers of the call option may choose the strike price that suits their risk aversion: high (low) strike

87

31.

32.

33. 34.

prices have small (high) premiums. Option premiums also work as substitute efficient signals compared to price signals generated by ICAPs (Singh, 2002). Likewise, the capital market might not be able to provide the long term financing for generation investments commensurate to the associated risk. This combined with inexperience with commodity trading in the electricity industry and the perceived regulatory risk, might raise the cost of capital so much that the investment level will be far below than the needed for an efficient resource adequacy level (Oren, 2003). When the market price p is greater than the strike price s, and the generator is unable to honor its obligation to produce, the generator will have to pay an additional penalty pen (apart from the difference p – s). The additional penalty is intended to discourage even more bids not backed by reliable capacity. I owe this observation to William Hogan. k would therefore contain terms such as “cfalla” and “k factor” of the 1999 Mexican reform proposal (see Carreón-Rodrígiuez and Rosellón, 2002).

References Borenstein, S. (2002) “The Trouble With Electricity Markets: Understanding California’s Restructuring Disaster,” Journal of Economic Perspectives, 16: 191–211. Bouttes, J.P. (2004), Roundtable “Market Design and Competition in Electricity,” presented at IDEICEPR conference “Competition and Coordination in the Electricity Industry,”, Jan. 16, 17. Carreón-Rodríguez, V.G. and J. Rosellón (2002), “La Reforma del Sector Eléctrico Mexicano: Recomendaciones de Política Pública”, Gestión y Política Pública, Vol. XI, No. 2. Carreón-Rodríguez, V.G., and J. Rosellón (2004), “Incentives for the Expansion of Electricity Supply and Capacity Reserves in the Mexican Electricity Sector,” Mimeo. Carreón-Rodríguez, V.G., Jiménez San Vicente, Armando, and J. Rosellón (2003), “The Mexican Electricity Sector: Economic, Legal and Political Issues,” Working Paper at the Program on Energy and Sustainable Development of the Center for Environmental Science and Policy of Stanford University (available at: http://iis-db.stanford.edu/ viewpub.lhtml?pid=20311&cntr=cesp). Chandley, J. D., and W. W. Hogan (2002), “Initial Comments of John. Chandley and William Hogan on the Standard Market Design NOPR,” November 11, available at: http://ksghome.harvard.edu/~. whogan.cbg.Ksg. Chao, H.P. and R. Wilson (2002), “Multi-Dimensional

88

Procurement Auctions for Power Reserves: Robust Incentive-Compatible Scoring and Settlement Rules,” Journal of Regulatory Economics, 22:2, 161–183. Commonwealth of Australia (2002), “Towards a Truly National and Efficient Energy Market,” available at http://www.ksg.harvard.edu/hepg/Papers/Australian.govt_National.energy.report_02.pdf Creti, A. and N. Fabra (2004), “Capacity Markets for Electricity,” Working Paper, UCEI, CSEM WP-124. Davis, T. (2004), “New England Officials, Utilities Yowl Over Installed Capacity Plan,” The Energy Daily, Wednesday, March 24. De Vries, L. and K. Neuhoff (2003), “Insufficient Investment in Generating Capacity in Energy-Only Electricity Markets,” paper presented at 2nd Workshop on Applied Infrastructure Research (Regulation and Investment in Infrastructure Provision-Theory and Policy), TU Berlin WIP, DIW Berlin, October 11. FERC (2002) Notice of Proposed Rulemaking. Remedying Undue Discrimination through Open Access Transmission Service and Standard Market Design, Docket No. RM01–12–000, July 31. (FERC: Federal Energy Regulatory Commission) Ford, A. (1999), “Cycles in Competitive Electricity Markets: A simulation Study of the Western United States,” Energy Policy (27): 637–658. Green, R. (2004), “Did English Generators Play Cournot?,”, Mimeo, University of Hull Business School. Gülen, G. (2002) “Capacity Payments,” presentation, 22nd Annual IAEE/USAEE North American Conference, Vancouver 6–8 October. Harvard Electricity Policy Group, 2003, “Rapporteur’s Summaries of HEPG Thirty-First Plenary Session,” May 21–22. Hobbs, B. F., Iñón, J. and S. E. Stoft (2002), “Installed Capacity Requirements and Price Caps: Oil on the Water or Fuel on the Fire,” The Electricity Journal, July, pp. 23–34. Hunt, S. (2002), Making Competition Work in Electricity, New York, John Wiley & Sons. IEA (2002), Security of Supply in Electricity Markets: Evidence and Policy Issues, OECD/IEA. (IEA: International Energy Agency). Joskow, P. (2003), “The difficult transition to competitive

Repsol YPF-Harvard Kennedy School Fellows Research Papers

electricity markets in the U.S.,” Working Paper, The Cambridge-MIT Institute Electricity Project, CMI 28, Joskow, P. and J. Tirole (2004), “Reliability and Competitive Electricity Markets,” paper presented at IDEICEPR conference “Competition and Coordination in the Electricity Industry,”, Jan. 16, 17. Knops, H. (2002), “Electricity Supply: Secure under Competition Law?,” proceedings, 22nd Annual IAEE/USAEE North American Conference, Vancouver 6–8 October. Madrigal, M, and F. de Rosenzweig (2003), “Present and Future Approaches to Ensure Supply Adequacy in the Mexican Electricity Industry,” paper presented at the Power Engineering Society Meeting, IEEE, vol. 1, pages 540–543, 13–17 July, Toronto,. Murphy, F. H., and Y. Smeers (2002), “Generation Capacity Expansion in Imperfectly Competitive Restructured Electricity Markets,” CORE Discussion Paper, 69. Newbery, D.M., (1995), “Power Markets and Market Power”, The Energy Journal, Vol. 16, No. 3, pp. 39–66. Oren, S. (2003), “Ensuring Generation Adequacy in Competitive Electricity Markets,” Energy Policy and Economics Working Paper, UCEI, EPE 007. Oren, S. and R. Sioshansi (2003), “Joint Energy and Reserves Auction with Opportunity Cost Payment for Reserves,” paper presented at IDEI-CEPR conference “Competition and Coordination in the Electricity Industry,”, Jan. 16, 17. Patton, D. S. (2002) “Review of New York Electricity Markets,” New York Independent System Operator, October 15, Summer. PJM, Monitoring Market unit (2003), “State of the Market Annual Report,” http://www.pjm.com (PJM: Pennsylvania, Maryland, and New Jersey ISO) Singh, H. (2002), “Alternatives for Capacity Payments: Assuring Supply Adequacy in Electricity Markets,” presentation, 22nd Annual IAEE/USAEE North American Conference, Vancouver 6–8 October. Vázquez, C., M. Rivier, and I. Pérez Arriaga (2002), “A Market Approach to Long-Term Security of Supply,” IEEE Transactions on Power Systems, Vol. 17, pp. 349–357, May 2002. Wilson, R. (2002) “Architecture of Power Markets,” Econometrica, 70: 1299–1340.

4

The Mexican Electricity Sector: Economic, Legal and Political Issues Victor G. Carreón Centro de Investigación y Docencia Económicas (CIDE) Armando Jimenez San Vicente Stanford University Juan Rosellón CIDE and Harvard University

1.

Introduction

and textile industrial areas as well as the largest cities—while leaving aside most rural areas. The Mexican Revolution period (1910–1917), and the political consolidation of the country (which included the assassination of President Álvaro Obregón) caused foreign private investment to trickle. By the late 1920s, two things were clear. First, electricity supply was (and still is) strongly associated with the concepts of “nationalism” and “sovereignty.”2 Second, private investment in the sector was declining and electricity demand was rising. Therefore, there was an urge for the government to step in and assume control of the power system. During the 1930s the industry was swept up in a broader process of reorganization as the Partido Revolucionario Institucional (PRI) consolidated its grip on power and unified the farflung Mexican states into an integrated federal country. As a result of this consolidation, Mexico had the Código Nacional Eléctrico (National Electric Code), and a newly created state-owned and state-financed enterprise—Comisión Federal de Electricidad (CFE)—which came to dominate all investment in new capacity. At the same time, worker unions were developed. Electricity being a key sector for the Mexican government (and mainly to the party in power), the Electricity Worker Union quickly gained political power. Since then, the strong correlation between the evolution of the electric sector and the political environment has become stronger. Through the 1940s and 1950s installed power-generating capacity continued to rise as the government and a few private generators invested heavily in the sector. In 1960, a constitutional amendment to Article 27 nationalized the electricity industry, formally giving the government “exclusive responsibility” for generating, transmitting, transforming, and distributing electricity. The private participation in generation ended and new challenges emerged.

This chapter aims to explain the motivations and strategies for reform in the Mexican electricity sector. Our focus is on the effects of politically organized interests, such as unions and parties, on the process of reform. We show how particular forms of institutions—notably, the state-owned enterprises (SOEs) within the power sector as well as the state firm that supplies most fuels for electricity generation—shape the possibilities and pace of reform. The tight integration of these SOEs with the political elite, opaque systems for cost accounting, and various schemes for siphoning state resources explain why these institutions have survived and the actual progress of reform has been so slow. Where private investors have been allowed into the market it has been only at the margin through the “independent power producer (IPP)” scheme, an oxymoron since the purchase agreements and dispatch rules that determine payment to these IPPs are dominated by the State. In its origins in the late 19th century, the Mexican power system grew as a series of privately owned, vertically integrated regional monopolies. Investors, mainly from firms based in foreign countries, built power systems in areas where they thought they could earn a profit—mainly mining

This paper was elaborated with the participation of Antonio Cervantes, and Luis Carlos Ugalde. Also, we want to thank Guillermo Govela and Gustavo Dector for their contributions. We are especially grateful to David G. Victor for his editing advice in the revision process. Finally we want to thank Mario Chocoteco for his research assistance. Juan Rosellón’s work was completed in residence under the Repsol YPF-Harvard Kennedy School Fellows Program. 89

90

Political issues, lack of credible data on the true cost of electricity, among other difficulties, raised barriers for setting economically efficient tariffs. Also in this decade, the government created the Compañía de Luz y Fuerza del Centro (LFC) to supply electricity to Mexico City and the neighboring states. Reinforced by these changes in the power sector, populist ideas claiming sovereignty and state autonomy as the government’s primary goals became more important than efficiency and economic growth. As was the case in many countries during the 1960s and 1970s, Mexico alienated private investment and insulated the power system from market forces, allowing it to grow without much consideration for the economics of the business. Moreover, the “soft budget” of state financing allowed these enterprises, CFE and LFC, to operate (albeit inefficiently) and to wield growing political power Nonetheless, a steady supply of new technologies (developed mainly abroad) as well as the economies of scale in building everlarger power systems made it possible to sustain low tariffs for end users without causing these firms to become a huge drain on the state budget. Although these improvements were largely exhausted by the 1970s, the surges in oil prices at that time delivered a windfall to oil-rich Mexico, much of which was directed to subsidies for electricity generation. On the other hand, when oil prices crashed in the early 1980s, a deep financial problem created both the urgent need and a political opportunity for reforms that would make the power sector more efficient while reducing the burden on the state to supply all new capacity. Even though those reforms started slowly and cautiously, successive financing crises have created additional pressure for reform. In the late 1980s and early 1990s, the Mexican government implemented swift market reforms in various economic sectors (like banking and pension systems) and started to open its markets to international free trade. These included foreign investment agreements allowing participation in several sectors (including electricity) and the creation of new economic institutions that were required to implement those reforms. The Comisión Federal de Competencia (Antitrust Federal Commission, CFC), the Comisión Federal de Telecomunicaciones (Telecommunications Federal Commission, COFETEL) and the Comisión Reguladora de Energía (Energy Regulatory

Repsol YPF-Harvard Kennedy School Fellows Research Papers

Commission, CRE) were created to regulate markets in order to get the desired social outcomes. More specifically, the CRE was created in 1993 to help build an electricity market. During the late 1990s, former President Zedillo attempted a comprehensive reform of the electricity sector, which included amending the Mexican Constitution, but faced strong political resistance. Finally, in year 2000, for the first time in modern Mexico’s history, a candidate from the opposition—the Partido Accion Nacional, PAN—won the Presidential Office election. The new government made a new reform attempt; but in a divided Congress its proposal did not achieve the required majority support. At the same time, both major parties in the opposition, PRI and Partido de la Revolución Democrática (PRD), presented their own proposals, which are presently being debated. The plan of this chapter is as follows. In section II we recount the history of the electric industry in Mexico to explain the structure of the SOEs that dominated during most of the 20th century and are now the subject of reforms. We analyze the performance of the system, to the best extent possible given the limited data, by looking at patterns of investment and tariffs. We also examine the spread of electrification to the rural poor, regulation of the environmental impact of electricity generation and other social dimensions of the power system. In Section III we examine motivations and outcomes from the various attempts to reform this state-dominated system, starting with the financial crisis in the early 1980s. We analyze the changes introduced in 1992 and the reform proposed by former President Zedillo in 1999. The main political actors: consumers, parties, government, unions, etc, are also introduced as they get involved in the discussion. This is very important since the Mexican electric system (as any other system in the world) should not be seen separately from the political and economical standpoint since both have shaped the power sector. By understanding the different scenarios and conditions that prevailed in the sector, the whole story should make sense. While there has been some progress in the process of reform, fundamental issues remain unsettled because of the combination of economical, political, and legal factors: the composition of both chambers (deputies and senators), the judicial decisions about the legality of the present regulatory schemes, the role of the public

Victor G. Carreón, Armando Jimenez San Vicente, and Juan Rosellón

opinion, especially on issues of nationalism and sovereignty, the new role of the CRE, the evolution of tariffs in the near future, etc. We summarize those in section IV, where we discuss the evolving agenda. Finally, conclusions are stated in section V. 2. 2.1

History of the Mexican Power Sector 1880–1979: Mexico’s political consolidation and power sector growth

The origins of the Mexican Power System can be traced back to the late 19th century when private investors built and operated electric networks that would provide traction, lighting and machine motors for industry (mainly textile and mining) and lighting in the major cities. The first plants deployed whatever source of primary energy was readily available—coal for thermoelectric plants and, where appropriate rivers were available, the power of running water. The first thermoelectric generation plant started operation in 1879, mainly to supply a textile mill at León in the state of Guanajuato; the first hydro plant produced electricity a decade later—for the mining industry at Batopilas in Chihuahua. In parallel, governments sold lucrative concessions for electrification of cities—the first of these, in 1881, awarded electric service for Mexico City to the privately held Compañía Mexicana de Gas y Luz Eléctrica.4 Through these vertically integrated monopolies, installed capacity grew at nearly 20% per year by the first decade of the 20th century.6 Private investors were drawn only to the wealthiest and most industrialized areas. However, investment concentrated in the center of the country around Mexico City. Low prices and generous terms for the concessions, along with the demographic growth of Mexico in the early 1900s, attracted investors—most from firms based in Canada, France, Germany, and the United States, with only a small share from Mexican investors. This private model of electrification was followed in all five of the countries examined in this book. It included few requirements to invest in activities that the private investors themselves would not find profitable, such as “universal access” to electricity or rural electrification. Moreover, this administrative law instrument (i.e., a concession) was laden with ambiguities that, usually, were interpreted in ways that benefited the investors, and there was no

91

authority with clearly articulated competence for setting policy and enforcing the terms of the concessions.8 Mindful of the increasing social, political and economic relevance of electricity, the government tried to tame the monopolistic tendencies of the electricity companies. But the task was daunting as the central government was weak and the industry itself was in the midst of a massive reorganization that would produce even larger monopolies that, by design, would not compete or even complement each other. The most important firms became holding companies by absorbing the many small retail companies—interestingly, this consolidation occurred at roughly the same time that Samuel Insull, in the United States, was creating a vast holding company by acquiring the assets of smaller isolated firms. The Mexican Light and Power Company adopted a single 50 Hz system across its entire network, but Impulsora de Empresas Eléctricas operated at eight local frequencies—from 25 to 58 Hz—and did not attempt to seize the economic advantage of full interconnection. These two firms, together with a much smaller one, Nueva Compañía Hidroeléctrica Chapala, dominated the market. In a few jurisdictions, local government regulators had discovered their ability to wield influence and, in most cases, demanded low tariffs set without regard to costs. Wary of such pricing schedules that, in effect, expropriated monopoly profits, private firms reduced their investments in new capacity. The result, arguably, was the worst of two worlds. Monopolistic pricing flourished where regulators were weak, but in the heart of the industrializing nation—Mexico City—arbitrary tariff rules set the stage for perpetual under-investment in the power sector. In 1926 the federal government adopted a policy strategy that would cast a long shadow over the century. The Código Nacional Eléctrico changed the Constitution and declared electricity a public service and conferred to Congress the attributions to legislate in related matters. In the short term, this constitutional move had little impact because the federal government remained weak—financing and regulation of local electric monopolies, for example, was controlled by state and city governments and the large industrial users of electricity. Congress adopted rules that demanded rural electrification, a politically popu-

92

lar mandate, but the private companies ignored new mandates that made no sense for their business plans. However, the Código did require homogenization of the frequency standards over the complete system—on that score it had large effect since it supplied the public good of coordination and required only two of the dominating firms (Impulsora and Chapala) to align their practices with the third. As it is the case in other countries examined in this book—such as Brazil and South Africa—the Mexican government tried to circumvent the difficulties of sustaining private investment by assuming the function of supplying electricity itself. The seeds of this effort are found in the creation, in 1934, of the Comisión Federal de Electricidad (CFE), with a modest initial budget (50,000 pesos, about USD $14,000 at the time), a tiny staff of 20 employees and two main objectives: 1) to operate as a regulatory agency and liaison between foreign companies and government, and 2) to supply electric service to those areas considered by private power companies as not profitable. With its loose mandate and tiny budget it was hardly clear, at the time, that CFE would emerge as the dominant force in the entire Mexican electric power system. At the same time, President Lázaro Cárdenas consolidated power around PRI, his Party. Although PRI had a strong influence on peasants, its key strength rested on Unions, and the best organized ones were those in the largest industries—mining and electricity. Indicative of the growing power of these unions was the Sindicato Mexicano de Electricistas (SME) (Mexican Electric Workers’ Union), which struck Impulsora de Empresas Eléctricas and its seven subsidiaries in 1936. The oldest and strongest union in México, SME, became a critical piece in President Cárdenas’ policy that became known as “Mexican Corporatism”—strong central government ruling in collaboration with organized labor unions. By the late 1930s the PRI was firmly in control. Land reform and nationalization of economic resources became symbols of Mexican national sovereignty and thus key planks in PRI’s policy platform. The 1938 Electricity Public Service Act, issued just as PRI was completing its consolidation of political power, required strong federal regulation of electric services, including tariffs. Foreign firms, already finding their investments squeezed by low mandated tariffs in a few

Repsol YPF-Harvard Kennedy School Fellows Research Papers

key jurisdictions, reduced their investments still further. In most cases, they maintained their existing capacity but invested little in expansion (for these firms, the Second World War was an additional discouragement to investment abroad.) From 1937 to 1943 private investment grew less than 1%. Wartime President Manuel Avila Camacho sought to nationalize the power system but feared a backlash if he simply appropriated the assets of powerful foreign investors. Rather, he launched a rolling process of nationalization— CFE was instructed to buy (at depressed prices) existing electric assets, and with state resources CFE also oversaw the construction of new generation, transmission and distribution services. Prior to the 1940s, private firms supplied all investment in new capacity (Mexican, Impulsora and Chapala). But then private investment flagged. Nationalization began in 1944 when CFE acquired Chapala (the third largest of the private electric companies) and built CFE’s first generating facility (Ixtapantongo). During the 1940s and 50s, CFE acquired and consolidated hundreds of regional electricity monopolies into a single firm— linking all with common technical standards and taking advantage of the ever-larger economies of scale offered by new generation equipment. From 1939 to 1950, ,52% of the total investment in the power system came from public resources and 30% from contracted credits by the government— essentially all of this within the growing CFE system. Only 18% was private investment from firms that remained outside CFE’s network.10 By 1959, total installed capacity had reached 8,547 MW of which CFE controlled about half (4,229 MW) and the remaining private networks accounted for smaller shares: the Mexican Light and Power Company with 1,821 MW and Impulsora de Empresas Eléctricas with 701 MW. See Figure 1 for the increase in new capacity during this process of nationalization and consolidation of the system— notice that all new capacity came from additions from CFE while the other firms kept the same installed capacity. The consolidation largely finished in 1960 when the Federal Government bought 95% of the common shares in Impulsora and also acquired a majority stake in Mexican. These new acquisitions also allowed for a reorganization of the sector. CFE was given control over all segments and regions of the power system except for the central states of

Victor G. Carreón, Armando Jimenez San Vicente, and Juan Rosellón

Figure 1 Percentage of New Capacity by Firm, 1930 to 1960 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0%

1930 Mexican

1940 Impulsora

1950

1960

Chapala

CFE

Source: CFE and Nafinsa

Mexico, Morelos, Puebla, Hidalgo, and Distrito Federal, which became the service area of a new state-owned enterprise, Compañía de Luz y Fuerza del Centro (LFC). This division of geographical responsibility between two state enterprises remains to the present. By the time nationalization reached the nation’s capital a politically well-organized and (at the time) efficient power company already existed; fearful of being rolled into CFE, instead the incumbents in the center carved LFC out of the remnants of Mexican and implored the Mexican President to establish their service as a separate enterprise. Having largely completed its nationalization already, in 1960 the government formalized the arrangement with a constitutional amendment (Article 27, paragraph 6) declaring: “It is the exclusive responsibility of the Nation to generate, transmit, transform, distribute and supply electricity that is intended for public service use.”12 As in most Latin American countries in the 1960s and 1970s, nationalization along with an import substitution strategy were part of the government’s effort to control economic development—to accelerate the rate of growth and to spread the benefits widely. Over the decades that followed, the government controlled power system connected millions to the grid, achieving nearly universal coverage, which is one of the reasons why the population at large—especially those from adverse social backgrounds—support state control of utilities in Mexico. Along the way, the notion of social justice was expanded to include a wide array of subsidies

93

for urban and agricultural consumers—as in many countries (e.g., India) electricity tariffs were constructed with an eye to political benefits rather than economic cost, and over time they led to a system characterized by mounting losses. Under this new legal framework, CFE continued its growth by acquiring the few regional companies that remained in private hands, buying the last, Compañía de Servicios Públicos de Nogales, in 1972. New functions accreted to the national electricity system built around these two stateowned enterprises. In 1974 President Luis Echeverría Álvarez sponsored yet another amendment to Article 27—this time to grant the State the exclusive right to use radioactive materials and nuclear fuel for generation of energy—as well as Article 73 that reserved for Congress the right to legislate on nuclear energy matters. The process of nationalization and consolidation of control into the hands of the state was finalized legally in 1975 with the Ley del Servicio Público de Energía Eléctrica (LSPEE) which declared CFE and LFC as public suppliers of electricity. State-controlled monopoly, it was thought, was essential for ensuring the real-time management of electric power. Only a state enterprise could be trusted with a technology that had large economies of scale— and thus natural tendencies to monopoly. Furthermore, private generators sought only profitable markets, leaving a large part of the population without electricity, and it was assumed that only a state-owned enterprise could deliver electric service more equitably As in most of the SOEs studied in this book— possibly with the exception of Eskom in South Africa—CFE and LFC were managed like government offices rather than private, competitive firms. Relying heavily on the public budget for financing, they were (and still are) a source of political patronage for senior appointments. These enterprises also host among the strongest of the nation’s unions—key elements of the PRI power base. In addition to these generic features of stateowned enterprises, management of these firms has been complicated by frequent changes in policy as well as by the difficulty of drawing a line between State and enterprise. We now turn to the task of evaluating the system that emerged from this historical context, with particular attention to factors that have affected the choice of fuels, the setting of tariffs and the financial performance of

Repsol YPF-Harvard Kennedy School Fellows Research Papers

94

the State-dominated system. We also examine the system’s performance on several politically important dimensions (beyond financial)— notably, its ability to connect people to the grid. The analysis of successes and failures sets the context for the goals that reform efforts, begun in the wake of a financial crisis in the early 1980s, sought to achieve. This system, controlled and financed completely by the State, appears to have performed adequately during the 1970s. Demand grew rapidly, but so did installed capacity. There was no boom and bust cycle of investment that is often evident in private-investor dominated merchant power systems. However, as in many State-dominated systems, over-building appears to have been commonplace—reserve margins were greater than 30% throughout the period from 1970 to 2002, as shown in Figure 2. As the power system expanded over the 20th century the need for primary fuel quickly outstripped the availability of high quality coal, so

generators turned to other locally available options: oil and hydroelectricity (see Table 1). Rivers were tapped from the earliest decades of electrification, but Mexico’s water resources in the north are scarce and the load factors on hydroelectric plant factors rarely exceed 30%. That left fuel oil from petroleum as the main fuel, particularly as Mexico became one of the top ten world oil producers in the 1970s and growing concerns about environmental pollution favored oil over coal. Since the 1980s, with the availability of cost-effective gas turbines, fuel oil fell out of favor as more costly than gas alternatives—much of the history of reform in the power sector, to which we turn to in the next section, is intertwined with efforts to secure a larger share for gas in the power sector. Within the political and organizational calculus of CFE and LFC, the preference for fuel oil is easy to understand. First, even though Mexico is rich in natural gas, the state-owned Petróleos Mexicanos (PEMEX) that is responsible for hydrocarbon production did not consider gas as part of

Figure 2 Demand and Generation Capacity Growth 45000.00 40000.00

Peak Demand

Installed Capacity

35000.00 30000.00

MW

25000.00 20000.00 15000.00 10000.00

0.00

1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002

5000.00

Source: Secretaria de Energia and Comision Federal de Electricidad

Victor G. Carreón, Armando Jimenez San Vicente, and Juan Rosellón

95

Table 1 Total Installed Capacity by Type of Generation

1879 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000

Hydro

Steam

14 99 192

1.8 4 0 0

355 559 1249 3228 5992 7805 9619

40 171 839 2353 6616 11367 14282

Combined Cycle

540 1687 2914

Turbo Gas

1190 1779 2360

Internal Combustion

72 174 205 271 137 86 116

Geothermal

Dual

Nuclear Wind

12 12 12 37

3 150 700 855

Coal

2100

1200 2600

675 1365

2

TOTAL 1.8 18 99 192 475 680 1234 3021 7414 16862 26267 36213

Source: SENER-CFE. this refers to generating plants in operation.

its core business throughout most of this period and thus could not guarantee a gas supply to the power sector. Even today, gas is a poor second cousin to oil extraction at PEMEX. Second, Mexico does not have coal with the quality needed to generate electricity. Third, the logic of “dependency” and the strategy of import substitution animated a self-sufficiency policy under which imports of technology and fuels would be minimized. Coal and gas plants typically require greater purchases of equipment overseas, whereas oil-fired facilities would be relatively easy to construct and supply with fuel from an oil-rich nation. Crucially for CFE and LFC in their internal decision-making, during the 1970s and 1980s PEMEX sold fuel oil to the power sector at around 30% of its opportunity cost (see Figure 3). From the perspective of managers within CFE and LFC, allocation of investment towards oil was actually efficient. Viewed from the vantage of the country as a whole, this strategy was extremely costly—the under-pricing of fuel oil amounted to a massive implicit subsidy to the power sector that averaged about USD $1.5 billion dollars a year.14 When world oil prices soared so did the subsidy; ironically, however, the subsidy proved easiest to sustain when oil was dear and thus large windfalls flowed to the State budget from Pemex oil sales overseas. Even when oil prices plummeted in the late 1980s the price charged to CFE and LFC for fuel oil was only 70% of its true opportunity cost. The philosophy of import substitution and “Mexican sovereignty”

had been built into every aspect of the Mexican power system; even today, a reliable political strategy for opposing reform of state-dominated enterprises is to hype the threat to Mexican sovereignty. Low fuel prices allowed for tariffs that were also set far below their opportunity cost. However, the exact relationship between tariffs and costs is difficult to assess because, even today, there are no credible statistics on the true cost of electricity production in Mexico. We attempt to compare tariffs with costs by comparing Mexican tariffs with those in the U.S. (see Figure 4). Mindful that there are many differences between the systems, such a comparison is nonetheless a useful place to begin in assessing Mexican tariffs. In most jurisdictions in the U.S. tariffs were set “in the public interest” by independent regulators and implemented by privately owned utilities whose stockholders demanded that the enterprise cover its cost and make a predictable profit. In Mexico, the function of regulating tariffs was (and still is) played by the Secretaría de Hacienda (Ministry of Finance) and is an extension of the development strategy that the government pursues at any given moment. Often, the agenda at Hacienda has not been compatible with the needs of a financially self-sustaining power sector—as in all the other countries examined in this book, such mismatches would not necessarily cause turmoil in the sector so long as the government was also willing to cover the difference (usually indirectly through its financing of new projects). Hacienda

Repsol YPF-Harvard Kennedy School Fellows Research Papers

96

Figure 3 World Oil Price and Relative Price for Fuel Oil in Mexico Until the early 1990s fuel oil prices were regulated in Mexico at a level far below the true opportunity cost. World oil prices (left scale) is the average cost of acquisition of Mexican heavy crude by U.S. refiners; the relative price of fuel oil (right scale) is the ratio of price charged for fuel oil in Mexico (available only since 1974) vs. a similar product in the U.S. (residual fuel oil #6, 3% sulfur, Gulf Coast average).

200%

70 Oil Relative Price of Fuel Oil

180%

60 160%

USD Per Barrel

50

140% 120%

40

100% 30

80% 60%

20

40% 10 20% 0%

19 7 19 0 71 19 7 19 2 73 19 7 19 4 75 19 7 19 6 77 19 7 19 8 79 19 8 19 0 81 19 8 19 2 83 19 8 19 4 85 19 8 19 6 87 19 8 19 8 89 19 9 19 0 9 19 1 92 19 9 19 3 9 19 4 95 19 9 19 6 97 19 9 19 8 99 20 0 20 0 01

0

consistently set tariffs for public purposes (e.g., street lighting) for agriculture and for residential service at levels below those of the U.S.—a reflection of the importance of rural agricultural and low/middle income class voters to the construction of PRI, and also the tendency not to draw a strong line between core public functions and the full cost of services supplied by state-owned enterprises. In many respects, the integrated state budget was like a gigantic shell game. In general, tariffs for the other classes remained well above U.S. levels, which we conjecture is the result of at least two forces. One was that the state-owned power system in Mexico was less efficient than the U.S. power system—payrolls were larger, the aversion to outside equipment meant that technical losses (although not known) were probably larger, and quality of service was lower.16 The other factor was the ability of Hacienda to extract higher rents from commercial and industrial consumers, which are not a power base for PRI.

Overall, the Mexican power sector’s tariff policy seems to have been broadly reflecting costs until 1973. Electricity prices were a bit higher than in the US but it was probably due to the oil-intensive and somewhat inefficient Mexican system. After 1973, there is a clear shift in the tariff policy that appears to mirror the shift in the country’s general economic policy—an inward policy that allowed Hacienda to lower tariffs with the help of oil money. Lower tariffs were used as an inflation control policy followed by the government during that time. The late 1970s through the 1980s marked a peak period for state control and budgetary shell games, thanks to the lubrication of oil revenues. This is evident in figure 4 which reveals that the Mexican tariff level has followed the availability of oil subsidies, with a delay of about two years for the normal cycle of state budgets. Tariffs declined sharply in 1981 (after the oil price windfall created by 1979 Iranian revolution) and then climbed in the late 1980s as the cost of subsidy mounted and oil prices softened.

Victor G. Carreón, Armando Jimenez San Vicente, and Juan Rosellón

97

Figure 4 Mexican Electricity Tariffs as a % of US Tariffs 200 Commercial

Industrial

Public

Residential

* Agricultural X Total

180 160 X

140

X X X

Percentage

120

X

* *

100

* *

X

X X X X

80

X

*

X

X

X

60 40 20

*

X X X

*

X * * * X

X

X

*

X X X

X

*

X X

X

X

X

X X

* * *

* * *

* *

* * *

19 7 19 0 7 19 1 72 19 7 19 3 74 19 7 19 5 76 19 7 19 7 78 19 7 19 9 8 19 0 81 19 8 19 2 83 19 8 19 4 85 19 8 19 6 87 19 8 19 8 89 19 9 19 0 91 19 9 19 2 93 19 9 19 4 9 19 5 9 19 6 9 19 7 98 19 9 20 9 0 20 0 01

0

* * * * * *

* *

*

Source: CFE–SENER.

3.

Shifting the State-Dominated Economy: 1982 to 2003

3.1

1980 to 1989: The lack of resources to increase infrastructure

Starting in the early 1980s the Mexican government’s framework shifted from a situation in which, in the words of former President José Lopez Portillo (who ruled from 1976 to 1982), “Mexico was managing its oil wealth” to a scenario with spiraling public debt and hyperinflation. By 1980 the Mexican Government was operating with a total deficit of 7.5% of GDP, and the electricity sector alone ran a deficit at almost 2.4% of GDP—all financed by extraordinary oil revenues. Despite oil prices that remained high in the early 1980s, shortly after Miguel de la Madrid assumed the presidency in 1982 Mexico defaulted on its external debt. The shock of this financial crisis created a window of opportunities for reformers who imposed tight fiscal controls, dismantled

the import substitution strategy, integrated Mexico into the world economy, and reduced the role of the state in the local economy; ever since, the role of state-owned enterprises in the economy has declined steadily—first as a fraction of GDP and then, after delays, in the aggregate workforce (Figure 5). Among the few industries that escaped privatization in the two decades of reform that followed were the two areas with the greatest implications for the state budget—electricity as a drain, and oil as a source of revenue. The failure to disengage the electric sector is evidence of key political and constitutional factors at work—to those factors we now turn. One of Hacienda’s immediate responses to the crisis was to adjust prices with the twin (often incompatible) goals of reducing financial losses caused by low tariffs while at the same time taming hyper-inflation. Hacienda increased the price of fuel oil burned for electricity and also reformed commercial and industrial tariffs, which in 1983 had reached a historical low while keeping flat the

Repsol YPF-Harvard Kennedy School Fellows Research Papers

98

Figure 5 Role of State-Owned Enterprises in the Mexican Economy 12.00 SOE economic activity (% of GDP)

SOE employment (% of total)

10.00

8.00

6.00

4.00

2.00

0.00 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998

more politically sensitive residential and agricultural tariffs (Figure 4). On the assumption that industry could pay a stiffer rate, the cross-subsidy from industrial and commercial users to the others grew over the following years. These modest reforms on fuel prices and tariffs bought time, but they did not fix the structural problems within the state-owned power sector. 3.2

1990–2000: Structural reforms towards building a new market architecture

During the 1990s, Mexico shifted from a country that avoided foreign direct investment to one that actively sought it—especially in export-oriented industries. Through expanded access to markets offered through trade agreements—notably NAFTA (1992) and the World Trade Organization (1994)—the value of Mexico’s exports almost quadrupled from 1990 to 2000. These investor- and trade-friendly reforms also created buffers for the Mexican economy that, in contrast with the 1982 crisis, have made it easier for Mexico to weather subsequent macroeconomic shocks. Nonetheless,

each financial crisis since 1982 has brought stern limits on public debt, which in turn has limited the ability of CFE, a public company, to raise the capital needed to build new plants at the pace of rising demand. In contrast with the 1970s (see Figure 2), during the era of crises—from 1982 through the 1990s (and perhaps the present)—the growth in supply and demand were more unpredictable; reserve margins varied widely because of lack of investment in capacity as demand was steadily growing. A new financial crisis in 1994–1995 proved a breaking point. Politically, this crisis induced a strong change in the electorate preferences that allowed for a new composition of the Mexican Congress after the midterm elections in 1997. After more than 65 years of control, the ruling party (PRI) lost its majority in Congress (see Table 2). PRI’s absolute majority in both houses of Congress had long been a crucial asset for the PRI-controlled presidency. Any policy that the President (and PRI) sought to implement—such as import substitution in the earlier era, and shifting from the state dominated economy along with free trade agree-

Victor G. Carreón, Armando Jimenez San Vicente, and Juan Rosellón

99

Table 2 Political Control of Congress: The Percentage of Deputies and Senators

PRI 1964 1967 1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 2000 2003

83 83 84 82 82 74 75 72 52 64 60 48 42 45

Deputies (lower house) PAN PRD 10 9 9 11 8 11 13 10 20 18 24 24 42 31

0 0 0 0 0 0 0 0 0 8 14 25 10 19

Others

PRI

7 8 7 7 10 15 12 18 38 10 2 3 6 5

100 100 100 100 100 100 100 100 94 95 74 60 47 47

ments during the era of reforms—could be assured a working majority. Any significant opposition came from within the establishment itself and could be addressed within the PRI apparatus. Economically, this crisis also had seismic effects because the government’s negotiated settlement with its creditors included a prohibition against state-owned enterprises incurring additional debt. For the power sector, this did not seem a substantial concession—the economy was expected to tip into recession and thus demand for power would be sluggish, and a considerable excess capacity was available from the years of over-building. Reality proved to be quite different—integration with the U.S. fueled a rapid growth in Mexico and led to power demand that rose at a much higher rate than expected. The government found relief in Amendments to Mexico’s Ley del Servicio Publico de Energía Eléctrica (LSPEE), which was altered in 1992 to allow private participation under different schemes such as Independent Power Production (IPP), Cogeneration and Self Supply (see Table 3). These legal reforms had been undertaken to comply with the energy chapter of NAFTA, which was artfully constructed to permit continued state control of oil and electricity sectors (as enshrined in Articles 27 and 28 of the Mexican Constitution) while at the same time allowing for private participation in the power sector. As shown in Figure 6, an interlocking array of constitutional, international

Senators (upper house) PAN PRD 0 0 0 0 0 0 0 0 0 2 20 26 36 36

0 0 0 0 0 0 0 0 6 3 6 12 12 12

Others 0 0 0 0 0 0 0 0 0 0 0 2 5 5

and national laws then applied in the power sector, and as we will see the interaction between these laws has strongly shaped the outcomes. With the LSPEE already on the books, though not yet implemented, the government jumpstarted the IPP program to alleviate the looming crisis in power supply caused by CFE’s inability to contract debt. The first tender (Merida III, a combined cycle gas-fired plant) was awarded in January 1997. In practice, IPPs on their own were not a miracle solution because generators still had to sell their power through one of the state-owned distributors—LFC or CFE—and the power purchase agreements (PPAs) that underpinned IPP investments were, in essence, a form of long-term debtlike commitment that the post-1995 settlement would seem to have forbidden. The proposed solution was a shell game established in December 21st, 1995 when the Mexican Congress approved reforms to the Public Debt and the Budget Laws that created a new scheme for the development of long-term infrastructure projects, currently known as PIDIREGAS. This scheme is tailor-made for IPPs.18 Under this scheme, only the capacity payments of a Power Purchase Agreement (PPA) of the starting and the following year are accounted for as liabilities. Future payments are considered as contingent liabilities but are not included in the government’s yearly budget. Since IPPs work under long-term PPA contracts, it would

Repsol YPF-Harvard Kennedy School Fellows Research Papers

100

Table 3 Activities considered for private participation Scheme

Description

Self supply

Generation of electricity to meet an industrial facility’s own energy needs. Refers to power plants owned and operated by private companies

Cogeneration

Refers to electricity generated simultaneously with steam or other types of secondary thermal energy to be used in an industrial process, or the generation of electricity from the surplus of thermal energy of an industrial process

Independent Power Production

Refers to power plants with installed capacity larger then 30 MW, built and operated by private companies. All generated power must be sold to CFE under a power purchase agreement

Imports and Exports

Exports refer to electricity produced under cogeneration, IPP or small scale generation categories. Imports refer to electricity exclusively used for self-supply purposes.

Small-scale generation

Refers to power plants with an installed capacity no larger than 30 MW built and operated by private companies. This electricity is to be sold solely to CFE.

seem normal that the capacity payments were not considered as liabilities. However, since the Mexican power sector is operated as a vertically integrated publicly owned monopoly, every single IPP project that is operating, or under construction, has sought and received explicit government guarantees. These guarantees are in essence contingent liabilities, which must be handled more like normal liabilities. As of June 2003, PIDIREGAS debt for CFE alone amounts to USD $4.3 billion, with payments distributed over the following 10 years. So far, there has never been a default on any of the PIDIREGAS liabilities; even as investors in power plants have lost vast sums in many other developing countries, all of the contracted IPPs are rewardFigure 6 Legal Framework for the Mexican Power Sector Mexican Constitution

North American Free Trade Agreement (Chapter VI)

Ley de Servicio Publico de Energia Electrica (LSPEE)

Rules from LSPEE

Rules from LSPEE related to contributions

Administrative rules

Administrative rules

ing investors more or less as expected. For investors and government managers the scheme is attractive; however, since PIDIREGAS backs PPAs denominated in 2003 dollars there remains a substantial devaluation risk—a constant feature of Mexican financing for the last three decades that might prove to be a major problem for power sector investors. Merida III entered into service in 2000; since then, 3,495 MW of capacity have been added through IPPs, which has contributed considerably to restoring the sector’s reserve margin. As shown in Figure 7, from 2000 to 2002 about half the new capacity came from IPPs. In 2002, one-third of the new capacity came from self-generation and cogeneration facilities—that is, power plants that are located at industrial sites outside the direct control of CFE and LFC. Barely one-third of the new capacity from 2000 to 2002 came from the traditional CFE and LFC-dominated model of power plant construction. Despite this new surge in investment, IPPs alone were not enough to meet all the growth in demand, and there are several indicators of the chronic underinvestment due to the continuing severe restrictions on public debt. First, reserve margins have slipped—to just 1% in summer 2002—and have been maintained in part by delaying the retirement of old plants, especially plants that burn high cost fuel oil. Second, the government has slashed the authorized budgets for maintenance and repair—typically, as shown in Table 4, to levels on average 30% lower than the

Victor G. Carreón, Armando Jimenez San Vicente, and Juan Rosellón

101

Figure 7 Growth in New Generating Capacity 10.00% LFC

CFE

PEMEX

Self Supply–Cogen

IPP

8.00%

6.00%

4.00%

2.00%

0.00%

1990

1991

1992

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

–2.00% Source: Secretaria de Energia and CFE

level that CFE executives think is required. These short-term measures helped to preserve resources for capital investment and helped to avert crisis in the power sector, but they were merely stopgap measures. Even more worrisome than these problems with current investment are the inconsistencies laden in CFE’s official planning forecast for the next ten years: a 25,000 MW increase in net installed capacity through the addition of 28,000 MW of new plants; with planned retirements amounting to only around 4,100 MW, about half the level expected.20 From both the financial and technical perspectives, the power sector would

appear to be in serious trouble. The Mexican State will be unable to meet these growth targets because it has no financial resources itself for investment in the required new capacity.22 IPPs can meet some of the shortfall, but the confidence of IPP investors may wane as the latest scheme to defer crisis—the PIDIREGAS mechanism— becomes exhausted. Two conclusions can be drawn from the experience, so far, with the IPP program. First, it has resulted in almost no change in the market architecture of the sector. Although a leap for private investors, IPPs are a stopgap measure. By design, they exist inside the LFC & CFE-dominated sys-

Table 4 Solicited and Authorized Budgets for Maintenance and Repair at CFE

Solicited Budget (Millions of Pesos) Authorized Budget (Millions of Pesos) Proportion Source: CFE

1995

1996

1997

1998

1999

2000

2001

2924 1904 65.1

3128 2011 64.3

3403 2502 73.5

2610 2045 78.4

2514 2002 79.6

2844 2045 71.9

3150 1937 61.5

Repsol YPF-Harvard Kennedy School Fellows Research Papers

102

Table 5 Mexican Electricity tariff/cost ratios Consumer Class

1995

1996

1997

1998

1999

2000

2001

2002*

Residential Commercial Public Service Agricultural Medium Industrial Large Industrial Average

0.47 1.31 0.88 0.33 0.88 0.81 0.71

0.42 1.16 0.79 0.28 0.84 0.83 0.70

0.40 1.13 0.81 0.28 0.91 0.91 0.74

0.43 1.21 0.94 0.30 0.92 0.90 0.79

0.41 1.19 0.92 0.29 0.91 0.90 0.74

0.41 1.07 0.88 0.28 0.85 0.85 0.70

0.42 1.07 0.90 0.29 0.87 0.83 0.70

0.50 1.05 0.90 0.30 0.93 0.90 0.74

* Estimated Source: Secretaria de Energia

tem and require minimal adjustment of that structure. They solve an immediate problem—surging demand but stagnant supply and aging incumbent plants—at considerable cost that is largely not transparent in current state accounts. Second, IPPs have had a dramatic effect on the technology available in the sector in ways that are probably quite beneficial for Mexico. All of the IPP projects have been built by foreign companies using stateof-the art combined cycle gas-fired technology— with gas purchased from the U.S. or from PEMEX. Whereas in other countries examined in this book the introduction of gas has been difficult because fuel costs in a gas system are higher than the incumbent coal (China, India, South Africa) or hydro (Brazil). In Mexico, the incumbent is expensive and gas plants not only have lower capital cost but also are less costly to operate—in addition to being much cleaner. Even as the creation of the IPP and the PIDIREGAS schemes offered new tools to avert crisis, Hacienda continued to reform fuel prices and tariffs with the aim of restoring some sustainability to the sector. Nonetheless, tariffs appear still to be at a level below cost—especially as the cost basis of the oil-intensive Mexican power sector is much higher than in the U.S., where low-cost coal is the dominant primary energy source. According to the Ministry of Energy, Mexican electricity tariff/cost ratios are as shown in Table 5.24 Despite these increases in tariffs, the sector still loses vast sums of money. Official figures estimate a net subsidy of USD $5 billion a year (see Figure 8), principally because of residential and agricultural tariffs are set way below cost—residential tariffs alone could be as

high as 3% of GDP. (The distributional effects of this subsidy are enormous; total tax collection, outside the oil sector, is only 10% of GDP.) In 2000, residential consumers received 64.1% of the total subsidy; the industrial sector, 17.9%; the agriculture sector, 11%; and the commercial sector, 5.3%. As a consequence of this policy, residential consumers face a tariff that is among the lowest in the world; but, it relies on a regressive scheme as shown by López-Calva and Rosellón (2002). A new 31-category tariff scheme adopted at the end of 2000 marks a further step at rationalization; still, the residential tariffs remain below cost—implying a subsidy for 98% of users. Politically it has proved extremely difficult, if not impossible, to raise residential and agricultural tariffs. Thus, most analysts conclude that the only practical way to make the sector financially sound is to reduce costs—yet that, too, is politically challenging as it requires confronting the powerful unions that are embedded in CFE and, especially, LFC. These unions—Sindicato Mexicano de Electricistas at LFC in a cross alliance with the leftist Party PRD, the leftist wing of PRI, and some other social organizations and unions and the Sindicato Enico de Trabajadores Eléctricos de la República Mexicana (SUTERM) at CFE (which has a mixed position on the reform issue)—have led a broad coalition to block any attempt to allow private investment into the sector or to modify significantly the market architecture (e.g., tariff reform) in ways that could hurt their interests. Since the SME and the SUTERM are well-organized interest groups with the capacity to mobilize votes, no political Party has been willing to face the political cost of supporting a modification of the electricity

Victor G. Carreón, Armando Jimenez San Vicente, and Juan Rosellón

103

Figure 8 Evolution of Total Subsidies to the Power Sector 60000

6 Total Subsidies (Million pesos)

%GDP

50000

5

40000

4

30000

3

20000

2

10000

1

0

0 1995*

1996

1997

subsidy policies or a substantial modification in the market architecture to allow for the implementation of competitive fares..26 If both consumers and unions oppose changes, then it becomes a risky business to pass a bill which eventually could costs votes or popular support for the involved parties. The strongest of the referred political opposition to reform became evident in 1999 when the first profound reform of the sector was attempted by President Ernesto Zedillo. Before then, policy makers under President Carlos Salinas, mindful of the political sensitivity of the energy sector, attempted only partial reform. In 1992, amendments to the LSPEE—the basic legal architecture for the power sector that had been codified in 1975—allowed IPPs into the sector (discussed above) and also empowered a new institution: the independent regulator. The strategy was to make the true costs of generating power more transparent—through market competition—and to empower independent regulators who would be able to scrutinize costs. In addition to promising the delivery of electric service at lower cost, a shift to competitive electricity markets would make it

1998*

1999

2000

possible to remove key operational decisions in the sector from the grip of the unions. Markets built around transparent rules as well as tariffs set at levels that ensured recovery of costs would attract private investment into new generating capacity and would also allow CFE and LFC to direct their scarce resources towards dire needs such as repair and maintenance of their existing assets. Moreover, they could implement better management of the system to reduce theft of electricity, which is a rising problem that threatens to undermine further the financial soundness of the system. Indeed, the experience in telecommunications, highways, the pension system, and the banking system, seemed to confirm, at the time, that privatization and the introduction of market forces would lead to an influx of private capital that could constrain the government’s ability to torque tariffs to its macroeconomic and political agendas.28 Today, the political case for privatization and market reforms is thus extremely difficult to make—indeed, policy makers often engage in verbal and legal contortions to argue that the proposed reforms do not involve privatization and unfettered markets.

104

3.3 The regulator in the power sector The 1992 Amendments to the LSPEE did not alter the Mexican Constitution, and thus the State still held the exclusive right to generate, transport, and supply electricity for public service. Reforms in this context—as we have seen with the emergence of IPPs—required careful balance to preserve the constitutionally-assured role for the central government. The creation of an independent regulator presented a new frontier in this balancing act— although formally part of the government, the regulator would have an arm’s length relationship with the traditional entities of government precisely to preserve the political control of electricity that was originally envisioned in Article 27 of the Constitution. In 1993 the government created by decree the CRE as an advisory body on gas and electricity issues. In October 1995, the Ley de la Comisión Reguladora de Energía (Energy Regulatory Commission Act, LCRE) transformed CRE into an autonomous agency in charge of regulating the natural gas and electricity industries. The CRE has its own budget (which is allocated via the Energy Ministry with few strings attached) and has technical and operational autonomy. It consolidates functions that had previously been scattered among several agencies, and pursuant to its enabling Act, in the electric sector CRE is empowered to perform several key tasks:30 (a) Participation in the setting of tariffs for wholesale and final sale of electricity, (b) Issuance of permits to generate electricity under the schemes allowed by the LSPEE, (c) Review and approval of the criteria for determining fees related to public electricity service, (d) Verification that entities responsible for the public electricity service purchase electricity at the lowest cost and also offer optimum stability, quality, and safety of electric service. (e) Approval of the methodologies for calculating payments for the purchase of electricity used in public service, and (f) Approval of the methodologies for calculating payments for electricity transmission, transformation and delivery services. In addition to these functions, CRE also performs similar functions in the gas sector, including the issuance of building and operating permits for gas infrastructures. By 2003 CRE had granted 218

Repsol YPF-Harvard Kennedy School Fellows Research Papers

permits in all schemes accounting for investment commitments over USD $12.2 billion for the construction and operation of nearly 20,000 MW of capacity. Nonetheless, CRE’s authority and power are not clearly specified in many areas, and CRE influence is hobbled in key areas—such as in tariffs, where the operations of LFC and CFE are far from transparent and thus rational tariff-setting is essentially impossible. CRE approves the methodologies for calculating payments for electricity transmission and distribution, but CRE does not have the authority to actually establish tariffs. Despite reforms to create an independents regulator and allow the entry of IPPs, the fundamental barrier to competition and private participation remained—Articles 27 and 28 of the Constitution. In February 1999, near the end of his tenure, President Zedillo proposed structural reforms that would have modified the Constitution, but these never passed the Congress. Many fractions inside PRI opposed reform that could erode a traditional power base—the unions in CFE and LFC—and they relied on a public that remembered the failed promises of earlier privatizations.32 In addition to opposition within his own party, Zedillo’s earlier political reforms meant that he didn’t have a working majority in the Congress (see Table 4), which required him to negotiate with many different parties to achieve the support needed for passage of his proposals. In order to amend the Constitution a majority vote of 2/ 3 of each House and 51% of Local Congresses are needed. February 1999 proved to be a difficult time for such negotiations as few were willing to compromise with the July 2000 Presidential elections on the doorstep. These political factors combined with the lack of general awareness about the problems in the sector. To the casual observer, everything appeared to be working well—costs and quality were not out of line with the experience of most Mexicans. Zedillo’s plan sought a comprehensive reform that would introduce competition in generation, distribution, and marketing of electricity. The proposal followed closely the UK model, although studies have shown that alternative systems— such as the Australian system with a regulated market for capacity reserves—would be more appropriate in the Mexican context (Carreón and Rosellón, 2002b). Under the Zedillo plan, nuclear generation, some hydro generation (mainly in the

Victor G. Carreón, Armando Jimenez San Vicente, and Juan Rosellón

south of the country), and the system operator would remain in the hands of the State—nuclear for reasons of security, and large hydro because the State manages the nation’s water supplies for multiple purposes including agriculture. The independent regulator, CRE, would also oversee the aspects of the system that were prone to monopoly, such as transmission and distribution. Regulators would also ensure that generation and marketing would remain contestable activities, through monitoring of market power, barriers to entry and other factors that would undermine a competitive market. The Zedillo plan envisioned three stages of effort. First, the government would implement basic organizational changes. CFE and LFC would be partially unbundled into several generation, transmission, and distribution companies kept at arm’s length; a government-controlled system operator would be created. Separate state enterprises would be created to hold nuclear and hydro assets. And basic rules for a competitive electricity market (and its regulatory framework) would be debated. Despite the failure to implement the Zedillo plan, some progress on this first stage was already accomplished when CFE created a “shadow market” in which generators compete for service at 1400 nodes through the use of a power flow model. Since September 2000, CFE’s “shadow market” has sought to emulate a truly open, competitive market; it uses a merit order rule for dispatching generators and includes a one-day-ahead market as well as a “real-time” balancing market. In the one-day-ahead market, bids for hourly slots are submitted to CFE’s system operator by thermal plants that are administratively separated so that they plan their strategy, to some degree, as different power producers.34 Payments to generators include a “capacity” payment intended to foster the development of generation capacity reserves. In this shadow market, distribution companies are also divided into several units; a MWMile method is used to set transmission tariffs.36 The second stage of Zedillo’s proposal envisioned opening the sector to private investment and the creation of a wholesale electricity market that included both short-term and long-term markets as well as competition for contracts with distributors and large users. In the third, final, stage the arm’s-length entities would be separated fully and privatized.

105

In addition to political obstacles, the Zedillo plan had some technical problems. It sought to balance state-of-the-art economic theory with the practical realities in the context of the Mexican power sector. One of the main omissions was the lack of a mechanism for creating incentives to expand transmission capacity. The plan envisioned that the State would not bear risks nor provide guarantees to private investors; yet it vested transmission planning solely within the Ministry of Energy and potentially created risks that most private investors would avoid unless given a state guarantee similar to the PPA that IPP generators required. Related to this problem was the lack of incentives to address problems of short-run congestion, which in turn could create bottlenecks for new generators.38 Nor was there clarity in the incentives that would govern the system operator. Finally, it was not clear how the IPPs were incorporated into the reform—the state would retain strategic control of the sector, but it was unclear how to square that vision with investors’ requirements for predictable returns on their projects. Alas, these flaws and the many possible remedies were never given serious consideration—the upcoming election and the fragmentation of political power eviscerated the Zedillo plan before the government ever had a chance to build a political coalition for its passage. 3.4

Electricity and the Social Contract

Although key choices about fuels and tariffs made during the 1970s would later undermine the financial sustainability of the power sector, that decade was a period of substantial progress in delivering benefits from electricity more widely to the society—what the editors call in the introduction to this book the “social contract.” We focus on three dimensions: electrification of rural and poor populations, protection of the environment, and investment in long-term research and development. On all three, the accomplishments rooted in the 1970s were notable, and ironically these achievements (especially that of electrification) have reinforced public support for a state-controlled power system. The greatest success in these three dimensions of the social contract is evident with electrification. Access to electricity more than doubled from 1970 to 1990 (see Figure 9). Residential and agricultural

Repsol YPF-Harvard Kennedy School Fellows Research Papers

106

Figure 9 Population with access to electricity 100,000

100 Population with access

Population in Thousands

80,000

90

Population without access % Percentage of population electrified

80

70,000

70

60,000

60

50,000

50

40,000

40

30,000

30

20,000

20

10,000

10

1910 1920 1930 1937 1940 1950 1952 1960 1970 1980 1990 1997 1998 1999 2000

Percentage

90,000

0

Source: CFE.

tariffs declined in the 1970s, which aided electrification, but the progress in electrification has continued even through the flat and rising tariffs of the 1980s. Even as the sector has experienced enormous financial difficulties in the 1990s, electrification continued apace. By 1997, 94.7% of the Mexican population had access to electric power. At this writing (2003), penetration has reached 96%, despite the country’s complicated geography and remoteness of small settlements in diverse rural areas. Despite this achievement in aggregate, some states have lagged markedly—notably, Oaxaca, Chiapas, and San Luis Potosi where there is a high percentage of indigenous communities living in remote rural areas where the cost of service is high. Many factors could explain the pattern of electrification. In Table 3 we report simple correlations using basic demographic and economic statistics from all states between 1970 and 2000 (one observation per state per decade). The correlation with electrification is highest for GDP (R2 = 0.81) and urbanization (R2 = 0.99). A multivariate regression

confirms these simple results—urbanization has been the main driving force for electrification, and there is little residual value that might indicate a role for policy. Similar results are evident for water services, but in telecommunications the correlations are much less robust—suggesting that public policies promoting access have been more important for telecommunications or, perhaps, the cost of telecommunications has declined so sharply that factors such as urban access and income have a less intense effect than in the public services where costly fixed infrastructures remain central. The story of successful electrification in Mexico is similar in many respects to that of China—factors outside the electric sector have spilled over to create dramatic progress in electrification. This history is quite unlike that of South Africa, where success in electrification in the 1990s is the direct consequence of an active government policy to promote electric connections. Second, on environment, the sector is subjected to increasingly strict regulation concerning siting and effluents. The relevant norms are under

Victor G. Carreón, Armando Jimenez San Vicente, and Juan Rosellón

107

Table 3 Correlation Coefficients for the Mexican Economy Coefficients calculated with single-year estimates for 1970, 1990 and 2000 (1980 is incomplete) reported by the government.

State GDP Urbanization Residential Electrification Water Services Access to Telecommunications

State GDP

Urbanization

1.00

0.81 1.00

renewed consideration at the present as Mexico considers the possibility for even stricter rules based on improved state-of-the-art technology. The government is in the midst of designing a credit trading system for regulating large sources of sulfur dioxide—including power plants as well as the many facilities of PEMEX. Progress on environmental issues depends heavily on the rate of technological change in the electric sector. As in the United States, most environmental laws in Mexico “grandfather” existing facilities with weaker regulations, and thus the difficulties in the power sector that have resulted in slowing the retirement of old plants have had negative consequences for the environment. The biggest news— and good news at that—is the arrival of gas in the sector, which is mainly a function of technological improvements (gas turbines) that occurred outside Mexico as well as decisions on IPP tenders that were taken in part because gas is cheaper than the oil alternatives. The environmental benefits are a windfall. Third, on investment in innovation, two institutions support long run research and development in the power sector: the Instituto de Investigaciones Electricas (Electric Research Institute, IIE) and the Instituto Nacional de Investigaciones Nucleares (National Institute for Nuclear Research, ININ).40 The IIE was created by Presidential Decree in December, 1975, as a public decentralized entity with legal personality and own patrimony, with scientific and technological character. The origin of the ININ goes back to 1956 when the Nuclear Energy Commission was created as the Institution in charge of research and regulation in nuclear issues. Later on, in 1979, the National Commission on Nuclear Security and the National Institute for Nuclear Research were cre-

Residential Electrification

Water Services

Access to Telecommunications

0.81 0.99 1.00

0.80 0.98 0.99 1.00

0.68 0.42 0.36 0.34 1.00

ated to separate those activities. Since then the ININ is in charge of basic and applied research and technological development on nuclear and related matters. 4.

The Evolving Agenda

Under the current legal framework for the electricity industry, private investment co-exists with the state in key areas, such as power generation. Nonetheless, the reforms implemented so far are stopgap measures—they are minor reforms in tariffs and fuel pricing implemented from 1982 to 1990, the IPP scheme created in 1992, the empowerment of CRE in the 1990s, and a new tariff schedule adopted in 2000. Each of these measures push crisis a bit further into the future; but, the sector remains financially unsustainable. Indeed, the engine for partial reforms is now running out of steam as the gap between expected demand and supply grows and the financial needs of the sector multiply. Yet the need for reform has not commanded adequate political support, and the fragmentation of political authority has made it even harder for government to assemble viable reforms. Serious reforms will require institutions such as a truly independent regulator with substantial powers and information—all conditions that are difficult to satisfy in the current context. Moreover, serious legal problems remain so long as reformers have attempted to navigate around the constitutional restrictions on private participation in the sector. The Mexican Supreme Court ruled in 2002 that the 1992 law—the cornerstone to the IPPs and CRE’s authority—might be unconstitutional, which has cast a shadow over investors. A myriad of proposals has induced madness in the public opinion on these topics as consumers (and some

108

key actors) do not know with certainty what is going on in the sector and what to expect in the near future. Meanwhile, time is running and Mexico is getting closer to a critical situation in its power sector. We examine each of the referred issues in turn—the need for new investment to close the gap between demand and supply; the financial sustainability of the sector; the constitutional challenge and the role of the Supreme Court; the authority and role for CRE; tariffs; public opinion; and the main current proposals for reform. 4.1 Demand, Supply and Gas The government expects that from 2001–2011 electricity demand will rise 5.6% per year. At present, most of the total capacity (about 41 GW in 2002) is supplied by hydroelectric and conventional steam plants fired mainly with oil (23% and 42% of the total, respectively). Combined cycle generation accounts for only 18%, although these plants are the newest. About 44% of the generating power plants are at least 30 years old. If the power sector expands as expected, about USD $25 billion in investment will be required through 2006; in total, from 2003–2011, the expected investment cost will exceed USD $50 billion, with about 40% for generation, 24% for transmission, and 21% for distribution. Of this total, the Ministry of Energy envisions that various private sector investment schemes, notably IPPs, will contribute USD $39 billion— about four-fifths of the total. Nonetheless, the (smaller) requirements in the public sector will impose extraordinary strain on the budget and could divert resources from other social priorities such as education, social security, or poverty relief. To serve the growing demand for power and replacing the retired plants a variety of fuels is available, but one (by far) is the most attractive: gas, especially gas burned in combined cycle baseload plants. About 90% of the 18,700 MW of new capacity scheduled to open by 2006 is gas-fired combined cycle. By 2011, half of Mexico’s expected total generating capacity of 64,000 MW will be gas fired.42 Demand for gas will rise accordingly— about 7.4% per year over the next decade.44 By 2010, perhaps 60% of all gas sold in Mexico will be burned for electricity generation. This shift to gas is good news for the environment and also promises to lower tariffs. However,

Repsol YPF-Harvard Kennedy School Fellows Research Papers

it is not clear how such a massive shift to gas will be achieved. Close to the end of the Zedillo’s administration, The Ministry of Energy and PEMEX announced an ambitious program, Plan Estrategico de Gas Natural (Strategic Plan for Natural Gas, PEG) that outlines a vision for meeting this demand, calling for PEMEX to double its natural gas production from 2002 to 2006. However, actual progress at PEMEX has been lackluster—the PEMEX budget is set by Hacienda, and as with CFE it has not received all that it requests. In tough times, PEMEX focuses on its core business (oil) and shunts gas aside. PEMEX lacks not just the capital but also the expertise to develop new gas fields, so it has turned to Contratos de Servicios Multiples (Multiple Service Contracts, CSM)—a scheme to allow private participation in natural gas extraction in the Burgos fields (in the northeast of Mexico) without actually conferring ownership on the fields to the non-PEMEX operators (which would contravene the Mexican Constitution, which assigns sole authority over hydrocarbons to the state). The CSMs, however, have come under a similar cloud that threatens the constitutionality of reforms in the electric sector, and most foreign operators remain wary of participation under those terms. Nonetheless, the first CSM were granted to the Spanish Oil firm Repsol-YPF during the fall of 2003. The monopoly position of PEMEX includes not just control over fields but also pricing and retailing of gas. The current regime sets gas prices on a “netback” basis to Texas markets, which made sense when most gas was imported from Texas but yields undue windfalls to gas suppliers (i.e., PEMEX) as large indigenous supplies are envisioned. It also creates difficulties for gas-on-gas competition as LNG terminals are built and will compete for contracts with local natural gas supplies. CRE has directed PEMEX not to discriminate in its pricing and marketing of gas, but the problems are structural. Gas is used in IPPs by private investors who are stuck between a monopsony (CFE) and a monopoly (PEMEX). This situation is damper for competition as, in general, fuel costs account for 60% of the total costs of gas-fired electricity. The pervasive problem of competition in the gas sector extends even to the siting of power plants which, in effect, is determined by PEMEX and its decisions about location of the gas transportation infrastructure.

Victor G. Carreón, Armando Jimenez San Vicente, and Juan Rosellón

4.2

Constitutional Integrity: The Supreme Court Decision and Beyond

The keystone to sustaining investment in the power sector, in the context of the very limited reforms that have been implemented so far, is the 1992 amendment to the LSPEE, which created the framework for IPPs. In May 2001, President Vincente Fox proposed further reforms to Articles 126 and 135 of the LSPEE which would have modified the terms and limits of the self-generation and co-generation schemes to make them more attractive to private investors. Banking on success of this proposal, the Fox administration was already projecting that by the year 2011, about half of the country’s generation would take place under the selfgeneration and co-generation schemes. However, on July 4th, 2001, the Mexican Congress filed a petition before the Supreme Court for review of the proposal and argued that the proposed articles envisioned giving the Executive Branch (which would control tendering and operation of these projects) more power than allowed under the Constitution. The Supreme Court ruled in favor of Congress, but the Court did not restrict itself just to the immediate issue of separation of powers. It also speculated about the consistency of the entire LSPEE framework for private generators with Article 27 of the Constitution. The Court implied, in its decision, that, if asked, it would rule against the IPP scheme that had become the bedrock of efforts to expand the power system. Important investors—such as Electricité de France, the largest private investor in the Mexican power sector—announced that they would not participate further in the IPP scheme until further reforms that clarified their constitutional position had taken place. The Supreme Court decision illustrates that reformers must focus on just the key legislative and regulatory actions but must, finally, achieve a reform in the Constitution itself. The Supreme Court decision has introduced yet another uncertainty into an already contentious and fractious debate. Long ago—with the Zedillo proposal of 1999—the subject of electricity reform left the technical arena and has almost totally evolved in the political arena. Participation of the labor unions as well as the multi-dimensional negotiations between political parties are the main determinants of reform proposal success. To

109

provide a sense of the range of options, Table A1 in the Appendix summarizes the main proposals under consideration in late 2003, they range from PAN and President Fox’s vision, for a liberalization of the sector that builds on the Zedillo proposal while rectifying important flaws, to the proposals of the PRI and the PRD (the main Parties in opposition to Fox’s Party), which foresee tinkering at the margins of a system that would remain vertically integrated and organized much as it is today.46 According to the negotiations taking place in late 2003, the proposal that is most likely to be discussed in the Congress is the one presented in the last column. This proposal was negotiated by PAN and some fractions of PRI. However, a common factor in all these proposals is the lack of technical discussion on the specifics of the electricity sector. Even the Fox proposal seems to be totally unaware of the highly complicated task of designing an electricity market under the presence of a vertically and horizontally integrated incumbent state firm. It is not clear how under such conditions there could exist a leveled playing field for entering private generators that would have to compete with generators that belong to a state holding company that is able to deliver subsidies across its different subsidiaries. 4.3

Evolution of Tariffs in the near future

The direction of tariff reform remains difficult to predict; yet continued alignment of tariffs with costs is essential. We attempt to develop different scenarios for future tariffs by looking at each of the major contributors to final tariffs. At present, costs are allocated for low-voltage supply with about 35% for generation, 5% for transmission, 50% for distribution, and 10% for marketing. For residential customers, who account for 24.4% of the total load, the gains from distribution and marketing could only arrive via efficiency gains given the proposals presented in Table A1. So in order to discuss some possible scenarios we assume that this 60% of the current cost will remain unchanged. With respect to transmission, all the current proposals under consideration envision that the government will retain control. Therefore, any reduction in this tariff will come from efficiency gains rather than outright competition. Lower tariffs from efficiency, however, will be offset with the creation of proper incentives to invest in transmis-

Repsol YPF-Harvard Kennedy School Fellows Research Papers

110

sion assets. Even large changes—positive or negative—will have little effect on the final. Generation costs are likely to have a much larger impact on the total final tariff. Improved efficiency through competition should lower tariffs, although generation tariffs are already artificially low due to under-pricing of fuel oil.. Gas fueled power plants should increase efficiency, but the difficulties of attracting investors for indigenous gas production, and the rising gas price trend in North America, has given rise to the need for imports of LNG. The first such facilities are slated to open soon at Ensenada, Baja California, Altamira, Tamaulipas and Lázaro Cárdenas, Michoacán, with delivered costs of perhaps 30% to 50% higher than the price of domestic gas. However, it is important to note that LNG terminals make economic sense with prices of LNG above USD $4.00 per MMBTU. Although new plants should replace costly old oil-fired facilities, already those old plants, through CFE’s “shadow market,” are being dispatched only during peak periods, and the shift to a peakier load in the future—with ever-larger residential demand—may actually result in a higher cost for peak power, which reforms will attempt to pass to final users. The most likely effect of all these forces, we estimate, is a higher cost of generation. Finally, one must consider the fate of subsidy policy in a reformed environment. While there will be pressure to maintain the current subsidies, we doubt that the high cost of this program— though offset a bit, perhaps, through efficiency improvements—will allow for continuation at current levels, which amount to around 50% of the true cost for residential consumers and 70% for agricultural users. Only in the case that investors mistakenly over-invest and produce a glut of lowcost baseload power is it likely that tariffs could be kept low while subsidies are also reduced. 4.4

New Role of CRE

Although the creation of an independent regulatory authority in the middle 1990s was an enormous accomplishment, the powers and authority of CRE require further clarification—especially as key functions that are performed by regulators in other countries, such as setting tariffs, are actually controlled by Hacienda as an extension of government policy. Indeed, the Fox proposal for continued reform includes a further specification of CRE’s

role, including its role in overseeing a transparent tariff policy. A consensus is emerging that CRE should be vested with independent authority to define the rules for market operations, set tariffs, and regulate natural monopolies (see Table A1). 4.5

The greatest challenge to Reform: Public opinion

The fragmentation of politics in Mexico has exacted a considerable toll on the process of reform. Not only has debate over reform left the technical arena and become a completely politicized issue, but the constant debate and the lack of control by any single party in the Chamber (see Table 4 above) has undercut any continuity in reform strategy and made it difficult for critical investors to anticipate outcomes. Moreover, available data shows public opinion opposes privatization as well as even private investment in the energy sector. Mexicans who are even aware of the existence of reform proposals (who are in a small minority of the total public) believe that the essence of the most comprehensive reform— proposed by the Fox government—is a privatization that will undermine Mexico’s sovereignty. This view is the result of a carefully manufactured public opinion by interest groups such as unions that fear (probably correctly) that reform will harm their narrow interests. Detractors have found fertile soil for sowing discontent. The 1995 financial crisis, which cost Mexico 7% of GDP, has led many to believe that neoliberal reforms are the cause of economic malaise. Opposition parties to then PRI-led government, especially the PRD, have found success in bashing technocrats as the cause of social problems and injustice in Mexico. Once the PRI lost the presidency in the year 2000 and PRI as a party has fragmented, the core of “anti neoliberals” has swelled in numbers. In many other Latin American countries, the decade of liberal reforms has yielded a similar (and powerful) coalition of illiberal crusaders.48 These voices have found it particularly easy to be heard in countries, such as Mexico, where the 1990s liberal reforms were ridden by corruption. These observations are illustrated in a recent poll conducted in 2002 by Coordinacion de Estudios de Opinion (CEO): 36% of those who know about President Fox’s reform bill think it is about privatizing the power sector, while only 5% mention attracting private investment. In that same

Victor G. Carreón, Armando Jimenez San Vicente, and Juan Rosellón

poll, 35% of the population opposed private investment—when asked—and only 17% supported a strategy of attracting new private funds in the industry. Citizens appear to fear private money almost as much as they loathe privatization. About half (49%) of Mexicans consider the country as having electricity problems, a figure far from being overwhelming—and one-third of those who think that electricity is poor cited high prices as the main problem in the sector. Of the whole sample, only 14% said that the quality of electric service was bad, while 33% who said that service quality is good. Additionally, according to the CEO survey, only 29% believe private investment would guarantee electricity supply in the decades to come; 28% believe poor and rural communities would be electrified, while 30% do not believe that would happen; surprisingly, only 24% believe the government would channel more resources to social spending, while 36% says that promise is false. In the same vein, only 23% think service would improve as a result of private investment. Moreover, the survey suggests that the public thinks many dangers lurk in reform—among them, 60% believe worker rights would not be respected, and a majority believes that private investors will force higher tariffs. Thus reformers face a problem of credibility with the public. Their mission of reform is viewed, by many, as unnecessary and harmful. Why should people believe this time benefits will be fairly distributed among the population? Why believe corruption will be absent this time? Why will reform yield better service and lower tariffs when similar (unfulfilled) promises were made for reform of the banking system, for example? Thus old-fashioned popular politics, not economics or technical design, has become the most important factor explaining the failure of power reform in Mexico. One strategy for fixing this problem would be a massive campaign to alter public opinion by explaining the benefits of reform (improved service and a chance of reduced tariffs in the long run, as discussed above) and the current hidden costs to the status quo, such as subsidies that could be redirected to other social purposes. However, available data suggests that such an effort would not be very effective as citizens distrust the ability and honesty of the government to reallocate one peso saved in electricity to other worthy goals. A second strategy would entail

111

waiting for a more favorable composition of reformers in the Chamber of Deputies and the Senate. However, this strategy depends on political variables that are outside the control of energy reformers. Moreover, the tide is turning against reform—elections for the period 2003–2006 of the Lower House have left the PRI with a greater number of representatives, and none of the three major political parties (PRI, PAN, and PRD) have absolute domain. The Senate will face new elections only in 2006—for this strategy, a large measure of patience and luck would be needed. A third strategy might entail striking a bargain with PRI to assure that it would not pay an electoral cost for its support of a reform bill. Despite the preferences of public opinion against energy reform, this is not a main issue in the minds of the electorate—polls show that voters care much more about employment and public security. At this writing, some efforts appear to be under way on this front. Indeed, with public opinion generally focused elsewhere, PRI may be over estimating the electoral cost, which is suggested to be minor. Although PRI may be over-estimating the cost for supporting reform, it is still unclear whether PRI would see a benefit from reform (especially if PAN, which would be most visibly identified with reform, were to reap most of the political gain if reforms were successful). Creating a winning political coalition for reform will be especially important because many unions in the electric sector have amplified their political power by forming alliances with other unions to block reform. Even if PRI’s leadership could be convinced to support reform, PRI’s traditional relationship with many unions would strain. The only group with a strong interest in mobilizing in favor of reform is industrial users. They clearly face costs from the status quo and would enjoy substantial benefits from better service and more competitive tariffs.50 To date, however, industrial consumers have been ineffective at influencing members of Congress and public opinion despite some lobbying and communication efforts. Moreover, with provisions for cogeneration and self-supply already on the books (and under lesser constitutional threat than for IPPs), the largest industrial users may actually find it cost-effective to create their own power systems and exit, largely, from the public system—as is already evident in India.

Repsol YPF-Harvard Kennedy School Fellows Research Papers

112

5.

Conclusions

Many outside analysts are surprised at how much quarrel and opposition arises from attempts to reform the power sector in Mexico. To some the rhetoric and populist claims of lost sovereignty and state autonomy—at the expense of economic efficiency and growth—are difficult to comprehend in the modern era of open economies. A close look, however, reveals that behind the rhetoric a set of powerful economic and political incentives are at work. In contrast to the late 1980s and early 1990s in which the Mexican government implemented swift market reforms in various economic sectors, today new reforms seem unlikely as many forces have emerged to shame market reforms as the cause of poverty, inequality and stagnation. For five decades the power sector has been dominated by consolidated, state-owned utilities that have entrenched themselves in the organization of the Mexican economy and the Mexican constitution—their position has, ironically, proved difficult to unravel now that reforms are implemented in the context of political fragmentation. These incumbents have also entrenched their position by touting important achievements such as 96% power access coverage—although we have shown that success on that front probably stems mainly from urbanization and economic growth. Power service appears to have improved; more importantly, the public (in general) believes that the power system is functioning well. In the last decade, niches for private investors have been created, but the broader judicial reforms have brought even these under a cloud of constitutional contention. Without any party holding a working majority in any of the chambers, further reforms have become gridlocked. The greatest challenges that remain in the sector are complex and not visible to average citizens. They include large and inefficient subsidies in tariffs, which have made it difficult for the government to contract additional debt and have skewed government spending on a wide array of other programs. We have suggested that further power sector reform is essential as the high growth in demand for electricity is narrowing the gap with available supply, and the various stopgap measures adopted to attract investment (and delay closure of old plants) are running out of steam. Even budgets to maintain old plants have been slashed.

Although it is difficult to assess, the competitiveness of the country is probably harmed—perhaps substantially—by this continued gridlock. Yet absent massive apparent difficulties—such as widespread blackouts (themselves a possibility as reserve margins recently dipped to just 1%) the needed consensus for reform remains elusive. Endnotes 1. The Mexican modern nation was built around the idea of sovereignty as a key element to keep the country united against external forces. To understand the importance of that concept it is necessary to recall that Mexico lost half of its territory in the 19th century to the United States. Since then Mexican leaders have used the discourse of nationalism and sovereignty as persuasive and unifying elements to protect Mexico’s borders and maintain the country’s independence. Although territorial sovereignty is no longer in danger, several decades of indoctrination can persist even if international conditions have changed. Today, privatization and foreign private investment are rejected because some groups perceive them as new forms of colonialism. 2. Rodríguez y Rodríguez (1994). 3. Rodríguez y Rodríguez (1994). 4. Rodríguez y Rodríguez (1994). 5. Bastarrachea S. and Aguilar, J. (1994). 6. Breceda, M (2000) 7. For the period 1974–1989 at 2001 constant dollars. 8. Efficiency measures are against CFE and LFC. First, the energy sold per worker it is only about 1.85 Gwh/worker in CFE and 1.6 Gwh/worker in LFC, compared to about 4.5 Gwh/worker in Australia. Second, the power interruption per user is 230 and 331 minutes, in CFE and LFC, respectively. In France and the United States, it is 115 and 120 minutes, respectively. Moreover, LFC is more inefficient that CFE. 9. Similar schemes were designed to hide and shift accruing debts in other areas such as airlines and PEMEX. 10. Secretaria de Energia (2002). Considering a life plant of about 30 years (for thermal plants), and Mexico’s thermal installed capacity of around 30,000 MW (excluding cogeneration and self supply) one would expect in the period from 2002 to 2011 retirements of around 8,300 MW—suggesting the need for constructing about 32,000 MW in new power plants, which is almost 15% higher than official figures. 11. As of June 2003, total debt—excluding contingent debt like social security debt, highway and sugar

Victor G. Carreón, Armando Jimenez San Vicente, and Juan Rosellón

12.

13.

14.

15. 16.

17.

18.

industries debt—was around USD $236 billion (around 36% of GDP). Moreover, the debt service in the first semester of 2003 amounted for 60% of the income and valued added taxes collected by government in the same period. We reiterate that exact costs of generation, transmission and distribution are unknown, and thus this unorthodox cost allocation makes it hard to actually sort them out. Data should be treated with caution because of the indiscriminate use of financial costs and long run marginal costs to calculate final tariffs. As a recent example we should remember that President Fox implemented some changes in the subsidy policy (reduction of subsidy for some classes of residential consumers) and must overturned it in the northern states because it faced very strong opposition from those consumers. In the years since, sober assessments of the privatization process have revealed a more subtle story. In highways and banking, privatization spawned corruption that required reassertion of control by the government; in telecommunications, the process of privatization was not accompanied by the creation of an adequate regulatory authority, with the result that competition and tariffs have delivered only a fraction of the potential benefits from privatization. A private monopoly, Telefonos de Mexico (TELMEX), has become the focal point for claims that privatization and liberalization yield changes that benefit only a few. McKinsey and Mookherjee (2003) analyze the distributive impact of privatizations in several Latin American countries, including Mexico. They find positive welfare effects that do not support the generalized bad public opinion towards privatization that exists in the region. See www.cre.gob Using evidence from other Latin American countries, McKinsey and Mookherjee (2003) show that this public perception contrast with actual empirical evidence. There is no clear pattern in prices—in half the cases, reform brings lower prices—and the impact on payrolls is not large, while the fiscal effects of reform are favorable. Non “programmable” generators are small producers that only supply power according to a previously set energy delivery schedule. Hydro generators also make available all their generation capacity, and face production constraints in the one-dayahead market. Both types of generators then have zero variable costs. Through this method, charges for transmission services for 69Kv and higher tension lines are calculated as the higher of “fixed plus variable costs” and “operation and maintenance costs”. To this amount, fixed administrative costs are added. Fixed costs are

19.

20.

21. 22.

23. 24. 25.

113

set at the long-run incremental cost of the transmission network and allocated among consumers of the current grid and consumers of the future expanded grid according to the impact that each has on congestion in the complete network. For a discussion of incentive mechanisms to expand the Mexican transmission network see Rosellon (2003). The Instituto Mexicano del Petróleo (Mexican Petroleum Institute, IMP) is in charge of research on issues related to the oil industry. In this sense it is important for the power sector because of its relationship with natural gas and fuel oil. Secretaria de Energía (2002c). Demand for gas in electricity is expected to rise rapidly (10.2% per year), but that rate will be offset by sluggish growth in self-consumption of gas in the oil sector rising. For a more complete discussion, see de Rosenzweig, F. and Femat, J. C. 2003. See McKinsey and Mookherjee (2003). However, industrial consumers in Mexico have had contradictory traditional positions. The price of natural gas is a clear example of this, and of the lack of credibility of the Fox government in engaging in truly efficient reforms (based on technical criterion and not on political pressures by industrial consumers). The famous story of the 4 by 3 PEMEX gas contracts is that case. Under such contracts the government offered a deal to sale one million btus for USD $4 during three years. This contract was offered when this price was over USD $6 to $8. Industrial consumers accepted the deal; but, when prices went below USD $4 they rejected the contract and asked for market prices.

References Bastarrachea S. and Aguilar, J. (1994). “Evolución de la Industria Eléctrica en México”, en “El Sector Eléctrico de México”. CFE y Fondo de Cultura Económica. Breceda, M. (2000). “Debate on the Reform of the Electricity Sector in Mexico”, North American Commission for Environmental Cooperation. Cambridge Energy Research Associates (2002). “Mexico’s Supreme Court Ruling: Opening Pandora’s Box?”, CERA Insight. Carreon Rodríguez, V.G. (2003). “Las Tarifas en el Sector Eléctrico Mexicano”. Boletín División de Economía. CIDE. Carreon Rodríguez, V.G. and Rosellón Díaz, J (2002a). “La Reforma del Sector Eléctrico Mexicano: Recomendaciones de Política Pública”. Gestión y Política Pública. Volumen XI, Numero 2.

114

Carreon Rodríguez, V.G. and Rosellón Díaz, J (2002b). “Incentives for Expansion of Electricity Supply and Capacity Reserves in the Mexican Electricty Sector”. Working Paper STDE-219. Centro de Investigación y Docencia Económicas, A.C. Carreon Rodríguez, V.G. and Rosellón Díaz, J (2000). “The Economic Rationale of the Structural and Regulatory Reform of the Mexican electricity Sector”. Working Paper STDE-199. Centro de Investigación y Docencia Económicas, A.C. Carreon Rodríguez, V.G. and Rosellón Díaz, J (2000). “El sector eléctrico mexicano: La necesidad de una reforma estructural”. Ejecutivos de Finanzas IMEF, Año XXXI, No. 4. de Rosenzweig, F. and Femat, J.C. (2003). “Mexican Electricity Sector: Is it moving forward?. Mimeo. International Energy Agency (2002). “México Energy Outlook”. Paris. Jiménez San Vicente, A. (2002). “The Political Economy of Tax Collection in Mexico, 1970–2000”. PhD Thesis, London School of Economics. Jiménez San Vicente, A. (1999). “The Social and Economic Cost of Structural Adjustments in Mexico”. Masters Degree Thesis, Harvard University. Joskow, P. (2000), “Deregulation and Regulatory Reform in The U.S. Electric Power sector,”, mimeo, MIT López-Calva, L.F. and Rosellon, J. (2002). “On the Potential Distributive Impact of Electricity Reform in Mexico”. Working Paper. CIDE.

Repsol YPF-Harvard Kennedy School Fellows Research Papers

Merrill Lynch & Co. (2002). “Energy Reform Mexico”. México, D. F. Mc Kinsey, D. And D. Mookherjee (2003) ‘The Distributive Impact of Privatization in Latin America: Evidence from Four Countries”, Economia, vol. 3, no. 2, Spring. Rodríguez y Rodríguez, G. (1994). “Evolución de la Industria Eléctrica en México” in El Sector Eléctrico de México, México, CFE y Fondo de Cultura Económica. Rosellón, J. (2003) “Pricing Electricty Transmission in Mexico,” Mimeo (http://www.ksg.harvard.edu /hepg/index.html) Rosellón, J. and J. Halpern (2001) “Regulatory Reform in Mexico’ Natural Gas Industry. Liberalization in the Context of a Dominant Upstream Incumbent”, Policy Research Working Paper, The World Bank, 2537. Secretaria de Energia (2002a). “Proposal for the electricity sector modernization”. México, D. F. Secretaría de Energía (2002b). “Prospectiva del Mercado de Gas Natural 2002–2011”. DGFPE. México Secretaría de Energía (2002c). “Prospectiva del Sector Eléctrico 2002–2011”. DGFPE. México. Secretaría de Hacienda y Crédito Público (2002). “Programa Nacional de Financiamiento al Desarrollo, 2002–2006”. Mexico. Wilson, R., (1999), “Market Architecture”, Stanford University, mimeo.

5

Strategic Behavior and the Pricing of Gas in Mexico Dagobert L. Brito Department of Economics and Baker Institute, Rice University Juan Rosellón Centro de Investigación y Docencia Económicas (CIDE) and Harvard University

Abstract

increased and the capacity of the pipelines connecting Mexico with the United States pipeline system has also increased. The arbitrage point may shift south to Cempoala. Thus, at this point there may some incentives for Pemex to behave strategically in supplying gas to the Mexican market. This paper is a study of the implications of such behavior and possible instruments that can be used to eliminate possible inefficiencies. The paper will look at different models of the varying complexity to study this problem. The Mexican pipeline system can be modeled as a line on the interval [0,b] with a distribution function f(n) that has mass points at the border with Texas, Los Ramones, and Cempoala.1 The pipelines also have intervals where the demand is zero. For convenience we will refer to any open interval (ni,nj ) where f(n) = 0 as a gap in the distribution. The point of arbitrage is defined as the point where gas from Ciudad Pemex in the south meets the gas from Burgos and Texas. Assume that the distance between the Texas border and Ciudad − Pemex is d , and the total demand is given by

This paper looks at various models that address strategic behavior in the supply of gas. The paper has three very strong technical results. First, the netback pricing rule leads to discontinuities in Pemex’s revenue function. Second, having Pemex pay for the gas it uses and the gas it flares increases the value of the Lagrange multiplier associated with the gas processing constraint. Third, if the gas processing constraint is binding, having Pemex pay for the gas it uses and flares does not change the short run optimal solution for the optimization problem so it will have no impact on short-run behavior. Key words: natural gas, strategic pricing, benchmark regulation, Mexico I.

Introduction to Problem

The initial economy studies of the efficiency of the netback rule were done under the assumption that the gas at Ciudad Pemex was produced as a joint product with oil and that Pemex did not behave strategically in the short run in supplying that gas to market. It was noted that in the long run there existed incentives for Pemex to shift the arbitrage point south, but at the time this did not seen like an important issue as a substantial amount of gas from Ciudad Pemex was reaching Los Ramones and there was little incentive for short run strategic behavior. Since that time, things have changed. Demand for gas in the south of Mexico has

.

(1)

Assume that Pemex supplies Q amount of gas to the market. Then distance from the arbitrage point to Ciudad Pemex is given by the solution of (2) which we will define as d(Q). The price of gas at a point n at the present time is given by: For all gas north of the arbitrage point, n > d(Q), the price of gas is the price at Houston plus the transport cost

The research reported in this paper was supported by the Comisión Reguladora de Energía through a grant to Centro de Investigación y Docencia Económicas (CIDE). The second author also acknowledges support from the Repsol YPF-Harvard Kennedy School Fellows program, and the Fundación México en Harvard.

(3)

115

Repsol YPF-Harvard Kennedy School Fellows Research Papers

116

For all gas south of the arbitrage point, n < d(Q), the price of gas is the price at the arbitrage point less the transport cost. (4) Thus the price of gas at Ciudad Pemex is given by (5) so the price of gas at Ciudad Pemex is a function of Q, pq(Q). Note that we are assuming that individual demands are not responsive to prices. The downward sloping demand curve faced by Pemex is strictly a function of the net back rule. In general, the demand curve faced by Pemex under the pricing rule is downward sloping in regions where the demand is positive and there are no mass points. This is because increasing sales move the point of arbitrage north. The price is constant in intervals of demand that correspond to mass points. This is because Pemex can sell more gas without moving the point of arbitrage. Finally, there are intervals in the pipeline where there are gaps. The demand curve faced by Pemex is discontinuous at these points; an infinitesimal shift in supply will move the point of arbitrage by a substantial amount and this leads to a discontinuity. II. Fixed supply of gas model We will initially assume that the amount of − pipeline quality gas Pemex produces is X1. This amount will be assumed to be fixed and not under the control of Pemex. Let Q be the amount of gas Pemex actually supplies to the market. We will investigate the optimal sales policy for Pemex under the assumption that it is maximizing profits for a distribution function that is uniform and a distribution function that has mass points. These assumptions are made to simplify the exposition and do not change any of the substantial results. The more general case is addressed later in the paper. Uniform Distribution Assume that the distribution of demand is uniform and f(n) = γ. Then (6)

and (7)

− − − for all Q < Q and pq = ph – cd for all Q ≥ Q For simplicity we are ignoring the cost of transport between Houston and the border. The demand and marginal revenue curve are given in Figure 1. Figure 1.

ph + cd

pˆ ph − cd

D(Q)



−t

Q

Qe

MR

In Figure 1 the quantity Qe is the point where the amount of gas is sufficient for Pemex to maximize revenue by exporting to the United States. If Q < Qe, then Pemex can maximize revenue by ˆ at a price pˆ . If Q ≥ Qe, then Pemex can supplying Q maximize revenue by exporting gas at a price ph – − cd . Note that the marginal revenue is discontinu− − − ous at Q where it goes from –t < 0 to ph – cd > 0. − Assume that Pemex has an amount of gas X1 it can supply to the market and define Qf as flared gas, Pemex would maximize (8) subject to (9) The Lagrangian is (10) and the first order conditions with respect to Q and Qf are: (11)

Dagobert L. Brito and Juan Rosellón

117

(12)

then all the gas available will be supplied to the market. Second, if

There are three cases. First if

,

(22)

(13) then λ > 0 and all the gas will be supplied to the market. Second, if

,

(14)

and Q < Qe. Then λ = 0 and Pemex will flare gas and supply an amount of gas that satisfies the condition

.

(15)

It is clear in this case that if Pemex is maximizing profits and behaves strategically, then gas will be withheld from the market and flared. − − − The third case is if X1 ≥ Qe. Then (ph – cd )X1 ≥ ˆ and Pemex would export all the gas not conpˆ Q sumed in Mexico. One instrument that would reduce the incentive to flare gas in the second case is to impose a tax on flared gas. Suppose a tax on flared gas was imposed on Pemex, then Pemex it would maximize (16)

gas will be flared. However, since t is a policy variable, it can be − chosen such that t > t and Pemex would not withhold gas from the market. (See Figure 1) The dis− does not create any problems continuity at Q because marginal revenue increases from the level –t−. Thus, if the domestic distribution of demand is continuous, then it is possible to regulate the supply of gas by imposing a linear tax on flaring gas. This result depends on the distribution of demand not having mass points and gaps. Mass Points and Gaps Now let us assume that the distribution has mass points at Cempoala, Los Ramones and Houston. The distribution is zero elsewhere. Assume the demand at Cempoala is Qc and the demand at Los Ramones is Qr. Further, assume Pemex is a price taker in the Houston market and can sell any quantity of gas at a price Ph. The demand curve for gas at Ciudad Pemex is given by pc = ph + c(dh – 2cdc )

for all

Q ≤ Qc

(23)

pr = ph + c(dh – 2cdr )

for all

Q ≤ Qc + Qr

(24)

pb = ph – cdh

for all

Q > Qc + Qr

(25)

subject to

The demand curve is illustrated in Figure 2. Figure 2.

(17) The Lagrangian is

pc

(18)

,

pr pb

and the first order conditions with respect to Q and Qf are: (19) (20) There are two possible solutions. First if

,

(21)

Qc

Qc + Qr

Note that the derivative of the demand function is undefined at Qc and Qc + Qr.

Repsol YPF-Harvard Kennedy School Fellows Research Papers

118

The revenue function π (Q) associated with this demand curve is Figure 3. π (Q) pr Q

pb Q

pc Q

Qc

Qc + Qs

Qc + Qr

Figure 4.

Qe

As seen in Figure 3, there are discontinuities in the revenue function at Q = Qc and at Q = Qc + Qr. Define Qk = Qc + Qs. The value of Qs is defined such that prQk = pcQc and Pemex is indifferent between supplying and amount Qc at a price pc and an amount Qk at a price pr. Thus, Pemex will not flare gas if the amount available is greater than Qk. If the amount of gas Pemex has available is in the interval [0,Qc) or in the interval [Qc + Qs), the maximization problem is simply

π(Q,Qf ) = piQ – tQf i =c, a

(26)

subject to

− Q + Qf = X1

(27)

Pemex will clearly sell all the gas it has, as it is a price taker and the objective function is locally concave. However, in the interval [Qc,Qc + Qs) the discontinuity in the objective function creates a problem and the problem cannot be solved using the standard optimization techniques such as the Kuhn-Tucker Theorem. Pemex has to supply more than Qc + Qs before revenues are greater than pcQc. Marginal revenue at the points of discontinuity is –y and a tax on the flaring of gas will not work. If Pemex has Q < Qc + Qs available, then it will supply the gas to market only if pcQc – tQf = pr(Qc + Qf)

the gas. It would likely be politically very difficult to implement. However, a policy that would induce Pemex not to withhold gas from the market can be implemented by defining the arbitration point by the amount of gas Pemex has the potential − to deliver. Thus the price is defined by X1 which we have assumed is not under the control of Pemex. We will now drop this assumption and assume that Pemex has some control over the amount of gas available.

(28)

or (29) As illustrated in Figure 4, the tax on flaring gas for would have to be very large if the amount of gas available is not much larger than Qc. The tax not have any relationship to the opportunity cost of

t

Qc

Qc + Qs

III. Joint Production of Gas and Oil We will now drop the assumption that the amount of gas Pemex has available to supply the market is fixed and assume that pipeline quality gas is a joint product with the production of oil, Z. We will also assume that the price of gas at Ciudad Pemex is given by a general demand function of the form pq = P(Q)

(30)

The short run production function for oil and gas is given by Z = F(X2)

(31)

X1 = βF(X2)

(32)

where Z is the oil produced, X1 is the total amount of gas produced, and X2 is the gas that is used to produce oil and gas. β is a constant that give the proportion between oil and gas. Let po be the price of oil and c be the cost of energy. Then Pemex would want to maximize the revenue from the sale of oil and gas less the cost of production.

π = P(Q)Q + poZ

(33)

subject to the production constraints

βF(X2) – Q – Qf – X2 = 0

(34)

Dagobert L. Brito and Juan Rosellón

F(X2) – Z = 0

119

(35)

0

The Lagrangian is L = P(Q)Q + poX + λ1[βF(X2) – Q – Qf – X2] + λ2[F(X2) – Z]

(36)

where λ1 is the Lagrange multiplier associated with the production of gas and λ2 is the Lagrange multiplier associated with the production of oil. The first order conditions are: (37)

λ1 ≥ 0

λ1Qf = 0

Po – λ2 ≤ 0

Z[po – λ2] = 0

(38) (39)

(43)

Strategic behavior will result in Pemex using too much gas in the production of oil. What is happening is that the shadow price oil is set equal to marginal revenue rather than price. Denote the ˆ solution by Q. A possible way to reduce the amount of gas Pemex consumes is to have Pemex pay the market price for the gas it uses. In that case Pemex would want to maximize the revenue from the sale of oil and gas less the cost of production where the cost of production includes the cost of gas used in the production of gas and oil as well as flared gas.

π = P(Q)(Q – Qf – X2) + poZ

(44)

The Lagrangian is (40)

=0

L = P(Q)(Q – Qf – X2) + poZ + λ1[βF(X2) – Q – Qf – X2] + λ2[F(X2) – Z]

If then Pemex will behave as a price taker and there are no problems. Problems can occur if

(45)

The first order conditions are: (46)

> 0 or if (47) is undefined because of a discontinuity. Let us first consider the case where

(48)

>0 . There are two possible solutions. First if gas is not flared and Qf = 0, then λ1 > 0 and all the gas will be supplied to the market. Second, if gas is flared, then Qf > 0, and λ1 = 0. Let us consider the case where λ1 > 0˜. . Then

(49) or (50)

(41) Let us first consider the case where gas is not flared. In that case Qf = 0 implies that

and (42) if λ1 < pq. This implies that Pemex will use more than the optimal amount of gas in the production of gas and oil. This is because the shadow price of gas to Pemex is the marginal revenue rather than the market price. Now suppose that gas is flared, then λ1 = 0 then Pemex treats gas as a free good and

.

(51)

ˆ The amount of distorDenote the solution by Q. tion depends on the magnitude of the term, . If, as is the current case,

,

Repsol YPF-Harvard Kennedy School Fellows Research Papers

120

the distortion will be smaller, but the shadow price ˜ of gas will be different from the market price. If (Q ˜ – X1) < 0 it is even theoretically possible that that Pemex will supply too much gas to the market. If Pemex were consuming more gas than it supplied to the market, it would be in its self-interest to lower the price of gas. Now suppose that gas is flared. Then from the ˜ Kuhn-Tucker condition given by (47) λ1 = –P(Q).

(54) (55)

. The Lagrangian is

.

(56)

Note that Qc is fixed. The first order conditions for Qf and Z are:

Then equation (51) can be written as (52) and the amount of gas supply is increased. Inasmuch as flaring gas produce carbon dioxide and is a negative environmental externality, charging Pemex for flared gas is a Pigou tax and helps the environment. Charging Pemex the market price for the gas it consumes will increase the amount supplied if it is facing a smooth demand curve. It does not lead to optimal pricing of gas except in the special case ˜ – X˜ 1) = 0. While it is necessary to underwhere (Q stand the case where the demand curve is smooth and there are no mass points to understand the economics of the problem, however, the case that is relevant at the moment is the possible shift of the arbitrage point from Los Ramones to Cempoala. This involves a shift between two mass points connected by a gap in the distribution. The derivative of the demand curve is not defined at Qc .

–pc – λ1 ≤ 0 po – λ2 ≤ 0

Qf (pc + λ1) = 0 Z[po – λ2] = 0

Continuous case Let us consider the case where the demand curve is characterized by two mass points connected by a gap in the distribution. Further let us assume that Pemex must pay the market price for the gas it uses or flares. Then Pemex must solve two problems and compare the solutions. First, it must solve the problem where it does not supply any additional gas to the pipeline and pays the penalty for flaring gas. The objective function in this case is

(58)

and since Qf and Z are strictly positive by assumption, the optimal conditions for the production of gas and oil are (59) The alternative is for Pemex not to flare gas and sell an amount of gas greater than Qa at a price pa. The maximization problem is then given by

π = pa(Q – X2) + poZ

(60)

which is also maximized subject to the production constraints

βF(X2) – Q – X2 = 0

(61)

F(X2) – Z = 0

(62)

The Lagrangian is L = pa(Q – X2) + poZ + λ1[βF(X2) – Q – X2] + λ2[F(X2) – Z].

IV. Production with Mass Points and Gaps

(57)

(63)

Note that Qf is assumed to be zero and is not included in the optimization. The first order conditions for Q and Z are: pa – λ1 ≤ 0

Qf (pa – λ1) = 0

(64)

po – λ2 ≤ 0

Z[po – λ2] = 0

(65)

Since Q and Z are strictly positive by assumption, the first-order condition with respect to X2 is (66)

(53) Define the net gas produced, X3, as which it maximizes subject to the production constraints:

X3 = βF(X2) – X2

(67)

Equations (62) and (67) can be solved for Z(X3).

Dagobert L. Brito and Juan Rosellón

121

Pemex’s profit can be written as a function of X3. π (X3), for X < Qc is given by

π (X3) = poZ(X3) + pcX3

(68)

and for X ≥ Qc Pemex’s profit is either

π1(X3) = poZ(X3) + pcQc

(69)

or

π2(X3) = poZ(X3) + paX3

(70)

So for X3 < Qc, the price of gas is based on the price at Cempoala, all gas is sold so Q = X3. For X3 ≥ Qc there are two possibilities. Pemex can either start flaring gas so the income for all X3 ≥ Qc comes from the sale of oil. Alternatively, Pemex can continue to sell gas, the price is then based on the price at Los Ramones and the profit function is discontinuous at Qc. The profit function is illustrated in Figure 5 under the simplifying assumption that Z is proportional to X3.

can be transmitted in a pipeline. This requires processing and gathering capacity. There is a question whether Pemex has sufficient capacity. The fact that over 200 million cubic feet of gas are currently flared suggests that this is a problem. Let us − assume that Pemex can process X amount of gas for sale in the pipeline. Then Pemex would want to maximize the revenue from the sale of oil and gas less the cost of production.

π = P(Q)Q + poZ

(71)

subject to the production constraints

βF(X2) – Q – Qf – X2 = 0

(72)

F(X2) – Z = 0

(73)

and the gas processing constraint − Q″X

(74)

− where X is the constraint on processing capacity. The Lagrangian is

Figure 5. π2

(75)

π1

where λ3 is the Lagrange multiplier associated with the gas processing constraint. The first order conditions are: Qc

QS

(76)

X3

If Pemex supplies more than Qc gas to the market the point of arbitrage will move from Cempoala to Los Ramones and the price of gas will drop from pc to pa. π1(Q) gives Pemex’s profit if it does not flare gas and accept the drop in price. π2(Q) gives Pemex’s profit if it flares gas and pays the penalty. For quantities of gas less that Qs it is optimal for Pemex to flare gas.

po – λ2 = 0

(77) (78)

λ1 ≥ 0

λ1Qf = 0

− X–Q≥0

− λ3(X – Q) = 0

(79) (80) (81)

V. Gas Processing Constraint Continuous case—Gas is free to Pemex We have studied the problem under the assumption that gas to be sold to the pipeline without processing. This is not a realistic assumption since it is necessary to remove butane, propane and other natural gas liquids from the natural gas before it

Denote the solution of this problem by a ~. If − ˜ ″X Q , then there is no change from the previous analysis, so assume that the constraint binds. Then − ˜ =X Q and (82) .

Repsol YPF-Harvard Kennedy School Fellows Research Papers

122

There are two cases. The first case is if gas is ˜ f > 0, and λ˜ 1 = 0. flared, then Q ˜f The second case is if gas is not flared. Then Q ˜ = 0 and λ1 > 0. If gas is flared and λ1 = 0 then equation (81) can be written as

(90) (91)

(92)

(83) ˜ 2 is determined by equation The value of X (83). Further, if gas is being flared λ˜ 1 = 0 and

.

Denote the solutions by a ^. Let us first consider ˆ f > 0 then from the case where gas is flared and Q − ˆ (91) λ1 = –P(X) so

(84)

The price of gas is imputed to the gas processing constraint. If gas is not flared, then all the gas that is not − ˜ ˜2= X – X . The values of X, sold will be injected, so X and λ˜ 1 are determined by the solution of

.

(93)

The term

(85)

(86) This solution will be used to compare the impact of charging Pemex of the gas it uses. Continuous case—Pemex pays for Gas Now let us consider what would happen if Pemex pays the market price for the gas it uses. Pemex would want to maximize the revenue from the sale of oil and gas less the cost of production where the cost of production includes the cost of gas used in the production of gas and oil as well as flared gas. (87) The Lagrangian is

so λˆ 3 > λ˜ 3 and the shadow price of gas processing facilities is greater in the case where Pemex must pay for the gas it uses. This result suggests that requiring Pemex to pay for the gas it uses will increase its incentives to invest in gas processing capacity. − If gas is flared, then since, λ2 = –P(X ) and equation (92) be written as (94) Note that equations (83) and (94) are identical ˜ 2. This gives the somewhat surprising so Xˆ 2 = X result that if Pemex is flaring gas, having Pemex pay for the gas it uses in production and that it flares does not change any of the short run economic decisions if the gas processing constraint is binding. Now let us consider the case when Pemex is not flaring gas. The first order condition for the sale of gas results in

(88) If the constraint on the processing of gas for sale in the pipeline is binding, the first order conditions are: (89)

(95) so λ1 + λ3 is greater in the case where Pemex must pay for the gas it uses if Pemex is not flaring gas. The amount of gas that is used in production − ˆ 2 =X ˆ –X is X . Let p be the price of gas to Pemex and

Dagobert L. Brito and Juan Rosellón

123

ˆ and λˆ 1 are define q = (λ1β + po). The values of X, determined by the solution of (96)

(97)

in the production of gas. Therefore making Pemex pay for the gas it uses in production decreases the value of λ1. Since the sum λ1 + λ3 is larger, it must be that the value of λ3 is larger. The incentive for Pemex to invest in gas processing capacity is increased. Mass Points

If we differentiate 95) and 96 with respect to p, we get

(98)

solving for

we get

Now let us consider the case where the distribution is characterized by mass points. The optimization is the same as with a continuous distribution, except that the price of gas is fixed. This is just a special case where P(Q) = pk. However, since this is the case that is most similar to the current situation, it merits a complete treatment. If it is optimal for Pemex to withhold processed gas from market, then the constraint is not binding and the analysis is as in section IV. Let us assume that the constraint is binding. Then the price Pemex would receive for the gas is given by the price at the mass point, Pk and again Pemex would want to maximize the revenue from the sale of oil and gas less the cost of production. The optimization is given by maximizing (102)

(99) and short run production is independent of the price of gas. Solving for

subject to the production constraints (103) (104) and the gas processing constraint (105) The Lagrangian is

(100)

. and

(106)

The first order conditions are: (101)

(107)

If Pemex is not flaring gas, it would never be optimal to produce at a point where

(108) (109)

, since this implies that more gas is being produced than the amount of gas being used

(110)

Repsol YPF-Harvard Kennedy School Fellows Research Papers

124

(111) Again, there are two possible solutions. First if gas is not flared and Qf = 0, then λ1 > 0. Second, if gas is flared, then Qf > 0, and λ1 = 0. In both cases λ1 reflects the value of gas and λ3 the value of the gas processing constraint. Clearly, if gas is being flared, the price of gas is all imputed to the gas processing constraint. The first order condition for the use of gas in the production of gas and oil is

The value of λ3 will increase so requiring Pemex to pay for the gas it uses will increase its incentives to invest in gas processing capacity. The first order condition for the use of gas in the production of gas and oil is (121) If gas is flared, then from the Kuhn-Tucker condition given by (118), λ2 = –pk and (122)

(112) We know that if gas is not flared, then all the gas that is not sold must be injected so the value of λ1 − is determined by X2 = X – X and equations (104) and (112). If gas is flared, the λ2 = 0 and

Thus having Pemex pay for the gas it uses will be equivalent to a lump sum tax in that it does not change any of the economic decisions if the gas processing constraint is binding and gas is flared. As we would expect, the results do not change from the results in the more general case.

(113) VI. Conclusions and Recommendations Now let us consider what would happen if Pemex pays the market price for the gas it uses. Pemex would want to maximize the revenue from the sale of oil and gas less the cost of production. (114) The Lagrangian is

(115) Assume the constraint on the processing of gas for sale in the pipeline is binding. The first order conditions are: (116) (117) (118)

The possibility that Pemex will behave strategically in its short run supply of gas creates some interesting technical and economic problems. The density function that characterizes demand has mass points and gaps. This results in a profit function that is not concave and standard economic analysis must be used with care. This paper looks at various models that address strategic behavior in the supply of gas. The models increase and complexity and understanding them is useful in developing a wellinformed intuition about the problem. The model that most closely resembles the current situation in Mexico is one where: 1. 2. 3. 4.

(119) Let us first consider the case where gas is flared. In that case Qf = 0 implies that λ1 = –pk so (120)

The distribution is characterized by mass points; Pemex uses gas in the production of gas and oil; The constraint in the processing of gas to pipeline quality is binding; Gas is being flared.

That model has three very strong technical results. First, the netback pricing rules leads to discontinuities in Pemex’s revenue function. Second, having Pemex pay for the gas it uses and the gas it flares increases the value of the Lagrange multiplier associated with the gas processing constraint.

Dagobert L. Brito and Juan Rosellón

Third, if the gas processing constraint is binding, having Pemex pay for the gas it uses and flares does not change the short run optimal solution for the optimization problem so it will have no impact on short behavior The first recommendation that follows from this analysis is that the arbitrage point be fixed by the amount of gas Pemex has the potential to supply in the absence of processing and gathering constraints. This policy is not strictly optimal in that it violates the Little-Mirrlees Rule, but the distortion is not large. The cost of distortion is less than the cost of moving the necessary gas from between the two arbitrage points in question. The reasoning behind this recommendation is that the discontinuities in Pemex’s revenue function create non-convexities in the optimization problem that cannot be addressed by policies that work at the margin. This is more of a political than an economic problem. In the absence of intuitional constraints on investment by Pemex, it would be economically efficient to invest in processing capacity so as not to flare gas and supply this gas to market. Given that there are intuitional constraints that restrict investment, the question is whether economic and political benefits of supplying gas to central Mexico at the price that would prevail in the absence of these constraints outweighs the cost of transporting gas between the two arbitration points. The second recommendation that follows from this analysis is that Pemex be charged for the gas it uses in production and the gas it flares. In the short run, this policy is neutral in that it does not distort behavior. It the long run, it creates incentives for Pemex to invest in gas processing capacity. The third recommendation that is suggested by this study is that investment in gas processing and pipeline be in a separate account from other Pemex investment. Pemex is under a strict capital constraint. The reasons for this constraint are beyond the scope of this paper. However, capacity constraints in gas processing appear to be a serious problem; Mexico is flaring a substantial amount of gas while it is importing gas from the United States. Pipeline capacity is not a binding constraint at the moment. However, demand is growing. If there is not sufficient investment in pipelines, capacity constraints may become binding.

125

Finally, a study should be done of the demand elasticity for gas in the production of gas and oil. At the moment Pemex appears to be treating gas as a free good. The question is how much gas would be available if Pemex had to pay for the gas it uses and the gas processing constraint was not binding. This is a question for petroleum and reservoir engineers. Endnotes 1. See Brito and Rosellon (2002) 2. We will assume that the demand of individuals for gas is not a function of price. This assumption is made for simplicity and does not change any of the results. We are also assuming that the pipeline system is not a binding constraint in Pemex supplying gas to market. This is a valid assumption at the moment, but it should be noted that the feasibility of the netback rule to serve as a pricing mechanism for gas in Mexico depend on gas being able to move freely to equilibrate markets.

References Adelman, M.A, 1963, The Supply and Price of Natural Gas, (B. Blackwell, Oxford). Brito, D. L., W. L. Littlejohn and J. Rosellón, 2000, “Pricing Liquid Petroleum Gas in Mexico, Southern Economic Journal, 66 (3), 742–753. Brito, D. L. and J. Rosellón, 2002, “Pricing Natural Gas in Mexico: an Application of the Little-Mirrlees Rule,” The Energy Journal, 23 (3), 81–93. Comisión Reguladora de Energía, 1996, “Directiva sobre la Determinacion de Precios y Tarifas para las Actividades Reguladdas en materia de Gas Natural,” MEXICO. (WEB SITE: http://www.cre.gob.mx) Little, I. M. D. and J.A. Mirrlees, 1968, Manual of Industrial Project Analysis in Developing Countries, (Development Centre of the Organization for Economic Co-Operation and Develpment, Paris) Pemex, 1998, “Indicadores Petroleros y Anuario Estadístico”. Rosellón, J. and J. Halpern, 2001, “Regulatory Reform in Mexico’s Natural Gas Industry: Liberalization in Context of Dominant Upstream Incumbent,” Policy Research Working Paper 2537, The World Bank. Secretaría de Energía, 1998, “Prospectiva del Mercado de Gas Natural, 1998–2007.” (WEB SITE:http:// www.energia.gob.mx/frame4.html)

6

Price Regulation in a Vertically Integrated Natural Gas Industry: The Case of Mexico Dagobert L. Brito Department of Economics and Baker Institute, Rice University Juan Rosellón Centro de Investigación y Docencia Económicas (CIDE) and Harvard University

Abstract

public ownership of energy resources in Mexico is politically critical. Foreign firms originally owned the oil industry and its nationalization was considered as a symbol of Mexican sovereignty. Nowadays, privatization of Pemex would be unthinkable in political terms. Technical and institutional difficulties are thus important problems in regulating the Mexican natural gas price. The Mexican regulator, Comisión Reguladora de Energía (CRE), solved the problem by using a natural gas price benchmark in Southeast Texas. The natural gas price at Ciudad Pemex in Southeast Mexico (where 80% of total natural gas is produced as a byproduct of oil extraction) is linked to the price at the Houston Ship Channel hub through a netback formula. The price of gas in Ciudad Pemex is equal to the price at Houston plus transport costs from Houston to the arbitrage point, (the arbitrage point is currently located at Los Ramones, in northeast Mexico)3 minus transport costs from the arbitrage point to Ciudad Pemex (see Figure 1).4 This pricing regulatory formula is an implementation of the Little-Mirrlees method, which proposes the use of world prices for pricing traded goods.5 Thus the price of gas in Houston is a measure of the opportunity cost to Mexico of consuming the gas rather than exporting it to the United States.6 The netback rule also implies that the Mexican gas price remains insensitive to variations in demand for gas in Mexico, and that consumers are facing a flat supply curve. The amount of gas imported or exported works as an equilibrating factor. The netback rule was published by the CRE in 1996.7 It has been debated during several North American price spikes such as the one in December of 2000.8 The price of gas in Houston rose from around $2.00 per MMBTU in January 2000 to almost $10.00 per MMBTU by January 2001. Many Mexican firms had not hedged and as a result

The Comisión Reguladora de Energía of Mexico has implemented a netback rule for linking the Mexican natural gas price to the Texas natural gas benchmark price in an industry structure characterized by a vertically integrated state-owned monopoly. This paper shows that in an open economy where agents can choose between gas and alternative fuels, and where the density function describing the distribution of agents along the pipeline can have mass points, the netback rule is Pareto optimal. 1.

Introduction

The Mexican oil and gas market is a peculiar one. The national State firm Petróleos Mexicanos (Pemex) has a monopoly in production of oil, gas and natural gas liquids, and it is vertically integrated in transportation and marketing of these products.1 Furthermore, gas is a joint product of oil, a fact that makes impossible the allocation of production costs to natural gas alone.2 In addition,

We would like to thank William Laney Littlejohn for his suggestions, as well as two anonymous referees for insightful comments. All remaining errors are our own. The research reported in this paper was supported by grants for the Baker Institute for Public Policy at Rice University, and the Comisión Reguladora de Energía to CIDE. Juan Rosellón carried out this research while he was a visiting Senior Fellow at the Kennedy School of Government of Harvard University and acknowledges the support from the Repsol YPF-Harvard Kennedy School Fellows Program, the Fundacion Mexico en Harvard, and the Harvard Electricity Policy Group. 127

Repsol YPF-Harvard Kennedy School Fellows Research Papers

128

found themselves in serious trouble. Plants were being forced to close. There was strong pressure on the CRE to drop the Houston benchmark in pricing gas. Pemex rescued the firms in trouble by offering a $4.00 per MMBTU three year take or pay contracts. The netback rule based on the Houston price remained. In this paper, we study the economics of the netback rule in a general model. It will be shown that under a very general set of assumptions, the netback rule is Pareto efficient. 2.

The Mexican Natural Gas Transportation System

The Mexican pipeline system is 9,043 kilometer long and reaches most of the main industrial centers in the South, Center and Northeast of the country (see Figure 1). Of the total natural gas transported in the system during 2000, 69% was associated gas, 24% was non-associated gas, and 7% was imported gas. As of 2000, 1795 kilometer of new private pipelines had been added to the existing transportation system and the capacity had increased to 7.4 bcfd. Additionally, the

pipeline linkage from the Reynosa-Burgos area to the Texas market is under expansion.9 The pipeline system in Figure 1 can be considered as a single pipeline connecting production in the South with production in the North. Ciudad Pemex is located at the south end of the system and produces most of natural gas (more than 80% of total gas) as associated gas (that is as a by product of oil extraction)10, while Reynosa-Burgos is located at the north end. Burgos produces 17.3% of total natural gas production and is a link to the Texas pipeline system. Two branches complement the pipeline system. One branch connects Ciudad Juárez (a place where gas is imported) and Los Ramones (the junction of the Southeast, Northwest and Northeast pipelines). The other branch of the pipeline connects the cities in the center of the country (including Mexico City) with the main network at Cempoala. The analysis of this pipeline network can be simplified exploiting the technical and institutional properties of the system. As shown in Brito and Rosellón (2002), the problem of pricing gas can be analyzed as a single pipeline connecting Burgos with Ciudad Pemex. The connections at

Figure 1.

Source: PEMEX

Dagobert L. Brito and Juan Rosellón

129

Los Ramones and Cempoala are mass points in the distribution of demand. The netback rule follows from the solution of the problem of a regulator that maximizes welfare subject to resource constraints in the pipeline network. The optimal prices of natural gas are the shadow prices in the optimization associated with the production of natural gas in Mexico. In particular, the neback rule is the shadow price of the scarcity constraint at Ciudad Pemex. This rule follows from the condition that Pemex should be indifferent between the sale or purchase of gas in Houston, and the sale of gas at any point in Mexico. When this condition is not met, the construction of a welfare-improving allocation of gas could be feasible. It would be enough to shift the allocation of gas from activities whose marginal benefit is less than the price of gas to activities whose marginal benefit is higher than the price of gas. We next study the netback rule in a general model. We will assume that individuals are located along a pipeline. They can spend their income on goods, an alternate fuel or gas. The price of gas is given by a nonlinear price schedule that is a function of location and the quantity of gas purchased. We show that under such conditions, the general optimal price of gas is the netback rule. A general optimal nonlinear price schedule for gas is a very powerful instrument in that it permits location specific taxation. However, the netback rule is also optimal without location specific charges if there are no income effects. Further, the netback rule is always Pareto efficient. The netback rule is the optimal way of pricing gas unless there are redistributional goals that must be met using this instrument and location specific charges are ruled out. 3.

Model

Assume that individuals are located on the inter−] with a general density function f(s) ≥ 0 val [0,n −. This denwhich represents a pipeline of length n sity function allows the possibility of intervals with no demand as well as mass points. A special case is where demand is on a set of discrete points along the pipeline. The typical individual located at point s has a utility function of the form. (1) where x is a bundle of goods, y is the consumption of natural gas and z is the consumption of a substi-

tute fuel for natural gas.11 Each individual is assumed to furnish one unit of labor at a wage w(s). Individuals maximize utility subject to the constraint (2) where t(y,s) is the price schedule for gas and q(s) is the market determined price of the substitute fuel. The price of x is normalized to one. The planner can redistribute income by location as a function of the consumption of gas, so

is a possible control instrument. Define

Individuals differ in their location and income, so using the envelope theorem it follows that the utility of individuals along the pipeline is given by the solution of the following differential equation (see the Appendix for the derivation of differential equation (3) and of most of the further mathematical derivations): (3) Let v(s) be the solution of the differential equation, then we can use the relationship v = u(x, y, z) to write (4) The variable v(s) is a state variable and the variables y(s) and z(s) are control variables. Define the aggregate amount x by X, of y by Y and z by Z. The good X1 is consumed and X2 is exported. For gas, Y0 is produced domestically, Y1 is imported at a price p, Y2 is used to produce X, Y3 is imported gas consumed by individuals and Y4 is domestic gas consumed by individuals. For the substitute fuel, Z1 is imported at a price q−, Z2 is used to produce X and Z3 is consumed by individuals. We will assume that the good X is produced by a technology that uses energy (5) where F(Y2, Z2) is a well behaved strictly concave function, and Y2 and Z2 is the energy used to produce the good. Production of good x is assumed to

Repsol YPF-Harvard Kennedy School Fellows Research Papers

130

occur at n = 0 and it is assumed that good x can be transported without charge (see figure 2).13 Production of natural gas is a joint product of oil extraction (associated gas) and is assumed −, while natural gas is imported at to occur at n = n

The aggregate constraints are: (13) (14) (15)

Figure 2

(16)

λ3

p

0



Constraints (13) through (15) make supply and demand equal in the markets for the private good, natural gas and the substitute fuel, respectively. Equation (16) represents the balance between the value of exports and imports. The constraints given by equations (8) through (12) can be converted to differential equations

n

(17)

n = 0. It is also assumed that gas can be transported at cost c. Define nˆ as the point of arbitrage. The cost of moving imported gas to point of arbitrage is nc ˆ and the cost of moving domestic gas to − – n)c. point of arbitrage is (n ˆ Define

(18) for n < n, ˆ (19) for n < n, ˆ 14

(6)

where β (s) is the welfare weight of individuals located at point s. Now let us consider a planner trying to maximize welfare (7) where G are public expenditures. The maximization is subject to the constraints that (8)

(20) (21) where A is aggregate redistribution. The aggregate constraints given by (13) to (16) define the transversality conditions for the differential equations. The planner’s problem can be written as maximizing the sum of aggregate welfare V and public expenditures G subject to the aggregate constraints (13) through (16): (22)

(9)

(10)

(11)

The variables δi ,i = 1,3 are the Lagrange multipliers associated with the aggregate constraints. This is an optimal control problem and the maximization with respect to the aggregate variables provides the transversality conditions. The Hamiltonian is:

Equations (8) through (11) represent the aggregate demand for goods and energy. If we assume that net redistribution is zero, then (12)

for n < n, ˆ and

Dagobert L. Brito and Juan Rosellón

131

Figure 3 f (n )

n1

ˆ n

n2

Proposition 1 The optimal non-linear price schedule for natural gas is the netback rule. Proof (see Appendix) for n > n, ˆ where λi ,i = 1,5, are the costate variables associated with (17) through (21) respectively, and θ is the costate variable associated with (3). The control variables are y, z and α. We assume that the point nˆ is in an interval (n1,n2) such that f(n) is strictly positive, continuous and there are no mass points for n in (n1,n2) (see Figure 3). Then it follows from the continuity of the Hamiltonian that

Proposition 2 If all individuals have equal weight in the aggregate utility function and there are no income effects, the optimal non-linear price schedule for natural gas is the netback rule and it is independent of redistribution. Proof (see Appendix) Proposition The netback rule for pricing natural gas is Pareto optimal. Proof (see Appendix)

(23)

(24)

These three propositions establish several optimality characteristics of the netback rule. Proposition 1 states that when in fact there exist no constraints in the flow of gas, gas will flow in equilibrium so that the marginal rate of substitution between the private good and gas is equal to the netback rule. Proposition 2 states that when there are no income effects15 the price of gas will be equal to the netback rule. Since all individuals have the same welfare weights and there are no income effects the aggregate welfare function will not be sensitive to redistribution. Proposition 3 states that given any allocation that results from the netback rule, then we can find a set of weights β (n) in the welfare function so that θ(n) = 0 for all n. Recall that

Since λ3 is the shadow price of equation (19) or, in other words, the shadow price of domestic gas at the point of production, this is in fact the netback rule.

is the contribution to the aggregate welfare of increasing the utility of an individual located at n.

Equation (23) links the shadow price of imported gas with the shadow price of domestic gas given the assumption that there are no mass points. Now suppose that nˆ is a mass point. Then imported gas and domestic gas are both consumed at nˆ and it follows from the first order conditions with respect to y that

which is identical to equation (23) and thus yields the netback rule. Intuitively the result follows from the law of one price. If imported gas and domestic gas are being sold at the point represent by n, ˆ they must have the same price. The value of λ3 is derived from (23) and results in

Repsol YPF-Harvard Kennedy School Fellows Research Papers

132

So, when θ (n) = 0 for all n it gives a first best. Therefore, the given allocation maximizes welfare and any deviation will not be optimal so that the allocation corresponds to the Pareto optimum for the set of weights. 4.

Vertical Integration and the Netback Rule

PEMEX is vertically integrated among production, transportation and marketing activities in the natural gas industry. This potentially permits PEMEX to carry out several strategies so as to preserve its vertical monopoly as well as to control pipeline capacity so as to circumvent the netback formula. When there is not enough capacity, the gas movement would not clear the markets, and Pemex could capture the rents associated with the capacity restriction. Vertical integration between transportation and marketing lets PEMEX’ transportation subsidiary (“PMX-Trans”) offer a preferential treatment to its marketing subsidiary (“PMX-Com”) 16 regarding access to capacity and transportation tariffs. This allows PMX-Com preserve its monopoly in gas marketing17 and, more importantly, avoid the netback rule. More specifically, PMXTrans monopolistically offers access to its pipeline capacity. This service is supplied to PMX-Com as well as to potential private competitors in a supposedly contestable market. However, if open access clauses are not carefully enforced by the regulator, PEMEX might allocate most of its capacity and gas sales to PMX-Com, and argue lack of capacity to meet gas sales to other consumers at a regulated price. PMX-Com could then sell gas at a price above the netback price. Even more, under constraints of capacity, Pemex could use cross subsidization from gas production and transportation activities to marketing activities so as to be able to further increase the final gas price. As an example on how a capacity pipeline constraint would be reflected in the final gas price, let us analyze the case of a capacity constraint in the import pipeline capacity. In such a case, it − would be necessary to add a constraint, Y1, for natural gas imports to optimization problem (22). This can be rewritten as: (25) where δ4 is the Lagrange multiplier for the import constraint and, when it is binding, it will be

reflected in the natural gas price. This multiplier reflects rents to the control of access to the pipeline and exemplifies the more general proposition that, when there are pipeline capacity bottlenecks, the Lagrange multiplier associated with the pipeline capacity constraint reflects such rents. In such a case, the netback rule will not work since gas flows cannot equilibrate the market, and the gas price would have to be adjusted so that rents accrue to the scarce factor: pipeline capacity. PEMEX’ vertical disintegration would contribute to a more competitive allocation of pipeline capacity, and hence to a better performance of the netback price regulation. However, institutional and political constraints preclude the vertical divestiture of Pemex. Alternatively, as argued by Brito and Rosellon (2003a), Pemex could equivalently be only allowed to establish gas spot or future contracts as long as it does not carry out any discount in the regulated production prices and transportation tariffs inside Mexico, even in a non discriminatory way. Pemex would then concentrate its marketing activities in the Houston market exclusively.18 Likewise, the CRE should enforce open access regulation as well as the monitoring of PEMEX’ investment in pipeline capacity. 5.

Other Policy Implications of the Netback Rule

We have shown that the netback rule results from a well structured welfare maximization general model, and hence has several desirable efficiency properties. However, policy makers should be aware that the netback rule also defines a peculiar structure of incentives. Small changes in the distribution of gas might result in big changes in its price. Pemex could then divert production from the regulated market (or simply reduce its production) to bring south the arbitrage point, and then cause an increase of the domestic natural gas price twice greater than the change in the marginal cost of transport.19 This does not change the efficiency properties of the netback rule, but it does change the allocation of economic rents. Policy makers should consider this when implementing location changes of the arbitrage point.20 Other policy measures that might have an effect on the netback rule include reductions in import tariffs, export restrictions and investments

Dagobert L. Brito and Juan Rosellón

in production facilities. A reduction in import tariffs does not imply an increase in natural gas imports and has a small effect on the domestic gas price (proportional to the tariff reduction). A pipeline capacity restriction that hinders gas exports is reflected in the domestic gas price through the shadow price of such a restriction. Additionally, the development of new gas production sites close to the arbitrage point—and possibly not close to consumption areas—is socially optimal.21 One more discussion on the netback rule is with respect to the benchmark price. The use of the South Texas price introduces to the Mexican market the competition of the US natural gas market but it also introduces the volatility of that market, as has been experienced in several cold winters. Then the netback rule must also be implemented with hedging mechanisms that smooth out such price distortions. Mexico has both followed decentralized and centralized policy measures regarding hedging procedures. However, Mexican industrial consumers have kept pressing the government so as to get gas price reductions.22 They would like to substitute the current Texas price reference with a (lower) “Mexico price” that takes an average of the international natural gas prices. Would such a policy be reasonable? The structure of the netback formula represents the value of domestic natural gas in a welfare maximization model. Likewise, the opportunity cost of the Mexican natural gas is given by the Texas gas price and not by other international markets (say in South America or Europe) because pipeline natural gas markets are local or regional. Natural gas is methane, it becomes liquid at very low temperatures (around –275ºF) and liquid natural gas (LNG) an only be transported by sea at a high cost.23 Therefore pipeline natural gas is not a typical commodity that can use any international price as a benchmark to determine its opportunity cost. However, there exists the theoretical possibility of changing the parameters of the formula and, consequently, modify the distribution of the rents associated with domestic natural gas sales. Therefore, a change of the benchmark price from Texas to, say, Venezuela would transfer rents from Pemex (and, hence, from the public budget) to natural gas consumers. This price-distortion policy would evidently be an inferior transfer policy to a lump-sum subsidy to such consumers.24

133

Nonetheless, the efficiency of using the Houston price as a benchmark relies on the assumption of competitive conditions in the Texas natural gas market. If such assumption does not hold, then the use of an alternative price might be justified. The recent increasing trend in the gas price and the expected future increase of LNG imports to the North American market might give good reason for the use of an alternative benchmark price. If it can be proved that pipeline network capacity restrictions in Texas preclude the arbitrage between the LNG import price and the Houston natural gas price, then the use of a net present lower benchmark price (that considers the possibly lower future gas price resulting from the increased entrance of LNG to the Texas market) might be justified. 6.

Concluding Remark

This paper studies the optimality of the netback rule based on the Houston Ship Channel price to price natural gas in Mexico that has been implemented by Comisión Reguladora de Energía in an open economy where agents can chose between gas and alternative fuels, and where the density function describing the distribution of agents along the pipeline can have mass points. The paper shows that the netback rule is Pareto optimal. However, a challenging policy issue for the adequate performance of the netback rule is the vertical disintegration of Pemex’ transportation and marketing activities so as to assure the existence of enough capacity on the Pemex’ pipeline network. As discussed, when there is not enough capacity the gas flows will not clear the markets, the netback rule will not work, and Pemex will capture the rents associated with the capacity bottlenecks. 7.

Appendix

8.1 First Order Conditions for Maximization of (1) subject to (2) The Lagrangian for the individual’s maximization is: (26) Then if we assume that there are no corner solutions, the first order conditions are:25

Repsol YPF-Harvard Kennedy School Fellows Research Papers

134

(27)

(35) The first order condition with respect to α is

(28)

(36) (29) Using the first order condition for x, the differential equation:

Assume that the point nˆ is in an interval (n1,n2) such that f(n) is strictly positive, continuous and there are no mass points for n in (n1,n2). Then it follows from the continuity of the Hamiltonian that

(30) can be rewritten as:

(37) 8.2 First Order Conditions for Maximization of (22) with respect to aggregate constraints (13) through (16)

so

To simplify notation we will not use the arguments of the variables. The Hamiltonian is Since X1, Y3, Y4 and Z3 are not in the Hamiltonian, (31) for n < nˆ and

The first order conditions for the aggregate variables in (22) are (32)

for n > n, ˆ where λi ,i = 1,5, are the costate variable associated with (17) through (21) respectively and θ is the costate variable associated with (3). The control variables are y, z and α. The first order conditions with respect to y are (33)

(38)

(39) (40) (41) (42)

for n < nˆ and

These first order conditions are the transversality conditions for G, X1, Y3 and Z3: (34)

(43) (44)

for n < n. ˆ The first order condition with respect to z is

(45)

Dagobert L. Brito and Juan Rosellón

135

−) are free end points, θ(0) = Since v(0) and v(n −) = 0. The value of λ is derived from (23) and θ(n 3 results in

8.3 Proof of Propositions (1) through (3) Proposition 1 The optimal non-linear price schedule for natural gas is the netback rule. Proof: −) are free endpoints so θ(0) = θ(n −) v(0) and v(n − = 0 at 0 and n. Since λ5 is constant, λ5 is 0 for all n and if α ↑ 0 then θ(n) = 0. Now suppose α = 0 and θ(n) ↑ 0, then since the Hamiltonian is linear in α, this would violate the necessary conditions of the Maximum Principle that the Hamiltonian be maximized with respect to the control variables and thus θ (n) = 0 for all n. The first order condition given by (33) can be written as (46) so that the marginal rate of substitution equals the netback price, which is the desired result. Proposition 2 If there are no income effects, the optimal non-linear price schedule for natural gas is the netback rule and it is independent of redistribution. Proof: A sufficient condition for the result to hold is that in the first order condition given by (33), the term

so that the condition holds. Denote derivatives by subscripts, then

(47) This is the income effect term from Slutsky’s equation. Further, since all individuals have the same welfare weights and there are no income

effects the aggregate welfare function is not sensitive to redistribution. Proposition 3 The netback rule for pricing natural gas is Pareto optimal. Proof: A sufficient condition for the result to hold is that the welfare weights, β(n) be such that the term −]. Since ν(0) is a free endθ(n) = 0 for all n in [0,n point, θ(0) = 0 so a sufficient condition for the term −] is that θ(n) = 0 for all n in [0,n

(48) so (49) is a sufficient condition for θ(n) = 0 for all n and the welfare weights are such that no redistribution is optimal. This implies that any redistribution cannot be Pareto improving and thus the solution is Pareto optimal. Endnotes 1. A partial reform in 1995 permitted new private transportation and distribution projects, but maintained Pemex’ monopoly in production. This reform has only been successful in attracting private investment to distribution systems (see Rosellón and Halpern, 2001). 2. See Adelman (1963) and Brito, et. al. (2000). The main substitutes for natural gas in Mexico are liquid petroleum gas (LPG) for residential consumption, and fuel oil and diesel for industrial consumption and power generation. Since the marginal cost of production for these products cannot be isolated, their prices are usually determined through international benchmarks that are not correlated with the natural gas price (see Brito et al, 2000). Thus, the price of natural gas cannot be determined through the price of its substitutes. However, using the price in Houston implicitly makes the price of Mexican natural gas reflect the price of competitive sources of energy. 3. The arbitrage point is the place where northern and southern gas flows meet, and where northern and southern gas prices coincide. The location of the arbitrage point might change according to gas flows. For example, if more gas is imported it will tend to move southwards displacing the arbitrage

Repsol YPF-Harvard Kennedy School Fellows Research Papers

136

4. 5. 6. 7.

8. 9. 10.

11.

12.

13. 14. 15. 16.

17.

18.

point from its current location at Los Ramones to the next demand point, Cempoala, which is located 700 kilometers to the south. See Comisión Reguladora de Energía (1996), section 4. See Little and Mirrlees (1968) p. 92. See Brito and Rosellón (2002). Pemex had been using a very similar rule base on another Texas benchmark (Tetco and Valero). See Rosellón and Halpern (2001). See Arteaga, and Flores (2002) and Arteaga and Flores, (2003). The export-import capacity at the Reynosa sector for 2002–2003 was 534 MMcfd. This means that it is impossible to allocate costs of production (or extraction) to most of natural gas produced in Mexico (see Adelman, 1963). As with natural gas, these main substitute goods (LPG, diesel and fuel oil) are also byproducts of oil extraction. Therefore, the supply of these products are conditioned by Pemex’ role as a source of revenue for the Mexican government (see Rosellón and Halpern, 2001). Note that we are implicitly assuming that ∂u/∂x is positive (a standard assumption for unsatiated goods) in order to invert ν = u(x) to get x = x(ν). As we explain later in the paper, λ3 is the shadow price of domestic natural gas at the point of production. If f(nˆ ) is a mass point in the distribution function the demand for domestic gas will be such that Y0 ≤ Y4. So that the second term of (33) is equal to zero. PEMEX’ natural gas transportation functions are carried out by the “Pipeline Area” of PEMEX’ subsidiary Pemex Gas y Petroquímica Básica (PGPB). PEMEX’ gas marketing functions are performed by PGPB’s “Natural Gas Area” while international marketing activities are made by PEMEX’ international subsidiary (PMI) PEMEX is able to exert its monopoly because the slight differences between marketing products (especially in financial futures markets) makes very difficult to technically prove that two marketing products are identical. For example, PEMEX might sell gas for a delivery 30 days in the future at a given price and the next day sell gas for delivery 29 days in the future at a different price. Technically, this transactions would not be discriminatory and would be very difficult to monitor by the regulator. The CRE actually tries to regulate Pemex’ gas marketing activities through requiring detailed information on PEMEX’ gas marketing, transportation and storage activities as well as gas sales, prices, contracts, availability, imports, exports, national balance, and price discount methods. Brito and Rosellón (2003a, p. 18–20) show that it is

Kaldor-Hicks superior to have the price of gas in Mexico equal to the Houston netback price and have Pemex sell the balance of the gas in the Houston market rather than sell gas in Mexico at a price below the Houston netback price 19. From equation (24) we get

20.

21. 22.

23.

24.

25.

Then, for example, a 700km southward displacement of the arbitrage point from Los Ramones (its current location) down to Cempoala (the next demand mass point) would increase the domestic natural gas price from USD0.50 to USD0.75 per MMBTU, which would imply an increase in Pemex annual rents in around USD170 million. Brito and Rosellón (2003b) propose two short-term measures that the CRE might use to provide PEMEX with incentives to increase supply in the domestic regulated market (and hence avoid a deliberated southwards displacement of the arbitrage point). The first measure is to temporally fix the arbitrage point at Los Ramones so that Pemex increases production (and investment) to a level corresponding to the price of gas implied by Los Ramones. The second strategy is to set a price based on the netback rule for internal gas transactions among PEMEX’ subsidiaries. Brito and Rosellón show that, although these measures can be at odds with long-run Pareto efficiency, their effects are minimal in the short run and help to deal with political pressures to keep domestic gas price low. See Brito and Rosellón (1998) Mexican consumers—as any other consumers in North America—could hedge in the developed US natural gas market (in particular the Houston market). However, political pressure by Mexican industrial consumers has proven to be effective. In addition to ask for a lower price for Mexican gas (unrelated to the netback rule), they have also put pressure so as to receive implicit subsidies through governmental hedging mechanisms. For example, the Mexican government implemented in 2001 three-year take-or-pay hedging contracts offered by Pemex with a fixed price of 4 dollars per Mmbtu, which eliminated any potential competition from private gas marketers. LNG projects only become economical when the long-run price of pipeline natural gas is more than approximately USD4.00 per MMBTU. Additionally, the social-welfare superiority of a government transfer to gas consumers as opposed to other social goals should also be gauged. This assumption does not change any of the results.

Dagobert L. Brito and Juan Rosellón

References Adelman, M.A (1963) The Supply and Price of Natural Gas, Oxford B. Blackwell. Arteaga, J. C. and D. Flores (2002) “Una Nota sobre la Regulación del Precio del Gas en Mexico,” El Trimestre Economico, vol. LXIX(1), No. 273, Mexico. Arteaga-García, Julio César; Flores-Curiel, Daniel (2003) “:Debe ser Texas la Referencia para fijar el Precio del Gas en México?,” EAWP2(10), available from: http://eawp.economistascoruna.org/archives/vol 2n10/ Brito, D. L. and J. Rosellón (1998) “Pricing Natural Gas in Mexico,” CIDE Working Paper, E-120. Brito, D. L. and J. Rosellón (2002) “Pricing Natural Gas in Mexico: an Application of the Little-Mirrlees Rule,” The Energy Journal, vol. 23, no. 3. Brito, D. L. and J. Rosellón (2003a) “Regulation of Gas Marketing Activities,” Estudios Económicos, Vol.18, No.1 January-June.

137

Brito, D.L. and J. Rosellón (2003b) “Strategic Behavior and the Pricing of Gas in Mexico,” 2003 , CIDE Working Paper, E-259 (under review at The Energy Journal) Brito, D. L., W. L. Littlejohn, and J. Rosellón (2000) “Pricing Liquid Petroleum Gas in Mexico,” Southern Economic Journal, Vol. 66, No. 3, January. Comisión Reguladora de Energía (1996) “Directiva sobre la Determinación de Precios y Tarifas para las Actividades Reguladas en materia de Gas Natural,” Mexico. (available at http://www.cre.gob.mx) Little, I. M. D. and J.A. Mirrlees (1968) Manual of Industrial Project Analysis in Developing Countries, Development Centre of the Organization for Economic Co-Operation and Development, Paris. Rosellón, J. and J. Halpern (2001) “Regulatory Reform in Mexico’s Natural Gas Industry: Liberalization in Context of Dominant Upstream Incumbent,” Policy Research Working Paper 2537, The World Bank.

7

Implications of the Elasticity of Natural Gas in Mexico on Investment in Gas Pipelines and in Setting the Arbitrage Point Dagobert L. Brito Department of Economics and Baker Institute, Rice University, and Centro de Investigación y Docencia Económicas (CIDE) Juan Rosellón CIDE and Harvard University

Abstract

Since the Houston market determines the price of gas in Mexico, a necessary condition for this policy to work is that gas be able to move to equilibrate supply and demand. Thus, it is essential that the pipeline system not be congested. If it does become congested, then it becomes impossible to supply the amount of gas that will clear the market at the Houston netback price. There will be excess demand and there are no institutions in place so that price can be the equilibrating factor. When the pipeline system becomes congested in the United States, such as in the summer of 2000, there can be disruptive peaks in the price of gas, rents accrue to agents who have access to the pipeline, but prices adjust to equilibrate supply and demand. If the pipeline system in Mexico were to become congested, the CRE’s netback pricing rule would not be feasible. Further, there would not be any market institutions to equate supply and demand and it would become necessary to use some political, ad hoc system to allocate the available gas. This would be very costly to the Mexican economy. Thus it is very important that there be sufficient pipeline capacity so that congestion does not occur. Unfortunately, the market is not a good guide to the allocation of resources in pipeline capacity. It can take as long as three years lead time to increase pipeline capacity, so it is necessary to rely on forecasts of future demands for the purpose of planning investment in pipeline capacity. These forecasts are at best uncertain. Mexico’s economy is to a large extent driven by economic activity in the United States. As we have seen in the recent past, forecasts of United States economic activity three years in the future are not always reliable. In this paper we will address the optimal timing of investment where the demand for gas is stochastic. We will show that this is a problem that can be solved in theory, but the solution depends

We address the optimal timing of investment in gas pipelines when the demand for gas is stochastic. We will show that this is a problem that can be solved in theory, but the practical solution depends on functions and parameters that are either subjective or cannot be estimated. We will then reformulate the problem in a manner that can Pareto rank investment strategies. These strategies can be implemented with reasonably straightforward policies. The demand for gas is very inelastic and thus the welfare losses associated from small deviations from a first best optimum are minimal. This implies that the gas pipeline system can be regulated with a relatively simple set of rules without any significant loss of welfare. Regulation of the gas pipeline system can be transparent and a result may be a good candidate for some institutional arrangement in which there is substantial private investment in gas pipelines. 1.

Introduction1

Mexico has adopted a policy of pricing natural gas based on the Houston price adjusted for transport cost. This is am application of the well known Little-Mirrlees Rule (See Brito and Rosellon, 2002) and results in the market for gas in Mexico having essentially the same character as the Houston market. Pemex behaves as a price taker and inasmuch as Mexico is importing gas from the United States, the price of gas to Mexican consumers reflects the marginal cost of gas to Mexico. The research reported in this paper was supported by grants from the Center for International Political Economy to Baker Institute for Public Policy at Rice University and the Comisión Reguladora de Energía to the Centro de Investigación y Docencia Económicas, A.C (CIDE). The second author also acknowledges support from the Repsol YPF-Harvard Kennedy School Fellows program, and the Fundación México en Harvard.

139

Repsol YPF-Harvard Kennedy School Fellows Research Papers

on functions and parameters that are either subjective or cannot be estimated. We will then reformulate the problem in a manner that can Pareto rank investment strategies. These strategies are not optimal in the strict sense of the word, but they can be implemented with reasonably straightforward policies. The demand for gas is very inelastic and thus the welfare losses associated from small deviations from a first best optimum are minimal. This implies that the gas pipeline system can be regulated with a reasonably simple set of rules without any significant loss of welfare. Regulation of the gas pipeline system can be transparent and a result may be a good candidate for some institutional arrangement in which there is substantial private investment in gas pipelines. 2.

The Production Function for Gas Pipelines

calculated under the assumptions that the real interest rate is 10 percent, the cost of pipeline is $25,000 per mile inch, maintenance costs are assumed to be 3 percent, and the cost of gas to power the pumps is $2.00 per thousand cubic feet (MCF). The cost of an installed horsepower was assumed to be $600 and the project life to be fifteen years. Figure 1 0.10

MC

0.09 0.08 dollars per 1000 cubic feet

140

0.07

AC

0.06 0.05 0.04 0.03

A simplified formula for computing the rate of flow of gas in a pipeline is given by

0.02 0.01

1,000

(1) where: D = internal diameter of pipe in inches L = length of line in miles Q = throughput in per day P1 = absolute pressure at starting point P2 = absolute pressure at ending point The amount of power needed compress a million cubic feet a day is given by (2) where: Z = horsepower R = the compression ratio, absolute discharge pressure divided by absolute suction pressure J = supercompressibility factor which we assume to be 0.022 per 100 pounds per square inch absolute suction pressure. Assuming as given the discharge pressure, equation (1) can be used to solve for the necessary pressure as function of the throughput. Equation (2) can then be used to compute the amount of power necessary. We can use these values to compute the cost of transporting gas. The costs were

2,000

3,000

4,000

5,000

6,000

million cubic feet

Pipelines have a high fixed cost, and for a substantial portion of their operating region low marginal costs. The capacity of the pipeline is ultimately limited by the pressure limits of pipe. Figure 1 illustrates the cost curves for a 48-inch pipeline 100 miles long. At a pressure limit of 1,500 pounds per square inch, the pipeline reached its limit at approximately 3,800 million cubic feet per day. The dashed line denotes this limit. At this point it becomes impossible to increase throughput by increasing power and it becomes necessary to add compressor stations that increases throughput without exceeding the line limit by increasing the pressure gradient. Note that this formulation leads to a cost of moving 1 MCF of gas 1000 miles to be $.50. We have shown in an earlier paper (Brito and Rosellon, 2002) that the netback-pricing rule is the solution of a static welfare optimization problem if the fee for transporting gas is the marginal cost of transporting gas. However, marginal cost pricing results in a loss or rents. (See Figure 1.) One solution to this problem is to set a fee that yields a regulated rate of return over the life of the project sufficient to cover all costs. An alternative, more sophisticated alternative is a two-part tariff with a

Dagobert L. Brito and Juan Rosellón

141

price cap. The sophisticated price cap mechanism is efficient in that it sets the marginal cost of transporting gas equal to the variable change for moving gas. The question is whether the more efficient allocation of resources merits the additional difficulties in regulation. Figure 2 p

AC

MC

D ( p) Q

The shaded area in Figure 2 illustrates the welfare loss associated with using average cost rather than marginal cost in transporting gas. The loss, L, is given by (3) where η is the elasticity of the demand for gas. Simple calculations suggest that for elasticities in the demand for gas in the range of –0.1 to –0.2 the welfare loss is of second order and can be ignored. If we calculate the dead weight loss for 4 million MCF the price of gas equal to $2.00 per 1,000 cubic feet, an elasticity for the demand for gas equal to –0.1, and a differential between AC and MC of $0.02, we get that the change in demand is 4,000,000 cubic feet and the deadweight loss is $40. Since the cost of moving gas is linear with distance, the deadweight loss over a distance of 1000 miles is $400 for 4 million MCF of gas. At a price of $4.00 per MCF, the welfare loss would be half. The welfare loss associated with using a rate of return fee structure for transport pipelines is so small that it is hard to see how the additional complexity in regulation can be justified given the low elasticity in the demand for gas in Mexico. The low elasticity of the demand for gas has some implications on the implementation of the

netback rule for pricing natural gas. The net back rule leads to the optimal price of gas in that the price of gas is the opportunity cost of gas. However, the price of gas is very sensitive to small in the geographical demand for gas. Since demand for gas tends to be concentrated at mass point along the pipeline system, a very small change in demand can result in a substantial change in the price of gas. Initially this was not an issue of policy concern. Gas from the southern fields was reaching Los Ramones. However, as of late, the demand for gas in the south of Mexico has increased to the point where the physical arbitration point is at Cempoala in the south of Mexico. There is pressure on the CRE to move the point used to price gas south to Cempoala. In a first best world there is no question that Cempoala is the correct point to price gas. The opportunity cost of gas to Mexico is the price of gas in Houston corrected fro transport cost. There are two separate independent arguments that can be made against moving the arbitration point to Cempoala. First is that it is not a first best world and, in theory, there exist incentives for Pemex to invest and produce so as to move the arbitration point south. Whether they do so or not is not a question we cannot answer. As economists all we can say is that the incentives to manipulate the price of gas exist. (See Brito and Rosellon 2003). The second reason is political. Because the demand for gas is so inelastic, pricing gas in Mexico is essentially a question of the redistribution of rents. For example, moving the arbitration point by 500 miles will cause the price of gas to change by $.50 per MCF. At a price of $3.50 per MCF the distortion cause by a subsidy is one-third cent per MCF. (See Figure 3 below). Given the other distortions in the economy, a distortion that small is simply not large enough to argue that economic considerations should trump political considerations in the setting of the arbitrage. Using Houston as a benchmark to price gas is a useful instrument in deciding whether to use natural gas to produce ammonia nitrate; it is not a particularly useful tool in allocating the use of gas between Monterrey and Puebla. Consider the following example. Suppose the arbitration points were at Los Ramones and 10 MCF a day of gas was reaching Los Ramones from the southern fields. Now a tortillería that consumes 20 MCF of gas a day moves form Monterrey to

Repsol YPF-Harvard Kennedy School Fellows Research Papers

142

Figure 3 $0.010 $0.009 $0.008

Welfare Loss

$0.007 $0.006 $0.005 $0.004 $0.003 $0.002 $0.001 $0.000 $0.00

$1.00

$2.00

$3.00

$4.00

$5.00

$6.00

$7.00

$8.00

$9.00

$10.00

P

Figure 4

Puebla. The arbitration point is now at Cempoala. Does it make sense to change the entire pricing structure of gas in central Mexico because a tortillería has moved from Monterrey to Puebela?

e αtQ0 D( p) Q

3.

Timing of Investment in Pipeline Capacity: The General Case

Q

Let us consider the case when gas is being transmitted a distance L over a pipeline of diameter D. The demand for gas is given by (4) where α is a random variable with mean α– and p − is a random variable with mean p . Some of the stochastic elements are short term such as weather and others are long term that can reflect macroeconomic conditions tine Mexico and in the United States. – The pressure limit on the pipeline is Q and we – will define T such that .

T

T

Define e –rTC(T – t) as the cost of building a pipeline at time t that will come on line at time T. It is assumed that the cost of construction drops as lead time increase, but that there exists some minimum feasible lead time, T – t = ∆*. Define f [s,Q(t)] as the probability at time t that – Q(s) = Q for some s > t, given that demand at time – t is Q(t) < Q . Define S(n,s) as the consumer surplus lost at time n if the constraint becomes binding at

Dagobert L. Brito and Juan Rosellón

143

Cost

Figure 5

C(T – t)

T–t

∆*

in welfare. The fact that they have chosen not to do so suggests that in some political or economic calculation that is more general than the timing of investment in pipelines it was decided that the benefits from taxing gas were out weighed by other economic or political factors. Strictly speaking, the calculation of the optimal timing of pipeline investment should be done in the context of the more general problem. This is not possible, but we can approximate the more general problem by assigning a cost [[GRAPHIC]] to the transfers so that the cost of the transfers is given by

time s. The welfare loss, W(s) of the constraint binding at time s is thus, (5) and the expected welfare lost at time t is: (6) If the constraint binds, the price of gas will have to increase as gas cannot move to equilibrate the market at the netback price. Figure 6 S(p)

pc

(9) where α = 0 means that there is no cost to the government associated with transfers cause by congestion of the pipelines and α = 1 means that the interests of the government and the consumers of gas are identical. We can then compare the outcome of this maximization with policies that are Pareto superior under the assumption that the government does not want to tax gas by collecting the transfers caused congestion. That is to say, we can assume the government does not want this revenue since they could have collected it by taxation and chose not to do so. Then, if gas consumers are willing to pay for a level of pipeline capacity that eliminates transfers, then they are better off and no one is worst off. Such a policy would be Pareto superior to one that could result in congestion and transfers.

pn

4.

D(p)

Q Q

Define R(s,n) as the rents at time n if the constraint becomes binding at time s. Define the total transfer that results from these rents as Z(s). Thus, if the constraint binding at time s,

Optimal Investment in Pipeline

Let us assume that Pemex is trying to time investment in gas pipelines to minimize a cost function that is the sum of the investment in pipelines, loss of consumer surplus and a weight sum of the transfers:

(10)

(7) this expression can be written as

and the expected value of transfers at time t is: (8) These are transfer from the consumers of gas to Pemex and as such they do not represent a loss

(11)

Repsol YPF-Harvard Kennedy School Fellows Research Papers

144

If we differentiate with respect to T, we get

5.

(12)

Timing of Investment in Pipeline Capacity: An Alternate Approach

Let us again consider the case when gas is being transmitted a distance L over a pipe line of diameter D. The demand for gas is given by (18)

The term

so

(13) and we get the expected result that the target date of completion of the pipeline is when expected marginal benefits are equal to the marginal cost. There are two problems. First, the distribution function on the probability that the constraint will be binding is not well defined and depends on such factors as the performance of the United States economy. Second, the solution depends on the subjective value of the parameter α. The outcome is substantially a function of the choice of α. If we assume that the demand function is locally linear then

where α is a random variable with mean α– and p is a random variable with mean –p. The pressure – – limit on the pipeline is Q and we will define T such – – – that Q = eαTQ0D(p– ). Assume that initial demand is given by so the expect time for the pipeline to reach full 1n(2) capacity is . Now let us consider a sequence of investment such that pipeline capacity is doubled every time the pipeline reaches full capacity. Thus there is a sequence of investments – at Ti , where Ti = Ti–1 + t . Let c1 be the charge for transporting gas. The present value of the revenues of the pipeline are given by

(19)

(14)

Now consider any other sequence of investment T˜i , where T˜i = T˜i–1 + t˜ and let c2 be the charge for transporting gas. Then

(15)

. (20)

(16)

If we assume the consumer of natural gas is paying for the buffer capacity, then PV1 = PV2 and

and

and

so ratio (17) (21) If we assume α– = .06 and ,

– then α–(t – T ) = .005 and ρ = 400. Note, however, that the solution depends on the value of α which is subjective.

or (22) and the difference in the costs can be expressed as a function of c1,

Dagobert L. Brito and Juan Rosellón

145

.

.

(26)

(23)

The cost per thousand cubic feet of gas transported – for maintaining a t – t˜ buffer of excess capacity, ∆C is given by substituting into equations (19) and (20).

(24) Let us calculate a simple example assuming that r = .12 and α– = .06, and that the cost without a buffer is $.10 per 1000 cubic feet. If there is no buffer then at a growth rate of six percent a year, –t = 11.5. Table 1 below gives the cost per MCF of maintaining excess buffer capacity. Table 1 Cost per Year of Pipeline Buffer Capacity

Note that there are three random elements in this expression, the net back price, p, at the time of congestion, the percentage of above full capacity , and the probability that the pipeline will be congested. Of these random variables, the net back price is the only one for which there exists published forecasts and historically these have not been very accurate.

Using the Mean Value Theorem

(27) Since we are evaluating the integral at the end point, T. The expression,

Year

Change in Tariff dollars

Present Value of Cost dollars

1 2 3 4 5

.006 .013 .020 .027 .035

9.37 19.32 29.12 41.12 53.04

,

is a lower bound of the expected cost of congestion to the consumer. If we assume that consumers are risk neutral, we can construct a variable such that (28)

Now consider a consumer that purchases an amount of gas Q1 over the period (0,t– ). The consumer faces two alternatives: First, the consumer can pay an transport charge c1 and run the risk that the pipeline will be congested; or second the consumer can run the risk that the pipeline will become congested. Suppose that it is possible to create a market mechanism to allocate gas if the pipeline becomes congested. This is unlikely, but it is a lower bound of the expected cost. The increase is price is given by

π θ= In this formulation, is the expected over capacity and t is the number of days the pipeline is congested. Thus we an express a lower bound of the tradeoff for consumers between buffer capacity to the pipeline and days of expected over capacity for a given value of θ.

(29) ,

(25) which can be solved for t.

for the period during which the pipeline is congested. Let g(t) be the probability that the pipeline will be congested at time t. The present value of the expect rents the consumer will pay over the planning period pay is:

(30) Figure A below gives the relationship for a price of gas of $3.00 per MCF. To illustrate, an indi-

Repsol YPF-Harvard Kennedy School Fellows Research Papers

146

vidual whose subjective expectation is that θ = .04 would rather pay the costs associated with two years of excess capacity rather than risk 31.6 days of congestion. An individual whose subjective expectation is that θ = .12 would rather pay the costs associated with two years of excess capacity rather than risk 10.6 days of congestion. Similar calculations can be performed for other assumptions about the price of gas. Alternatively, it is possible to examine the relationship between days of congestion and the price of gas for a fixed amount of amount of buffer. This is illustrated in Figure B. Suppose the price of gas is expected to be in the range of $3.00 to $6.00, then individuals whose subjective expectation of θ was greater than .04 would rather pay for two years of excess capacity rather than risk 30 days of congestion. An Example To get an intuitive insight as to what could lead to 30 days of congestion, it is useful to compute a simple example. Assume that a pipeline has an

increase of throughput that grows at six percent a year. If initial throughput is 2 where the capaci– ty of the pipeline is Q we can expect the pipeline to be congested in 11.5 years. Now suppose that after 9.5 years the growth rate increased by a one percent so that α = .07. The question is how days of congestion will result at θ = .04? The quick answer is 34. If throughput is growing at a rate α = .06, then after 8.5 years throughput will be equal to 1.67 . At a growth rate of .07 after the ninth year 2 the pipeline will reach capacity after 11.12 years. The number of days of congestion at θ = .04 is

.

. The numerator is the cumulative θ and the dominator normalizes it for θ = .04. Using very naïve calculations, a growth rate of .07 rather than .06 in the last three years of the planning period would result in over 30 days of congestion. The

Figure 7 Price of Gas $3.00 MCF .04 35 .06

30

.08

Days of Congestion

25

.10

20

.12 15

10

5

1

2 Years of Buffer Capacity

3

4

Dagobert L. Brito and Juan Rosellón

147

Figure 8 50

Max & Min for Two Year Buffer

45 40

Days of Congestion

35 30 25 20 15 .04 10 5

.12

0 0

1

2

3

4

5

6

7

8

Price of Gas real world is very much more complicated and there are problems such as construction delays, weather, macro-economic shocks, or war in the Middle East. The cost of buffer capacity is low and the cost of transfers that result from congestion to the consumers of gas of congestion is very high. This completely ignores social and political costs that would result if the gas pipeline system becomes congested and gas cannot flow to clear the market. 6.

Conclusions

The fact that the demand for gas is very inelastic in Mexico is a two edged sword with respect to the administration of the net back rule for pricing gas. On one hand, a very small change in the demand for gas can lead to a large change in the arbitration point, however on the other hand the fact that the demand for gas is very inelastic means that the welfare loss associated with the pricing of gas based on an artificial pricing point is very small. Cempoala is about 500 miles from Los Ramones so a shift of the arbitrage point from Los Ramones to

Cempoala would lead to a change in the price of gas of approximately $.50 per MCF. However at a price of $3.50 per MCF the welfare loss associated keeping the arbitrage point at Los Ramones is on the order of one third cent per MCF. Since very small changes in the demand for gas can lead to substantial changes in the net back price and since the welfare losses from maintaining an artificial point for price are low, the question is more political than economic. The opportunity cost of gas based on the Houston market can be used to argue why natural gas in Mexico should not be used to produce ammonia nitrate. It is harder to use that price to justify why a factory in Puebla should pay substantially more for gas than a factory in Monterrey. As illustrated in the example of the tortillería, this is particularly true when a very small change in the pattern of demand can lead to a substantial change in the price of gas. The fact that the demand for gas is very inelastic means that the welfare cost of keeping price of gas stable in Mexico is low. Similarly, the fact that the demand for gas is very inelastic in Mexico is a two edged sword with

148

respect to pipeline capacity. A ten percent increase in demand would result in a one hundred percent increase in the price that would clear the market is gas is not free to flow to maintain the net back price. However the fact that the demand is so inelastic permits the implementation of a very simple rate structure and appears to justify investment in substantial buffer capacity. Such capacity may be Pareto superior. Substantiation of the latter conjecture is beyond the limited scope of this paper. However, calculations suggest that users would prefer to pay for excess capacity in the pipeline system than to risk the consequences of congestion. Since the parameters needed to calculate this result are subject, it must remain a conjecture. Experience in the United States suggests that such periods of congestion do occur. The price of gas in the United States is set by market forces and an equilibrium can be reached. The netback rule, however, requires that gas be able to flow to achieve equilibrium.

Repsol YPF-Harvard Kennedy School Fellows Research Papers

References Adelman, M.A, 1963, The Supply and Price of Natural Gas, (B. Blackwell, Oxford). Brito, D. L., W. L. Littlejohn and J. Rosellón, 2000, “Pricing Liquid Petroleum Gas in Mexico, Southern Economic Journal, 66 (3), 742–753. Brito, D. L. and J. Rosellón, 2003, “Strategic Behavior and Pricing of Gas,” manuscript. Brito, D. L. and J. Rosellón, 2002, “Pricing Natural Gas in Mexico: an Application of the Little-Mirrlees Rule,” The Energy Journal, 23 (3), 81–93. Comisión Reguladora de Energía, 1996, “Directiva sobre la Determinacion de Precios y Tarifas para las Actividades Reguladdas en materia de Gas Natural,” MEXICO. (WEB SITE:http://www.cre. gob.mx) Little, I. M. D. and J.A. Mirrlees, 1968, Manual of Industrial Project Analysis in Developing Countries, (Development Centre of the Organization for Economic Co-Operation and Development, Paris) Pemex, 1998, “Indicadores Petroleros y Anuario Estadístico”. Rosellon, J. and J. Halpern, 2001, “Regulatory Reform in Mexico’s Natural Gas Industry: Liberalization in Context of Dominant Upstream Incumbent,” Policy Research Working Paper 2537, The World Bank. Secretaría de Energía, 1998, “Prospectiva del Mercado de Gas Natural, 1998–2007.” (WEB SITE: http:// www.energia.gob.mx/frame4.html)

8

An Environmental Kuznets Curve Analysis of U.S. State-Level Carbon Dioxide Emissions1 Joseph E. Aldy Department of Economics, Harvard University

some states with nonstationary income and emissions data. Finally, I find that cold winters, warm summers, and historic coal endowments are positively associated with states’ CO2 emissions.

Abstract Most environmental Kuznets curve (EKC) theories do not apply to carbon dioxide—an unregulated, invisible, odorless gas with no direct human health effects. This analysis addresses the hypothesis that the income-CO2 relationship reflects changes in the composition of an economy as it develops and the associated role of trade in an emissions-intensive good, electricity. To test this hypothesis, I use a novel data set of 1960–1999 statelevel carbon dioxide (CO2) emissions to estimate pretrade (production-based) CO2 EKCs and post-trade (consumption-based) CO2 EKCs. As the first EKC analysis of CO2 emissions in the U.S. states, I find that consumption-based EKCs peak at significantly higher incomes than production-based EKCs, suggesting that emissions-intensive trade drives at least in part the income-emissions relationship. I have also investigated the robustness of the estimated income-CO2 relationship through a variety of specifications. Estimated EKCs appear to vary by state, and the estimated income-emissions relationships could be spurious for

1.

Introduction

Empirical researchers have characterized the relationship between economic development and environmental pollution with the environmental Kuznets curve (EKC)—pollution follows an inverted-U shape with respect to per capita income. The application of EKC analyses to greenhouse gas emissions, such as carbon dioxide (CO2), has raised several important questions. First, some empirical studies have estimated an inverted-U shape of per capita CO2 emissions with respect to per capita income, but with a peak in this function occurring well outside the range of incomes in the studies’ samples. Since these studies often rely on restrictive regression specifications (i.e., quadratic income), these results do not support unequivocally an inverted-U for per capita CO2 emissions. Moreover, recent empirical analyses have challenged the robustness of estimated environmental Kuznets curve relationships (Harbaugh et al. 2002, Millimet et al. 2003, Perman and Stern 2003). Second, some studies have suggested that the inverted-U reflects an economy’s changing structure as it develops: economies are characterized by agriculture at low incomes, then move to more emissions-intensive manufacturing at middle incomes, and then transition to less emissionsintensive services at high incomes. Trade in emissions-intensive goods facilitates this specialization. The downward slope of the inverted-U estimated for higher incomes may reflect the combination of a transition from manufacturing to services and an increase in imports of emissions-intensive manufactured goods. Since all economies cannot import

This paper was prepared in residence under the Repsol YPF-Harvard Kennedy School of Government Energy Policy Fellowship program. Additional support was provided by the Environmental Protection Agency STAR Fellowship program, and the Switzer Environmental Fellowship program. This paper is supported by EPA Contract No. 4W-2131-NTTX. Thanks to Arthur Rypinski for providing state energy consumption data and Richard Bonskowski for state coal production data. The author acknowledges the useful comments provided by Gary Chamberlain, Niko Dietsch, Bryan Graham, Mark Heil, Bill Hogan, Jane Leggett, Michael Leifman, Rob Stavins, Kip Viscusi, and two anonymous referees, and participants of the Environmental Economics Program at Harvard University seminar. None of the opinions expressed in this paper represent the views of the Environmental Protection Agency. All errors remain solely those of the author. 149

150

emissions-intensive goods, failing to account for these trade effects could bias downward long-term emissions forecasts. This paper addresses these two questions and makes several additional contributions to the environmental Kuznets curve literature and related policy debates. First, by using a novel data set constructed by the author of state-level CO2 emissions for the 1960–1999 period, this paper presents the first panel-based environmental Kuznets curve analysis for CO2 emissions in the U.S. states. Second, by focusing on the U.S. states—a set of economies that have achieved advanced stages of economic development—this analysis provides better evidence of whether per capita emissions actually do fall at high income levels. Third, with a states dataset, the empirical analysis investigates explicitly the effects of emissions-intensive trade (specifically, electricity trade) on the income-CO2 relationship. Fourth, this research evaluates other determinants of CO2 emissions, such as energy endowments and the variation in winter and summer climates across the country. Finally, by employing a variety of econometric methods, this paper assesses the robustness of the income-emissions relationship. The empirical results based on standard EKC specifications illustrate that per capita CO2 emissions may follow an inverted-U pattern with respect to per capita income for the U.S. states over the 1960–1999 period. The estimated peak in the environmental Kuznets curve does occur at incomes that fall within the sample range. The estimated EKCs, however, are sensitive to a number of modifications to the analysis. First, as an explicit test of the effect of trade in emissionsintensive goods, I have estimated pre- and postelectricity trade environmental Kuznet curve regressions. I find that consumption-based CO2 per capita (post-trade) EKCs have peaks at much higher incomes than the standard (productionbased or pre-trade) CO2 per capita measure. Further, consumption-based CO2 emissions appear to remain much higher at high incomes than production-based CO2 based on more flexible regression specifications. This suggests that individuals in high income states don’t consume less carbonintensive goods than those in lower income states, but that they consume more imported carbonintensive goods and lower income states may be net exporters of carbon-intensive goods. Second, I

Repsol YPF-Harvard Kennedy School Fellows Research Papers

assess whether the income-emissions relationship is the same across the states. Tests of heterogeneous slopes models show that this relationship varies across the states. Third, an evaluation of the time series properties of the state-level data suggests that some estimated relationships could be spurious and that less than one in five states follow an inverted-U environmental Kuznets curve for which one could reject the possibility of a spurious relationship. Finally, I find that states’ cold winter and warm summer weather and historic coal endowments are positively associated with states’ per capita carbon dioxide emissions. By characterizing the income-CO2 relationship for the U.S. states, this analysis can also help inform our understanding of greenhouse gas emissions in an international context. First, illustrating the economic dynamics of per capita CO2 for highincome states provide evidence of what may occur for countries as they achieve advanced stages of development. Second, the potential role of trade in emissions-intensive goods (e.g., electricity) in the income-CO2 relationship may be valuable for other regions of the world that may share similar characteristics to the U.S. states. For example, the European Union, with converging incomes, policies, and institutions and substantial cross-border trade in emissions-intensive goods and energy may follow similar production- and consumption-based CO2 trends as the U.S. states. Third, if trade in emissions-intensive goods is as important in the international context as for the U.S. states, then this work suggests that studies that attempt to forecast carbon dioxide emissions may produce biased results without attempts to correct for trade. The next section briefly reviews several key hypotheses of the environmental Kuznets curve literature as it relates to greenhouse gas emissions. The third section provides an overview of the data used in this paper, including the novel state-level carbon dioxide emissions dataset constructed by the author. The fourth section describes the empirical methods and presents the results of the analysis. The final section concludes. 2.

Hypotheses of Environmental Kuznets Curve Related to Greenhouse Gas Emissions

A variety of theories have been posited to motivate the empirical work of the environmental Kuznets curve (see Selden and Song 1994, Arrow

Joseph E. Aldy

et al. 1995, Stern 1998). First, environmental quality may be income elastic. As individuals enjoy greater incomes, they demand better environmental quality either through markets or regulatory policies. Second, and related to this first point, is the increasing role of democracy with economic development. Since emissions of many environmental pollutants reflect missing markets, government institutions are necessary to address them. More responsive democracy may be necessary in order to translate individual demand for environmental quality into policies that restrict pollution. In contrast to air or water pollution, which can have immediate, identifiable local health effects, carbon dioxide emissions are locally innocuous and only impact the global environment over the long term. Moreover, per capita carbon dioxide emissions have no local impacts and reflect at best an indirect measure of the impact on the global environment. It is not clear that either of these phenomena would explain the per capita carbon dioxide environmental Kuznets curve.2,3 Third, the inverted-U shape may reflect changes in production associated with an economy’s stage of development and a wedge between the emissions-intensity of production and the intensity of consumption (Arrow et al. 1995, Rothman 1998). For example, a decrease in pollution in one economy may simply represent a shift in the polluting production activity to another economy, which would then experience an increase in pollution. This second economy would then export pollution-intensive goods to the first economy. This could follow the development path from agriculture (low income) to heavy industry (middle income) to services (high income). Since agriculture tends to be less energy-intensive (carbonintensive) than heavy industry, which is also more energy-intensive (carbon-intensive) than services, this development path could result in an environmental Kuznets curve for carbon dioxide. Note, however, that the inverted-U would only be temporary, since every economy cannot specialize in services and export its heavy industry to other economies. While a variety of theories may explain the shape of the environmental Kuznets curve for many local air pollutants, the developmentinduced changes in production coupled with trade story seems most plausible for carbon dioxide. An empirical test of this theory could attempt to dis-

151

cern the production of carbon-based goods from the consumption of these goods. Modifying measures of carbon dioxide emissions to reflect the location of consumption and comparing these with the standard measures that reflect location of production could allow for an assessment of whether the inverted-U shape of per capita emissions with income simply reflects shifts in production or substantial changes in the carbon-intensity of consumption at higher income levels. Several papers have attempted to test for this production-location hypothesis in an international context by simply expanding the set of regressors to include measures of trade and manufacturing intensity. For example, Harbaugh et al. (2002), in an analysis of sulfur dioxide concentrations and other pollutants, included a measure of trade intensity, which nearly doubles the income at which the estimated environmental Kuznets curve peaks. Suri and Chapman (1998) investigated energy use per capita and compared EKC regressions with and without a number of such controls: the ratio of manufacturing exports to domestic manufactured production, the ratio of manufacturing imports to domestic manufactured production, and the ratio of total manufacturing value added to GDP. For the Suri and Chapman analysis without these controls for trade, the estimated EKC peaked at about $55,000 (1985$). Including these additional variables resulted in a peak of nearly $144,000. In contrast, Cole (2003) included measures of trade intensity in EKC regressions of SO2, NOX, and CO2, and found that these tended to have a minor impact on the estimated peaks in these environmental Kuznets curves. Frankel and Rose (2002) have also investigated the role of trade and economic growth on environmental quality, and they find in-sample environmental Kuznets curve peaks for SO2 and NOX concentrations when accounting for the trade share of GDP. Interestingly, Frankel and Rose estimate an alwaysincreasing environmental Kuznets curve for CO2 per capita when controlling for trade.4,5 While some papers suggest that accounting for trade generally may increase the income at which environmental Kuznets curves peak, caution should be exercised when considering this approach. Some researchers have noted previously that as a reduced-form framework, EKC regressions should not include regressors that may be endogenous to the income variables, or the eco-

Repsol YPF-Harvard Kennedy School Fellows Research Papers

152

nomic growth process more generally (Holtz-Eakin and Selden 1995, Heil and Selden 2001). It may not be appropriate to include the trade share of output or the economy’s manufacturing intensity as regressors and then estimate the income at the EKC peak from the regression’s income variables holding everything else, including trade and manufacturing variables, constant (as is typically done in the literature). Since manufacturing intensity and trade intensity are likely to be systematically related to the economic growth process, constructing the EKC peak only from the income variables would result in a biased estimate because it omits the information in the manufacturing and trade variables that may explain in part the relationship between development and emissions. The approach in this paper to modify the measure of emissions directly to reflect trade aims to circumvent this potential bias. 3.

Emissions and Income Data

3.1 Carbon Dioxide Emissions Data Since long-term CO2 emissions data sets do not exist for the U.S. states, I have constructed statelevel emissions estimates based on fossil fuel combustion data (refer to Lutter 2000 and Marland et al. 2003 for similar applications of this approach).6 The Energy Information Administration (2001b) has compiled state-level energy consumption by fuel type and sector for the 1960–1999 period. I converted energy consumption by sector-specific fuel type to CO2 emissions using U.S. emissions factors provided in EIA (2001a; Appendix B).7 As one test of the plausibility of this construction, I aggregated these state-level CO2 emissions values to yield annual national estimates and compared these to the Marland et al. estimates for U.S. emissions. Over the 1960–1999 period, my constructed U.S. values differ on average 1.7 percent from the Marland et al. estimates (1.0 percent standard deviation). The maximum annual differential between the data sets is 3.8 percent. A comparison with EIA (2001a) CO2 emissions estimates for the United States over this time period yields very similar results. While some states likely have measurement error in excess of 1.7 percent, this comparison illustrates the plausibility of the energy-based construction of state carbon dioxide emissions developed for this paper. These carbon dioxide estimates reflect all

within-state fossil fuel combustion emissions. They represent emissions associated with producing all goods and services in a given state, so we can also denote them production-based CO2 emissions. In the presence of interstate trade, the emissions intensity of a state’s production may differ from the intensity of this state’s consumption. To explore this distinction, a second carbon dioxide emissions data set was constructed to account for interstate electricity trade. To modify the carbon dioxide emissions data set, I first calculated the annual average carbon-intensity of each state’s electricity sector. For a state that is a net exporter of electricity in a given year, the carbon emissions associated with the exported electricity (reflecting the state’s average electricity carbon intensity) are deducted from that state’s total emissions for that year. For a net importer, that state’s emissions are augmented based on the average carbon intensity of electricity imports.8 Since this modified measure reflects post-trade emissions and attempts to approximate for consumption-based emissions, as opposed to the production-based or standard measure of emissions, I refer to it as consumptionbased CO2 throughout this analysis.9, 10 3.2 State Income and Population Data The income variables used in the statistical analyses represent state personal income data provided by the Bureau of Economic Analysis (2000).11 These data have been used in a variety of economic analyses, including papers on economic growth and the environmental Kuznets curve (e.g., Barro and Sala-i-Martin 1992, List and Gallet 1999, and Millimet et al. 2003).12 The Bureau of Economic Analysis also provides annual statelevel population estimates. These population values were used to construct all per capita estimates. 3.3 Climate Data Previous research has shown that weather fluctuations can cause short-term shocks to energy demand that influence carbon dioxide emissions (Considine 2000, EIA 2001a). The National Oceanic and Atmospheric Administration (NOAA) has developed so-called “heating degree-days” (HDD) and “cooling degree-days” (CDD) metrics to characterize the effects of high summer temperatures and low winter temperatures on energy demand.

Joseph E. Aldy

153

Annual data for state heating and cooling degreedays were compiled from Heim et al. (1993a, 1993b) and NOAA (multiple dates). 3.4 Energy Endowment Data The energy endowment variable reflects a state’s coal endowment. To focus on the persistent effects of historic coal production on the industrial composition and associated emissions in a state, I focus on lagged coal production (ten-year lag). The historic coal production data reflect state-level production of all types of coal. These data were compiled and provided by the U.S. Energy Information Administration (Bonskowski 2002), and measure coal in thousands of short tons. 3.5 Descriptive Statistics

varies from about four in the 1960s to the high teens over the past several decades.13 This variation in per capita emissions is not too dissimilar from the variation evident in international data sets: the ratio of U.S. per capita emissions to India per capita emissions in 1999 is almost identical to the ratio of the maximum to the minimum per capita emissions among the states in that year. Per capita incomes vary by about a factor of two in any given year and by a factor of more than five over the entire 40-year sample. Incomes at the beginning of the sample ranged between about $7,000 and $16,500, about on par with current middle-income developing countries (e.g., Venezuela or Brazil) and lower income OECD countries (e.g., the Czech Republic or South Korea), in purchasing power parity terms. 4.

The variables used in the regression analyses presented below are summarized in Table 1. The state-level data for the 1960–1999 period reveal substantial variation in the per capita carbon dioxide emissions data and per capita income data. State average per capita carbon dioxide emissions is about 5.75 tons. Within a given year the ratio of maximum to minimum per capita carbon dioxide

Methods and Results

4.1 Estimation Strategy To estimate the environmental Kuznets curve for state-level carbon dioxide emissions per capita, I employed four econometric approaches with a panel of 48 states over the 1960–1999 period.14 The general regression specification takes the following form:

Table 1 Summary Statistics of U.S. State-level Data, 1960–1999 Variable

Description

Carbon Dioxide (CO2)

Per capita carbon dioxide emissions from fossil fuel combustion (metric tons of carbon).

Consumption-Based Carbon Dioxide (CO2cons)

Per capita carbon dioxide emissions from fossil fuel combustion corrected for emissions associated with net interstate trade in electricity (metric tons of carbon).

y

Personal income per capita, 1999 dollars.

CDD

Mean in levels (standard deviation) 5.74 (3.81)

Mean in logs (standard deviation) 1.61 (0.49)

5.34 (2.34)

1.59 (0.40)

$19,571 ($5,123)

9.85 (0.27)

Annual cooling degree days.

1051 (758)

6.69 (0.77)

HDD

Annual heating degree days.

5,321 (2,041)

8.48 (0.50)

coal production

State coal production (10-year lag), 1000s short tons.

13,151 (30,403)

8.85 (1.98)

Notes: N = 1,920.

154

where CO2it represents per capita carbon dioxide emissions in state i in year t, yit represents per capita income, X represents the vector of other explanatory variables (lnCDD, lnHDD, and coal production), β and γ are vectors of parameters to be estimated, α i and τ t are state and year fixed effects, εit is the error term which is characterized by a strict exogeneity assumption:

and Zis represents the vector of all regressors.15 In the first set of regressions, the income function is specified as quadratic, following Grossman and Krueger (1995), Holtz-Eakin and Selden (1995), and Heil and Selden (2001). These specifications were estimated by ordinary least squares (OLS) and a feasible generalized least squares (FGLS) approach that corrects for cross-sectional heteroskedasticity and incorporates a one-lag autoregressive error structure.16 For the OLS regressions, I report both robust standard errors and standard errors corrected for within-group heteroskedasticity (the errors have been “clustered” by state). The robust standard errors correct for cross-sectional heteroskedasticity, but assume independence in the residuals in the time series dimension, while the clustered standard errors allow for state-specific arbitrary serial correlation but assume independence across states. In the second set of regressions, I characterize income by a cubic spline function, a more flexible variant of the piecewise linear spline approach used in Schmalensee et al. (1998). The cubic spline ensures that the estimated environmental Kuznets curve is smooth (twice everywhere differentiable) by fitting cubic functions of income in-between analyst-chosen points or knots in the distribution of the data. For example, one could choose nine knots in the data, one at each decile in the income distribution, and fit cubic functions specific to each decile but constrained to be smooth at every knot. I experimented with a variety of specifications, including two to as many as eleven knots in the cubic spline function. The results presented below are robust to setting the distance between knots based on quantiles of the data or based on an equal inter-knot distance rule.

Repsol YPF-Harvard Kennedy School Fellows Research Papers

Third, I have made the regression specification more flexible by allowing the slope parameters (β ’s) to vary by state, following List and Gallet’s (1999) environmental Kuznets curve analysis of state NOX and SO2 emissions. This approach allows for an explicit test for whether the invertedU income-emissions shape is common across all economies or whether this relationship is economy-specific. I find evidence of the latter. Fourth, based on the evidence that the EKC shape is state-specific, I conduct analyses on the 48 separate time series. This facilitates an investigation of state-specific relationships and an evaluation of the time series properties of the data. I find that the emissions and income data are stochastically trending, and that these two time series are cointegrated for only nine states. I conduct dynamic ordinary least squares (DOLS; Stock and Watson 1993) for these states and compare the results with state-specific OLS. The estimation strategy allows for an examination of two important issues typically not addressed in the environmental Kuznets curve literature. First, I have undertaken a number of regressions with a modified measure of per capita carbon dioxide emissions that reflects the effects of trade on emissions. The standard, productionbased measure of carbon dioxide emissions may not best characterize the carbon-intensity of a state or region’s standard of living. Consider a hypothetical world with two regions: region 1 specializes in carbon-intensive goods, and region 2 specializes in carbon-lean goods. Suppose that region 1 exports its carbon-intensive goods to region 2, and region 2 exports its carbon-lean goods to region 1. This standard approach to estimating carbon dioxide emissions reflects a region’s carbon-intensity of production, and region 1 would have higher emissions than region 2, ceteris paribus. The carbon-intensity of consumption, however, would vary from the production carbon-intensity because trade allows the import of carbon-intensive goods in region 2 (and the import of carbon-lean goods in region 1). If one assigned carbon dioxide emissions to the region in which the good associated with those emissions is consumed, then estimates of region 1’s emissions would fall while region 2’s emissions would increase. Modifying the measure of carbon dioxide emissions to reflect electricity trade is a first step

Joseph E. Aldy

towards employing a true consumption-based carbon dioxide measure. The analyses with this modified measure can serve to illustrate whether the failure to account for trade yields a distinctly different income-emissions relationship and test for the production-shifting hypothesis of the environmental Kuznets curve. Second, I have included a number of control variables in addition to per capita income that would reasonably be expected to influence per capita carbon dioxide emissions. This analysis can quantitatively assess questions such as: Do warm weather regions have lower per capita emissions than cold weather regions? Do historical energy endowments influence current per capita emissions? To address these questions, I have included the following variables in a subset of the regressions reported below: annual measures of weather-related heating and cooling demand and lagged (10-year) coal production by state. Colder winter states and hotter summer states would be expected to have higher per capita emissions resulting from greater heating demand (winter) and cooling demand (summer). Historic energy endowments would be expected to influence per capita emissions by a combination of high transportation costs and geographic specialization. In the former case, a state may exploit its local energy resources because the high cost of shipping fuels or transmitting electricity makes it uneconomic to do otherwise. In the latter case, even the decline in transportation costs over time may not substantially alter a region’s fuel mix since years of capital investments have already been made geared towards that region’s historic fuel mix (see Krugman 1991). The coal endowment variable would be expected to be positively associated with per capita emissions. 4.2 Regression Results 4.2.1 Quadratic Income Specifications Table 2, columns 1–4 present the results for the quadratic income specifications with the standard (production-based) carbon dioxide per capita emissions measure as the dependent variable. All specifications reveal the typical inverted-U shape in income—the linear term is positive while the squared term is negative. The income variables in all specifications are statistically significant at the 1 percent level. The coefficient estimates of 15.06

155

and –0.78 on the log income per capita and its square in column 1 yield an environmental Kuznets curve that peaks at an income of $14,708 (refer to the last row of table 2).17 In the OLS specifications, all of the additional control variables have the expected signs, although the weather variables are marginally statistically significant at best. In the FGLS specifications, all explanatory variables have their expected signs and all are statistically significant at the 1 percent level. The quadratic income specifications in columns 1–4 yield income-emissions relationships with similar incomes at emissions peaks, ranging from $14,708 to $16,840. To assess whether the non-income controls are associated with CO2 emissions, I have conducted F-tests of the hypothesis that the X vector variables—ln(CDD), ln(HDD), and lagged coal production—equal zero. The F-statistic for the comparison of the specifications (columns 1 and 2) is 42. Including these additional controls have little impact, however, on the estimated peak in the environmental Kuznets curve. As the bottom row of Table 2 illustrates, the incomes associated with the maximum of the EKC are fairly similar between specifications that only vary in terms of the inclusion of the X vector of variables ($14,708 versus $15,295 in columns 1 and 2; $20,389 versus $23,870 in columns 5 and 6). To place some of these estimated coefficients in context, consider their implications for expected per capita emissions across the states based on their 1999 values.18 North Dakota, with the most heating degree-days in 1999, would have approximately 1.2 tons per capita higher emissions than Florida, with the fewest heating degree-days, ceteris paribus. The variation in emissions associated with cooling degree days is negligible. Wyoming, with the most coal production in 1989 (the lagged value for 1999), would be expected to have about 2 tons per capita more emissions than states that do not produce coal. 19 Table 2, columns 5–8 present the regression results for the quadratic income specifications with the consumption-based CO2 per capita dependent variable. Similar to the results with the productionbased CO2 variable, these specifications all reveal an inverted-U in income, and all income variables are statistically significant at the 1 percent level. In the OLS specifications, all explanatory variables have their expected signs and are statistically sig-

0.080 (0.047)* [0.042]* 1.67x10–6 (2.65x10–7)*** [1.03x10–6]***







ln(CDD)

ln(HDD)

coal production

$15,295 ($900) [$2,579]

13.85 (0.77)*** [2.94]***

$16,449 ($336)



–0.43 (0.013)***

8.34 (0.26)***

$16,840 ($354)

1.45x10–6 (9.03x10–8)***

0.059 (0.0034)***

0.0082 (0.0015)***

–0.43 (0.016)***

8.45 (0.31)***

ln(CO2) Production

FGLS

(4)

$23,870 ($2,556) [$8,117]

2.44x10–6 (2.34x10–7)*** [4.08x10–7]***



$20,389 ($1,717) [$5,694]

0.078 (0.036)** [0.034]**

–0.0032 (0.018) [0.016]

–0.35 (0.035)*** [0.12]***





–0.41 (0.035)*** [0.13]***

7.14 (0.66)*** [2.30]***

ln(CO2cons) Consumption

ln(CO2cons) Consumption 8.12 (0.66)*** [2.38]***

OLS

(6)

OLS

(5)

$22,072 ($481)







–0.31 (0.011)*** –

6.23 (0.22)***

ln(CO2cons) Consumption

FGLS

(7)

$21,491 ($428)

9.27x10–7 (5.14x10–8)***

0.056 (0.0020)***

0.0086 (0.00093)***

–0.31 (0.0085)***

6.13 (0.16)***

ln(CO2cons) Consumption

FGLS

(8)

Notes: N = 1,920. Robust standard errors presented in parentheses and robust standard errors clustered by state in brackets for OLS specifications. Standard errors corrected for first-order serial correlation and for heteroskedasticity for FGLS specifications. *, **, *** statistically significant at 10, 5, and 1 percent levels. All regressions include state and year fixed effects.

$14,708 ($801) [$2,364]



–0.022 (0.024) [0.018]

–0.78 (0.040)*** [0.15]***

ln(y)2

income at emissions peak (1999$)



0.72 –(0.041)*** [0.16]***

15.06 (0.76)*** [2.78]***

ln(CO2) Production

ln(y)

ln(CO2) Production

ln(CO2) Production

FGLS

Dependent Variable

OLS

(3)

OLS

(2)

Specification

(1)

Table 2 Regression Results for Quadratic Income Specifications

156 Repsol YPF-Harvard Kennedy School Fellows Research Papers

Joseph E. Aldy

157

nificant at the 1 percent level, with the exception of the weather variables.20 The income coefficient estimates of 8.12 and –0.41 in column 5 yield an estimated income at the peak of the consumptionbased EKC function of $20,389. The consumptionbased EKCs have incomes at emissions peak ranging from $20,389 to $23,870. In the FGLS specifications, all explanatory variables have their expected signs and are statistically significant at the 1 percent level. The per capita incomes associated with the environmental Kuznets curve peaks are higher for the consumption-based carbon dioxide emissions specifications than for the standard carbon dioxide emissions specifications. This indicates that higher income states may experience a decline in per capita carbon dioxide emissions, not because individuals consume less carbon-based goods in those states than before, but because they consume carbon-based goods (in this case, electricity) produced in lower income states. These differences in the income levels associated with EKC peaks between the two specifications are not inconsequential. The per capita incomes associated with the peak of the consumption-based carbon dioxide emissions EKC range up to 56 percent higher than the incomes associated with the production-based carbon dioxide emissions specification. The differences between the production- and consumption-based CO2 environmental Kuznets curves are statistically significant. To statistically evaluate the properties of the income-emissions relationships for the production- and consumption-based measures in an integrated regression framework, I undertook multivariate least squares regressions in which the dependent variables vector includes production- and consumption-based CO2 per capita: Table 3 Test Statistics Comparing Production- and Consumption-Based CO2 Environmental Kuznets Curves

Specification State and Year Effects X Vector F-Statistic, H 10 F-Statistic, H 20

(1)

(2)

OLS Yes No 112.54 47.50

OLS Yes Yes 86.70 40.52

Notes. N = 1,920. Based on multivariate regressions in which the dependent variables vector includes ln(CO2) and ln(CO2cons).

This two-equation system allows for correlation in the errors across the two equations, and a robust estimated variance-covariance matrix.21 Table 3 shows the results of F-tests applied to two specifications of this system: one with and one without the X vector of controls. These tests evaluate two hypotheses: Equality of Parameter Estimates:

for k = 1, 2, and Equality of Estimated EKC Peaks:

With F-statistics of 87 and 113 for the first hypothesis and 40 and 48 for the second hypothesis, these results strongly recommend rejecting the null hypotheses. The differences in the magnitudes of the estimated production- and consumption-based CO2 EKC peaks are economically and statistically significant. Employing the standard, productionbased carbon dioxide environmental Kuznets curve for long-term emissions forecasts could result in a substantially different forecast for the U.S. states than using the consumption-based curve since per capita emissions would continue to increase for a longer period of time with the latter function.22 4.2.2 Cubic Spline Income Specifications As an alternative to quadratic income specifications, a cubic spline function of income can also characterize the emissions-income relationship. Employing a cubic spline specification allows for a second approach to investigate the potential effects of emissions-intensive trade on estimation of the carbon dioxide per capita environmental Kuznets curve and to begin assessing the robustness of the EKC estimates. As Schmalensee et al. (1998) note, a spline specification is more flexible than power functions and may better describe the EKC relationship. I have conducted a series of OLS regressions with the cubic spline function divided into as few as 2 and as many as 11 knots for both the production- and consumption-based measures

Repsol YPF-Harvard Kennedy School Fellows Research Papers

158

Figure 1 Production- and Consumption-based Carbon Dioxide Environmental Kuznets Curves from Cubic Spline Regressions Production CO2 per capita

X

Consumption CO2 per capita

Tons of carbon

6.25

3.00 5,000

40,000 Income per capita (1999$)

Notes: Functions derived from regressions of the natural logarithm of per capita emissions on a ten-knot cubic spline function of the natural logarithm of per capita income and state and year fixed effects. The knots are placed at equally-spaced quantiles of the income distribution. The cubic spline function specification fits a unique cubic function of income between each knot subject to the constraint that the estimated function is everywhere smooth, including at the knots. The fitted values for the 1,920 income per capita observations are presented.

of emissions, and including state and year fixed effects.23 Based on F-tests evaluated at a 5 percent significance level, I could reject specifications with fewer than 8 knots for the production-based regressions and those with fewer than 10 knots for the consumption-based regressions. Figure 1 presents the estimated EKCs for the production- and consumption-based measures based on 10-knot cubic spline specifications.24 The figure shows a striking divergence in the income-emissions relationship between the production- and consumption-based measures at higher incomes. The two relationships track each other quite closely over the $7,000-$14,000 range, at which point the consumption-based measure crosses the production-based measure and remains at relatively high levels (between about

5.5 and slightly more than 6.0 tons per person) through the rest of the sample range for income. In contrast, the production-based EKC declines substantially with increasing per capita income. At the maximum income level in the sample (about $39,000), the production-based EKC has a per capita emissions level (3.1 tons per capita) that is just slightly more than half of its peak (5.7 tons per person, occurring at about $16,000). The consumption-based EKC however, has a per capita emissions level (5.5 tons per capita) that is about 90 percent of the value associated with its peak (6.1 tons per person, occurring at about $20,500). Per capita carbon emissions with the consumptionbased measure are up to 75 percent greater than the production-based carbon emissions measure at high incomes. While the results from the cubic

Joseph E. Aldy

159

spline specifications confirm the quadratic income results that emissions peak at higher incomes with the consumption-based measure, these estimated EKCs also show that per capita emissions decline at a much more gradual rate after peaking on a consumption basis than on a production basis. The consumption-based results could be more consistent with a “peak and plateau” shape than the more typical inverted-U EKC shape. 4.2.3 Heterogeneous Slopes Specifications The previous two sets of results, and the vast majority of the empirical EKC literature, assume that all economies follow a common income-emissions relationship. While the typical EKC (panel) regression only allows economies’ emissions to vary in terms of level (fixed) effects, following List and Gallet (1999), I have relaxed the assumption of common slope terms (β ’s) for all states by allowing these parameters to vary by state. I have conducted regressions with both the production-based CO2 and consumption-based CO2 measures specified as functions of state-specific quadratic income, state fixed effects, and year fixed effects. This heterogeneous slopes model provides a third approach both for assessing the hypothesis that emissions-intensive trade affects the estimated EKC and for determining the robustness of the earlier results. Table 4 presents summary statistics of these two regressions. Most states follow an inverted-U

Table 4 Summary Statistics for the Heterogeneous Slopes Regression Models ln(CO2)

ln(CO2cons)

Number of states with inverted-U income-emissions relationship

44

40

Number of states with statistically significant estimated EKC peaks

40

37

Median EKC peak for states with inverted-U relationships

$15,844

$18,231

F-Statistic, H 30

53.78

54.52

F-Statistic, H 40

56.45

52.43

Dependent Variable

Notes: Regressions specified as state-specific quadratic income, and include state and year fixed effects. These regressions estimated with robust standard errors.

shape for the production-based CO2 measure, although fewer of the states have such a relationship for consumption-based CO2 per capita. Forty states have a statistically significant estimated EKC peak (and all estimates fall within their states’ respective income range) for standard CO2 per capita, and 37 states do for consumption-based CO2 per capita. The median EKC peak income for the consumption-based measure is about 15 percent greater than the median production-based measure. This more general model that allows for heterogeneity in the slope coefficients yields results that question the typical assumption of common parameter estimates for all states. I evaluated the following two hypotheses: Equality of Parameter Estimates:

Equality of Estimated EKC Peaks:

F-tests of the equality of the income parameters reject the hypothesis of a common income-emissions relationship among the states. Likewise, tests of equality of the estimated EKC peaks also recommend rejecting the hypothesis that the states can be characterized by one EKC function (refer to the last two rows of Table 4). These results are consistent with the findings of List and Gallet (1999) for SO2 and NOX environmental Kuznets curves for the U.S. states. Recent research by Brock and Taylor (2004) provides some insights into economy-specific EKCs. State-specific EKC analyses may be appropriate in light of this finding, which motivates the time series analyses in the next section. 4.2.4 State-Specific Time Series Specifications State-specific EKC regressions can provide several checks to the preceding analyses. First, every state has income per capita observations on both the upward and downward sloping regions of the emissions-income inverted-U function estimated with the production-based carbon dioxide per capita measure in Table 2. If this reduced-form relationship estimated from a panel is robust, it should be evident individually among these

160

states. Second, the analysis with heterogeneous slopes indicates that the relationship may not be identical for all economies. A better understanding of which economies experience such trends in emissions as their incomes grow could further inform the theory that motivates the empirical work. Third, several papers in this literature have questioned the empirical methods that fail to account for possible stochastic trends in emissions and income data (see Perman and Stern 2003, Stern 2004). Focusing on state-specific time series will allow for an explicit investigation of these properties for the state-level data. Before undertaking state-specific regressions, I conducted several time series tests of the income and emissions data. First, I tested for whether the state-specific natural logarithm of per capita income and natural logarithm of per capita CO2 emissions time series are nonstationary (stochastically trending) with the Elliott et al. (1996) GLS version of the Dickey-Fuller test. As a test with higher power than the augmented Dickey-Fuller test, this GLS version is more likely to reject the null hypothesis of a unit root against the alternative of a stationary distribution when the root is close to but less than one. For only one state do the data suggest rejecting the null of a unit root at the five percent significance level for the log per capita income time series, and only four additional states at the ten percent significance level. Likewise, I reject the null for only one state for the log per capita carbon emissions time series at the five percent level, and for only two states at the ten percent level. To determine if the data are integrated of order one, I conducted the same tests on the first differences in log per capita emissions and incomes. For all but one state, the tests recommend rejecting the null hypothesis of a unit root for the changes. The state-specific log per capita income and emissions data appear to be integrated of order one time series. This suggests the need to evaluate for whether the state-specific time series are cointegrated to determine if the reduced-form EKC regression analyses are actually spurious.25 If the per capita emissions and income time series for a given state share a common stochastic trend (are cointegrated), then regression analysis can still be conducted to estimate consistent parameters on the income variables. Since the cointegration vector is not known, I made a prelimi-

Repsol YPF-Harvard Kennedy School Fellows Research Papers

nary estimate of the cointegration vector and tested for whether the emissions and the income and income squared variables share a common trend with the Engle-Granger Augmented Dickey-Fuller test (EG-ADF; Engle and Granger 1987). First, I conducted an OLS regression for each state of the natural logarithm of per capita emissions on the natural logarithm of per capita income and the square of the natural logarithm of per capita income. Second, I constructed the predicted residuals from these regressions and tested these for a unit root with an augmented Dickey-Fuller test. If the data reject the hypothesis of a unit root, then the estimated cointegrating vector (the parameter estimates on the income variables) yields a stationary relationship between the emissions and income variables. Following Stock (1994), I used the Akaike Information Criterion to determine the lag structure of the second stage test of the residuals. The results of the cointegration tests showed that few states have cointegrated data. Using a 10 percent critical level cutoff, I could reject the hypothesis of a unit autoregressive root for the residual for nine (eleven) of 48 states for the production-based (consumption-based) analyses.26 This suggests that the income-emissions relationship estimated through OLS could be spurious for many states, and may also be in a panel context (see Perman and Stern 2003 for similar findings for an international EKC analysis of sulfur dioxide emissions). To provide consistent estimates of the parameters and their associated standard errors for the states with cointegrated data, I estimated statespecific dynamic ordinary least squares (DOLS) regressions following Stock and Watson (1993):

where Yt represents the natural logarithm of income per capita and its square, ∆Yt–j represents lags, leads, and current values of the changes in the two income variables, p represents the number of lags/leads (assumed to be 2 in this analysis), and β and δ are parameter vectors to be estimated. The t-statistics constructed from DOLS are based on Newey-West standard errors (I have assumed a two-period lag structure). The β parameter vector can be considered the long-run effect of the income variables on the emissions variable (assuming that the income variables are exoge-

Joseph E. Aldy

nous). I compare these with OLS regressions of the natural logarithm of emissions on a constant, the natural logarithm of income per capita and its square, with Newey-West standard errors assuming a two-period autocorrelation structure. Appendix Tables 1 and 2, present all DOLS regression results, even for states with data that do not reject the null of no cointegration in the EG-ADF tests, and all OLS regression results. Thirty-two states have inverted-U EKCs that have estimated peaks that are statistically different from zero (p βV cn(Ω kt;θ )] using functional form assumptions on profits and distributional assumptions on µit:

(23) This equation can be rewritten in vector form as:

. Substituting in the

empirical average

for M n, we can solve for a

fixed point

.

6.2

Step 2: Generalized Method of Moments

After obtaining estimates of the continuation values in the first step of my econometric estimation technique, I estimate the parameters θ in the second step using generalized method of moments. To do so, I first substitute my estimates

(24)

and for the continuation values into equations (10) and (5) for the exploration and development probabilities to get the predicted probabilities:

(25)

(29)

Moreover, this equation can be inverted to:

Once again, empirical averages and can be used to estimate M n and g e, respectively, so that our inversion (I) estimate for becomes:

and (29)

(26) Fixed point for policy function (G) This estimator uses the equation for the continuation value derived after solving for both the expected truncated exploration profits E[π e(Ω kt, µit;θ )| π e(Ω kt, µit;θ ) > βV cn(Ω kt;θ )] and the policy function g e(·).14 The intuition is similar to that for the development stage, so I won’t repeat it here.

The moments I construct involve matching the probabilities of exploration and development predicted by my model, as given by equations (30) and (20), with the respective empirical probabilities in my data. Similarly, I construct moments that match the profits for tracts that develop predicted by my model with the actual profits observed in the data. For the exactly identified case, my moment

210

function Ψ (zikt, θ) is:

where I ite is an indicator for whether exploration began on tract i at time t; I itnot_yet_e is an indicator variable for tract i not yet being explored before period t; I dit is an indicator for whether development began on tract i at time t; I ite_not_yet_d is an indicator variable for tract i being explored but not yet developed before period t; and where π di is the profits earned on tract i after development. For the overidentified case, my moment function is:

and the optimal weighting matrix as specified by Chamberlain (1987) is used. Endnotes 1. [email protected] 2. In my broad definition of an externality, I say that an externality is present whenever a non-coordinated decision by individual firms is not socially optimal. 3. If firms are subject to a lease term by the end of which they must begin exploratory drilling, or else relinquish their lease, then the information externality would result in too little exploration at the beginning of the lease term and duplicative drilling in the final period of the lease (Hendricks & Porter, 1993; Porter, 1995). In contrast, the optimal coordinated plan would entail a sequential search in which one tract would be drilled in the first period and, if productive, a neighboring tract is drilled in the next (Porter, 1995). 4. As will be described in this paper, strong assumptions need to be made in my model in order for these policy evaluations to be valid. In particular, I need to assume that the equilibrium does not change when the policy environment changes. The development of a structural model that enables policy evaluations to be made without these strong

Repsol YPF-Harvard Kennedy School Fellows Research Papers

assumptions will be the subject of future work. 5. In addition to changing the lease term and/or tract size, other possible modifications to the current federal offshore lease auction mechanism might include encouraging unitization of exploration programs, perhaps by limiting the amount of nonunitized acreage that a firm can possess; requiring firms to make their seismic reports publicly available (Hendricks & Kovenock, 1989); changing the quantity or location of the tracts offered in each lease sale; using multi-unit auctions; or making the contractual environment more conducive to coordination. 6. The validity of policy evaluation is subject to the caveats outlines in a previous footnote. 7. I hope eventually to further extend the Pakes et al. framework by using both continuous and discrete variables. In contrast, the state variables in the framework developed by Pakes et al. must be discrete. 8. I currently do not have firm fixed effects or correlations between tracts owned by the same firm, but should think about how I might incorporate them. 9. Firms conduct and analyze seismic studies in order to help them decide whether or not to begin exploratory drilling (John Shaw, personal communication, April 18, 2003; Bob Dye, personal communication, January 21, 2004; Jon Jeppesen, personal communication, January 21, 2004; Mark Bauer, personal communication, January 21, 2004; Billy Ebarb, personal communication, January 22, 2004). 10. The assumptions that both types of shocks are i.i.d. and independent of each other, while restrictive, is needed in order for the estimation technique used in this paper to work. If either type of shock were serially correlated (or if, at the extreme, there were tract fixed effects), then firms would base their decisions not only on the current values of the state variables and of their shocks, but also on past values of the state variables and shocks as well. The state space would then be too large. If the distribution of the post-exploration shock ε it depended on the realization of the pre-exploration shock µit (e.g., the µit at the time of exploration), then µit would be a state variable in the development stage of production. As a consequence, the econometrician would need to observe µit, which she does not. In future work I hope to develop econometric techniques that allow these assumptions to be relaxed. 11. Thus, as with Harsanyi’s (1973) purification theorem, a mixed distribution over actions is the result of unobserved payoff perturbations that sometimes lead firms to have a strict preference for one action, and sometimes a strict preference for another. 12. A fourth estimator would be to solve for the policy function but not for expected truncated profits.

C.-Y. Cynthia Lin

13. The nonparametric (N) estimator does not work for estimating Vcn because one cannot estimate E[π e(Ω kt, µit;θ )|π e(Ω kt, µit;θ ) > βV cn(Ω kt;θ )]. 14. Another estimator would be to solve for the policy function but not for expected truncated profits.

References Chamberlain, G. (1987). Asymptotic efficiency in estimation with conditional moment restrictions. Journal of Econometrics, 34, 305–334. Dixit, A., & Pindyck, R. (1994). Investment under uncertainty. Princeton, NJ: Princeton University Press. Harsanyi, J. (1973). Games with randomly disturbed payoffs. International Journal of Game Theory, 2, 1–23. Hendricks, K., & Kovenock, D. (1989). Asymmetric information, information externalities, and efficiency: The case of oil exploration. RAND Journal of Economics, 20 (2), 164–182. Hendricks, K., & Porter, R. (1993). Determinants of the timing and incidence of exploratory drilling on offshore wildcat tracts. NBER Working Paper Series

211

(Working Paper No. 4605). Cambridge, MA. Hurn, A., & Wright, R. (1994). Geology or economics? Testing models of irreversible investment using North Sea oil data. Economic Journal, 104 (423), 363–371. Libecap, G., & Smith, J. (1999). Regulatory remedies to the common pool: The limits to oil field unitization. Mimeo, University of Arizona and Southern Methodist University. Libecap, G., & Wiggins, S. (1985). The influence of private contractual failure on regulation: The case of oil field unitization. The Journal of Political Economy, 93 (4), 690–714. Paddock, J., Siegel, D., & Smith, J. (1988). Option valuation of claims on real assets: The case of offshore petroleum leases. Quarterly Journal of Economics, 103 (3), 479–508. Pakes, A., Ostrovsky, M., & Berry, S. (2004). Simple estimators for the parameters of discrete dynamic games (with entry/exit examples). Mimeo. Harvard University.

12

Estimating Annual and Monthly Supply and Demand for World Oil: A Dry Hole? C.-Y. Cynthia Lin1 Department of Economics, Harvard University

chosen were both strong and credible. However, although the efficiently identified monthly supply and demand curves were consistent with economic theory in the cases of world demand, non-OPEC demand and two specifications for supply, this was not the case for either OPEC demand or for most specifications for supply. The assumptions of a static and perfectly competitive world oil market thus appear to be unrealistic, especially in modeling oil supply.

Abstract This paper uses instrumental variables and joint estimation to obtain efficiently identified estimates of aggregate supply and demand curves for world oil under the assumptions of a static and perfectly competitive oil market. When annual data spanning 1965–2000 were used, the instruments chosen, while credible, were weak, and, as a consequence, neither supply nor demand was identified. In contrast, with monthly data spanning 1981–2000, the instruments

1.

Introduction

One of the most important resources on the planet today is oil. Indeed, oil is a form of power, not only because it is a primary source of the energy needed to power modern industrialized society (Yergin, 1992), but also because its possession itself is a source of power. Oil not only fuels our cars, heats our homes and runs our factories, but also drives national economic, political and military policy around the world. Because oil is such a valuable resource, academics, businesspeople and policymakers alike have spent an inordinate amount of time and energy studying the oil industry. Yet, although their efforts have yielded many important insights, models and theories, the world oil market still remains somewhat a mystery, and many questions remain unanswered. As with any other commodity, one of the fundamental questions economists would want to ask and answer about oil is: “How do we model the world market for oil?” In particular, what deter-

This paper was prepared in residence under a Repsol YPFHarvard Kennedy School Fellowship. I would like to thank Gary Chamberlain for his advice and guidance throughout this project. This paper also benefited from discussions with William Hogan, Michael Kennedy, Howard Stone, and Martin Weitzman. Some of the data used in this study were acquired with the help of Brian Greene and with funds from the Littauer Library at Harvard University. I an indebted to Bijan MossavarRahmani (Chairman, Mondoil Corporation) and William Hogan for arranging for me to visit Apache Corporation’s headquarters in Houston and a drilling rig and production platform offshore of Louisiana, and for their support of my research, and I thank the Repsol YPF-Harvard Kennedy School Energy Fellows Program for providing travel funds. I thank Mark Bauer (Reservoir Engineering Manager, Apache), Robert Dye (VP, Apache) Steve Farris (President, CEO & COO, Apache), Richard Gould (VP, Wells Fargo), Paul Griesedieck (Manager, Apache), Thomas Halsey (Corporate Strategic Research, ExxonMobil Research and Engineering), Becky Harden (Land Manager, Apache), David Higgins (Director, Apache), Jon Jeppesen (Sr. VP, Apache), Adrian Lajous (President, Oxford Institute for Energy Studies), Kenneth McMinn (Offshore District Production Manager, Apache), Kregg Olson (Director, Apache), and Bob Tippee (Editor, Oil and Gas Journal) for enlightening discussions about the petroleum industry. I thank Derrick Martin for flying me offshore by helicopter, Mike Thibodeaux for giving me a tour of the production platform, Joey Bridges for giving me a tour of the drilling rig, and especially Billy Ebarb (Production Superintendent, Apache) for accompanying me throughout the entire offshore trip. Ivan

Dong provided useful suggestions on how to acquire the R statistical software package. I received financial support from an EPA Science to Achieve Results (STAR) graduate fellowship and a National Science Foundation (NSF) graduate research fellowship. All errors are my own.

213

214

mines the supply for oil, what determines the demand for oil, and by what equilibrium process are oil prices and quantities determined? Economic theory has much to say about how commodity markets might function. The most basic economic model of a market, as first envisioned by Adam Smith in 1776, posits that, under assumptions of perfect competition, the market price acts to equilibrate supply and demand (Mankiw, 1998). In addition to assuming perfect competition and price-taking on the part of both producers and consumers, this most basic model is also agnostic about the time period over which transactions take place, and, in particular, assumes that there are no dynamic considerations linking the static markets from one time period to the next. Ever since Adam Smith introduced the notion of a perfectly efficient market, economists have developed an impressive corpus of theoretical models to explain how markets might function when one or more of Smith’s simplifying assumptions are relaxed. While the economic theory of markets is fairly well developed, however, plausible empirical applications of this theory to actual real-world commodities are less so. As with most fields in economics, empirical studies lag behind the theory, not only because theoretical models can serve as the motivation behind empirical studies, but also because econometric techniques that confront the myriad statistical and identification problems that arise in any attempt to apply theory to actual data must be developed before any credible empirical application can take place. One central econometric question in empirical studies of markets is how to infer the structure of supply and demand from actual observations of equilibrium prices and quantities (Manski, 1995). Indeed, it owed in part to the desire of economists to analyze competitive markets that statistical models for estimating and identifying simultaneous equations were first developed (Angrist, Graddy & Imbens, 2000). To this day, econometricians are still developing techniques to analyze the functioning of markets and to tackle the identification problem that plagues such analyses. Although there have been countless empirical studies of the world oil market, not one has produced a satisfactory model that adequately explains historical data, much less accurately predicts future developments (William Hogan, personal communication, February 23, 2004).

Repsol YPF-Harvard Kennedy School Fellows Research Papers

Moreover, the preponderance of these studies were conducted over two decades ago (see e.g., Adelman, 1962; Berndt & Wood, 1975; Gately, 1984; Hausman, 1975; Kennedy, 1974; Nordhaus, Houthakker & Sachs, 1980). As a consequence, while frontier econometric methods have been used to estimate the basic economic model of static competitive markets for a variety of commodities, including the demand for fish (Angrist et al., 2000) and the labor supply of stadium vendors (Oettinger, 1999), new methods have yet to be applied to the market for oil. In this paper, I use a variety of econometric methods to estimate supply and demand curves for oil under the simplifying assumptions of a static and perfectly competitive world oil market. This paper makes two main contributions. First, by re-examining the timeless issue of oil supply and demand estimation using updated data and more recent simultaneous equation estimation techniques, I innovate upon the existing literature on the world oil market. Second, results of my econometric model of oil supply and demand under the simplifying assumptions of a perfectly competitive and static world oil market is in part a test of whether these simplifying assumptions are indeed correct. By providing a benchmark against which one can compare more complicated econometric models incorporating oligopoly behavior, dynamics, or both, an estimation of the world oil market using the most basic but perhaps unrealistic simplifying assumptions enables one to sense the tradeoffs that might occur as one moves toward the more complex—but also more realistic—models. I hope to develop these more complex models in future work. According to my results, while monthly world oil demand, monthly oil demand in countries that are not part of the Organization of Petroleum Exporting Countries (OPEC), and two specifications for monthly oil supply appear consistent with static perfect competition, monthly OPEC oil demand and most specifications for monthly oil supply do not. Thus, in the latter cases, the simplifying assumptions of a static and perfectly competitive oil market appear to be unrealistic. The balance of the paper proceeds as follows. In Section 2, I present my model of the world oil market and explain the identification problem that arises in empirical analyses of supply and demand. In Section 3, I outline the econometric

C.-Y. Cynthia Lin

methods I use to address this identification problem. I describe my data set in Section 4. My results are presented in Section 5. Section 6 concludes. 2.

A Model of Oil Supply and Demand

In this section, I present my model of world oil supply and demand, and explain the identification problem that arises in its estimation. More thorough treatments of the identification problem that arises in empirical analyses of supply and demand are given by Angrist et al. (2000), Goldberger (1991), and Manski (1995); the notation and exposition that follows were inspired in part by these sources. 2.1 The General Framework Suppose there are T oil markets isolated in time and indexed by t = 1,...,T. For each market t, let pt denote the price of oil, qt denote the quantity of oil transacted and xt denote a vector of covariates characterizing the market. For each market t, the market demand function q dt (·) gives the quantity of oil that price-taking consumers would purchase, while the market supply function q ts (·) gives the quantity of oil that price-taking firms would offer, both as functions of price. Markets are assumed to clear, which means that the transaction (pt,qt) is assumed to be an equilibrium outcome. In other words, for all markets t, the price pt acts to equate supply and demand: (1) Markets vary in their values of (q dt(·),q ts(·),pt, qt,xt). For each market t, the econometrician can only observe the equilibrium price pt, the equilibrium quantity qt and the covariates xt, but cannot observe either the demand function q dt(·) or the supply function q ts(·). Econometric analysis therefore seeks to learn about the supply and demand functions when only equilibrium transactions and covariates are observed. The identification problem that arises when observations of market transactions are used to infer the structure of supply and demand is called the simultaneity problem. More formally, the simultaneity problem is as follows. Econometricians would like to infer the distribution Pr(q dt(·),q ts(·)|xt) of demand and supply

215

functions conditional on the covariates xt. However, they can only observe the variables (pt, qt,xt). If the observations (pt, qt,xt) were obtained by a random sampling process, then the distribution Pr(pt, qt,xt) of the observed variables could be inferred. The simultaneity problem is that, although the econometrician can infer Pr(pt, qt,xt), knowledge of Pr(pt, qt,xt) is not sufficient for identifying Pr(q dt(·),q ts(·)|xt). Thus, it is possible that neither supply nor demand is identified. 2.2 A Linear Market Model In my study, I assume that both demand and supply functions are linear with fixed coefficients and additive residuals. Though perhaps unrealistic, the linearity and additivity assumptions simplify the estimation techniques and provide a useful benchmark for assessing whether they need to be relaxed in future work.2 The structural form of my model is given by:

which simplifies to: (2) (3) The demand equation (2) and the supply equation (3) are the structural equations of my linear oil market model. Because economic theory predicts that demand curves should be downward-sloping while supply curves should be upward-sloping, we expect that β pd ″ 0 and β ps ≥ 0. Solving the structural equations (2) and (3) for price and quantity as functions of the covariates, one obtains the following reduced-form equations for my linear oil market model: (4) (5) Econometric analysis seeks to efficiently identify the structural parameters (β pd,β xd,β ps ,β xs ). Unfortunately, estimating the demand equation (2) and the supply equation (3) separately by ordinary least squares (OLS) will not efficiently identify these structural parameters, for two reasons.

Repsol YPF-Harvard Kennedy School Fellows Research Papers

216

The first problem is a lack of identification. Because prices are endogenously determined in the supply-and-demand system, equation-byequation ordinary least squares estimators of the coefficients (β pd,β ps ) on price will be biased and not consistent (Goldberger, 1991). Thus, unless one uses instruments for price, these coefficients will neither be identified nor consistent.3 This lack of identification has not been addressed in most of the empirical work on the oil market to date (see e.g., Kennedy, 1974; Nordhaus, Houthakker & Sachs, 1980). The second problem with equation-by-equation ordinary least squares is a lack of efficiency. If there are restrictions on the parameters in the model, then joint estimation of the demand and supply equations will be more efficient than equation-by-equation OLS is (Goldberger, 1991; Ruud, 2000). Thus, equation-by-equation OLS will not efficiently identify the structural parameters because it yields estimates that are neither identified nor efficient. I now turn to describing the econometric methods I will use to improve upon equation-byequation OLS. 3.

Methods for Efficient Identification

As explained above, equation-by-equation OLS suffers from both an identification problem and an efficiency problem. In order to address the identification problem, I will use instrumental variables techniques that exploit exclusion restrictions on both the supply and demand equations. In particular, I will assume that the vector of covariates xt can be decomposed into four components:

where the demand shifters x dt are exogenous covariates that shift the demand curve but not the supply curve; where the supply shifters x st are exogenous covariates that shift the supply curve but not the demand curve; where the endogenous covariates x nt may enter the structural equation for supply or demand, or both; and where the market controls x ct are exogenous covariates that affect both demand and supply. into the Substituting structural equations (2) and (3) for demand and supply, respectively, one gets:

(6)

(7)

Formally, my exclusion restriction is the following: Assumption 1. (Exclusion) In the expanded structural equations (6) and (7) for demand and supply, Under Assumption 1, the structural model can be rewritten as: (8)

(9)

With the above exclusion restriction, I can now identify each equation by using the exogenous variables excluded in that equation as instruments (Manski, 1995). In particular, because the exogenous demand shifter x dt do not affect supply except through their effect on price, they can be used as instruments for price in the supply equation. Similarly, because the exogenous supply shifters x st do not affect demand except through their effect on price, they can be used as instruments for price in the demand equation. Exogenous market controls x ct can serve as instruments for both equations. My vector of instruments zt is therefore given by zt = (x dt,x st,x ct). So that these proposed instruments zt are indeed valid, I also make the following additional assumptions: Assumption 2. (Correlation) The instruments zt have a non-zero correlation with price pt. Assumption 3. (Monotonicity) The instruments zt have a monotonic effect on price pt. Under Assumptions 1–3, the instruments zt can be used to obtain consistent and identified estimates of the structural parameters. Analogous

C.-Y. Cynthia Lin

arguments and assumptions can be used for why exogenous demand shifters, supply shifters and market controls might be valid instruments not only for price, but also for any endogenous covariates x nt as well. Thus, in order to address the identification problem, I use instrumental variables techniques that exploit exclusion restrictions on both the supply and demand functions. Unfortunately, if the exclusion restriction in Assumption 1 holds, then efficiency becomes an issue. As mentioned above, the second problem with equation-by-equation OLS is that if there are restrictions on the parameters in the model, then equationby-equation OLS would be inefficient, and joint estimation of the equations would be preferred. More generally, in the presence of any parameter restrictions, joint estimation will be more efficient than its equation-by-equation analog (Goldberger, 1991; Ruud, 2000). Because Assumption 1 imposes exclusion restrictions on the structural parameters, joint estimation of the structural equations should be used to improve efficiency. In this paper I use several estimation methods to obtain estimates that are identified, efficient, or both. First, as a benchmark, I estimate the demand equation (8) and supply equation (9) separately by OLS. As explained above, these estimates are neither identified, nor consistent, nor efficient. Second, to enhance the efficiency of my OLS estimates, I estimate the structural equations (8) and (9) jointly. I thus treat the system of simultaneous equations as seemingly unrelated regressions (SUR) that I can estimate using feasible generalized least squares. Under Assumption 1, estimation of the SUR using feasible generalized least squares is more efficient than OLS. However, though SUR estimation may be efficient, it erroneously assumes that all the dependent variables in the structural equations, including price, are exogenous. Hence, SUR still lacks identification. In order to identify the price coefficients, the third technique I use is that of equation-by-equation two-stage least squares (2SLS). Each of the two structural equations (8) and (9) is estimated using the instruments zt. The estimates obtained via 2SLS are identified and consistent.4 However, although the estimates yielded by 2SLS are identified, they are not efficient because, in estimating each equation individually, 2SLS does not make use of all the available information.5 Owing to the

217

cross-equation restrictions imposed by Assumption 1, estimating the equations jointly can enhance efficiency. In order to address both the identification and the efficiency issues, the fourth estimation method I employ is that of three-stage least squares (3SLS). In 3SLS, not only are instruments used to help identify the structural parameters, but the equations (8) and (9) are also jointly estimated via generalized method of moments to improve efficiency. 3SLS is more efficient than its equation-byequation analog, 2SLS, because 3SLS uses all the available information at one time.6 Thus, 3SLS estimates are both identified and efficient. In this paper I therefore use a variety of methods (OLS, SUR, 2SLS, and 3SLS) to estimate the world supply and demand for oil under the assumptions of a perfectly competitive static oil market. If the theoretical and econometric assumptions of my model are correct, then the 3SLS estimates should be identified, consistent and efficient. I now proceed to describing the data used in my study. 4.

Data

In my empirical analysis of the world oil market, I use two data sets: an annual data set spanning the years 1965–2000 and a monthly data set spanning the years 1981–2000. Because the preponderance of empirical studies of the world oil market were conducted over 20 years ago, both my annual and monthly data sets include newer data not used in previous work on the topic. 4.1 Annual Data (1965–2000) Table 1 provides summary statistics for the variables in my annual data set. For oil price pt, I use the real annual spot price for crude oil, averaged over the Brent, Dubai, and West Texas Intermediate (WTI) prices. This average price time series was obtained from the World Bank and deflated to 1982–1984 U.S. dollars using the consumer price index (CPI). For oil quantity qt, I use two possible measures: world oil consumption as reported by BP and world petroleum consumption as reported by the U.S. Department of Energy. For comparison, summary statistics are also provided for oil and petro-

Repsol YPF-Harvard Kennedy School Fellows Research Papers

218

Table 1 Summary statistics for annual data (1965-2000) Variable

mean

s.d.

min

max

trend

Price real spot price for crude oil: average of Brent, Dubai & WTI (1982–1984$/barrel)

16.07

11.18

3.12

44.75

0.15 (0.18)

world oil consumption (million barrels/day)

58.88

11.37

31.23

75.25

world petroleum consumption (million barrels/day)

58.78

12.88

26.20

76.90

OECD oil consumption (million barrels/day)

39.25

6.01

23.23

47.61

OECD petroleum consumption (million barrels/day)

39.10

6.40

22.44

47.88

0.99*** (0.07) 1.06*** (0.10) 0.46*** (0.06) 0.49*** (0.06)

real world GDP (trillion 1982–1984$)

13.00

4.13

6.22

19.16

real GDP in Middle East and North Africa (trillion 1982–1984$)

0.30

0.12

0.08

world population (billions)

4.65

0.83

3.31

world electricity production (trillion kwh)

8.76

3.93

2.38

electricity production from oil (%)

19.03

8.16

7.82

electricity production from natural gas (%)

11.32

3.06

7.48

world crude oil reserves (billion barrels)

739.20

214.90

341.27

OPEC crude oil reserves (billion barrels)

539.61

181.88

254.90

Quantity

Covariates 0.38*** (0.02) 0.48 0.01*** (0.00) 6.05 0.08*** (0.00) 15.30 0.37*** (0.00) 33.09 –0.35** (0.12) 17.45 0.13** (0.04) 1034.27 19.31*** (1.13) 802.48 16.09*** (1.07)

Notes: The trend is the coefficient on year when the variable is regressed on year and a constant (standard error in parentheses). Significance codes: * 5% level, ** 1% level, and *** 0.1% level.

leum consumption aggregated over only those countries in the Organization for Economic Cooperation and Development (OECD). For covariates xt, I use data on the following variables: real world gross domestic product (GDP), real GDP for the Middle East and North Africa, population, total electricity production, electricity production from oil, electricity production from gas, world oil reserves, and Organization of Petroleum Exporting Countries (OPEC) oil reserves. GDP, population and electricity production data were obtained from the World Bank Group World Development Indicators (WDI) online database. Reserve data were obtained from the Oil and Gas Journal. GDP data were deflated to 1982–1984 U.S. dollars using the CPI.

As can be seen in the last column of Table 1, most variables exhibit a significant positive trend over 1965–2000, with two exceptions. The first exception is oil price, which has no significant trend. The trendless nature of oil price over time is in accordance with many empirical studies (see Krautkraemer, 1998, & references therein).7 The second exception is the percent electricity production from oil, which has a significant downward trend. While the electricity production from oil is declining, however, that from natural gas is increasing. Over the period 1965–2000, the world thus appears to be substituting away from oil and towards natural gas as its source of electricity. Table 2 presents the correlations among my two measures of world quantity and their OECD

C.-Y. Cynthia Lin

219

Table 2 Correlations between various measures of annual quantity

oil consumption, world oil consumption, OECD petroleum consumption, world petroleum consumption, OECD

oil consumption, world

oil consumption, OECD

petroleum consumption, world

petroleum consumption, OECD

1.00 0.96 0.93 0.97

1.00 0.88 1.00

1.00 0.94

1.00

Note: Consumption is measured in million barrels/day.

analogs. Oil consumption and petroleum consumption are highly correlated both for the world and for the OECD; moreover, the world aggregates are highly correlated with the OECD aggregates as well.

Figure 1 plots the time series for price and for the various measures of quantity. Once again, my various measures of quantity are highly correlated and upward-sloping. The real oil price jumps in 1974 and again in 1979, and drops in 1986, in

Figure 1

Repsol YPF-Harvard Kennedy School Fellows Research Papers

220

Figure 2

accordance with historical developments in the world oil market (Yergin, 1992). Figure 2 presents scatter plots of price versus each of the measures of quantity. The least-squares regression line is plotted as well. Observations before the 1973 Arab oil embargo are denoted with an open circle; observations from 1973 onwards are denoted with a filled circle. Several features of these plots should be noted. First, the least-squares regression line has a positive slope. Second, this least-squares regression line is neither the supply function nor the demand function. Since the prices and quantities are observations of equilibrium

transactions, it is impossible to identify either supply or demand: this is precisely the simultaneity problem described in Section 2. The third feature to note is that observations that took place before the 1973 oil embargo appear markedly different from those that took place after it. In particular, when compared with the pre-1973 market, the post-1973 market had higher prices and quantities. Moreover, while quantities varied more than price before 1973, prices varied more than quantity after 1973. Figure 3 is directly analogous to Figure 2, except that the prices and quantities used for the scatter plots are in logs rather than in levels. The

C.-Y. Cynthia Lin

221

Figure 3

qualitative features of the scatter plots of price versus quantity appear robust to whether these variables are logged or not. Since the data used to measure oil quantity in my annual data set were consumption data, one may wonder whether using production rather than consumption as a measure of quantity might yield different results. Although I was unable to obtain data for either oil or petroleum production that spanned the years 1965–2000,8 my monthly data does include a series on world oil production that spans 1970–2003. In order to compare measures of consumption with those of production, I compare my annual consumption time series with an annual average of my monthly production time

series.9 As seen in Figure 4, oil consumption and oil production are highly correlated, with a correlation of 0.94. Thus, my results should be robust to whether I use consumption or production as my measure of quantity. Having described my annual data set, I now turn to my monthly data set. 4.2 Monthly Data (1981–2000) While my monthly data covers fewer years than my annual data set, it has several advantages.10 First, there are many more observations in my monthly data set than in my annual data set, which increases the number of degrees of freedom

Repsol YPF-Harvard Kennedy School Fellows Research Papers

222

Figure 4

in my estimations. Second, my monthly data set includes many more variables than my annual data set does. Third, all the observations in my monthly data set took place after the 1973 Arab oil embargo. As seen in the previous section, the oil market appeared to have changed dramatically after 1973. My monthly data thus enables me to focus on the post-1973 oil market. Table 3a presents the summary statistics for the monthly variables in my data set; Table 3b presents the summary statistics for the annual variables used in my monthly analyses when restricted to the years 1981–2000 and when, for each year, the same annual value is repeated for all months in that year.11

I use two measures of price: the real average OPEC crude oil price and the real average nonOPEC crude oil price. Both were collected by the U.S. Department of Energy and were deflated to 1982–1984 U.S. dollars using the CPI. I use three measures of quantity: world oil production, OPEC oil production, and non-OPEC oil production. The world and OPEC production data are from the Oil and Gas Journal. The non-OPEC production data were constructed as the difference between the two. The other monthly variable I collected is that total world rig count as reported by Baker Hughes, Inc. For my annual covariates, I use several new variables in addition to the variables used in my annual analyses. I am able to use additional annu-

C.-Y. Cynthia Lin

223

Table 3a Summary statistics for monthly data (1981–2000): monthly variables Variable

mean

s.d.

min

max

trend

Price real average spot price for crude oil: total OPEC (1982–1984$/barrel)

17.02

9.00

5.91

39.72

real average spot price for crude oil: total non-OPEC (1982–1984$/barrel)

17.28

8.82

5.75

43.94

–0.10*** (0.01) –0.10*** (0.01)

world oil production (million barrels/day)

59.40

4.48

49.39

69.32

OPEC oil production (million barrels/day)

22.50

3.93

13.90

29.59

non-OPEC oil production (million barrels/day)

36.89

1.52

33.17

40.86

25.02

11.71

11.56

62.31

Quantity 0.06*** (0.00) 0.05*** (0.00) 0.011*** (0.001)

Monthly Covariates total world rig count (100 rigs)

–0.13*** (0.01)

Notes: The trend is the coefficient on month when the variable is regressed on month and a constant (standard error in parentheses). Significance codes: * 5% level, **1% level, and ***0.1% level.

al variables in my monthly analyses both because the greater number of monthly observations increases the degrees of freedom and thus enables me to increase the number of variables used in my regressions, and because the restricted range of years enables me to use variables that were not available for all the years covered by my annual data set. One annual variable I use in my monthly analysis but not my annual analysis is world commercial energy use, obtained from the World Bank Group World Development Indicators (WDI) online database. I also break down both the electricity production from oil and that from natural gas into several regional aggregates: world; highincome OECD, high-income non-OECD; and Middle East and North Africa.12 Unlike for the annual time series, monthly oil price has a significant negative trend over 1981–2000. Production has a significant positive trend. World rig count is declining. The signs of the trends for the annual variables when converted to a monthly series and restricted to 1981–2000 are similar to the analogous variables in the annual data set. One exception is real GDP in the Middle East and North Africa, which now has a significant negative trend. As for the new annual variables, world commercial energy use has a significant positive trend, and electricity production from oil and that from gas in different parts of the

world all have trends of the same sign as the world aggregate: as in the annual data spanning 1965–2000, electricity production from oil is decreasing while that from natural gas is increasing in the monthly data spanning 1981–2000. How correlated are my different measures of price, and how correlated are my different measures of quantity, both over 1981–2000? My two measures of price, OPEC oil price and non-OPEC oil price, are highly correlated, with a correlation of 0.99. Table 4 presents the correlation between my various measures of quantity. While world oil production and OPEC oil production are highly correlated with each other, non-OPEC oil production is not highly correlated with either of the two. Figure 5 presents the time series for the various measures of price and quantity. Once again, over the 1981–2000 time period, OPEC quantity and world quantity are highly correlated and all three measures of quantity are increasing. OPEC and non-OPEC prices are also correlated over this time period, but are declining. As expected, prices collapse in the mid-1980s. Figure 6 displays a scatter plot of the following combinations of price and quantity that I will later use for my supply and demand estimations: (1) OPEC price and world quantity; (2) OPEC price and OPEC quantity; (3) non-OPEC price and world quantity; and (4) non-OPEC price and non-

Repsol YPF-Harvard Kennedy School Fellows Research Papers

224

Table 3b Summary statistics for monthly data (1981–2000): Annual covariates mean

s.d.

min

max

trend

real world GDP (trillion 1982–1984$)

15.77

2.76

11.36

19.16

real GDP in Middle East and North Africa (trillion 1982–1984$)

0.35

0.04

0.30

0.42

world population (billions)

5.28

0.48

4.50

6.05

world commercial energy use (million kt of oil equivalent)

8.53

0.86

7.10

9.94

world electricity production (trillion kwh)

11.70

2.08

8.39

15.30

electricity production from oil, world (%)

15.84

6.74

7.82

27.09

electricity production from oil, high-income OECD (%)

8.62

2.35

5.40

15.24

electricity production from oil, high-income non-OECD (%)

32.08

10.08

25.41

62.18

electricity production from oil, Middle East and North Africa (%)

51.40

5.05

42.32

60.16

electricity production from natural gas, world (%)

11.86

3.70

7.48

17.45

electricity production from natural gas, high-income OECD (%)

11.30

2.06

8.70

15.72

electricity production from natural gas, high-income non-OECD (%)

18.86

2.86

14.14

24.13

electricity production from natural gas, Middle East and North Africa (%)

37.28

6.96

27.90

49.90

world crude oil reserves (billion barrels)

881.68

152.95

648.53

1034.27

4.14

0.83

2.63

5.15

0.04*** (0.00) –2e–4*** (0.0000) 0.007*** (0.000) 0.01*** (0.00) 0.03*** (0.00) –0.09*** (0.00) –0.03*** (0.00) –0.10*** (0.01) –0.06*** (0.00) 0.05*** (0.00) 0.03*** (0.00) 0.03*** (0.00) 0.10*** (0.00) 2.00*** (0.06) 0.012*** (0.0002)

Variable Annual Covariates

world natural gas reserves (1015 cubic feet)

Notes: For the annual covariates, each annual value is repeated for all twelve months in the corresponding year. The trend is the coefficient on year when the variable is regressed on year and a constant (standard error in parentheses). Significance codes: *5% level, **1% level, and ***0.1% level.

Table 4 Correlation between various measures of monthly oil quantity

world oil production OPEC oil production non-OPEC oil production Note: Production is measured in million barrels/day.

world oil production

OPEC oil production

non-OPEC oil production

1.00 0.94 0.51

1.00 0.20

1.00

C.-Y. Cynthia Lin

225

Figure 5

OPEC quantity. As before, because these prices and quantities are equilibrium observations, one cannot identify either a supply curve or a demand curve. However, unlike before, the least-squares regression line now has a negative rather than a positive slope. Figure 7 plots the analogous scatter plots using logs rather than levels. Because the qualitative features of the plots are the same whether the variables are in logs or in levels, I will only present the complete results for the levels form for the estimations; for the logarithmic form, only the results from the efficient and identified 3SLS estimation will be presented. Having described my annual and monthly data sets, I now proceed to estimating world oil demand and supply.

5.

Results

5.1 Annual Supply and Demand (1965–2000) For my annual analyses, I make the following exclusion restrictions: (1) The following covariates are exogenous annual demand shifters x dt that affect the demand for oil but not its supply: world GDP, population, electricity production, electricity production from oil, and electricity production from gas. (2) The following covariates are exogenous annual supply shifters x st that affect the supply of oil but not its demand: world oil reserves and OPEC oil reserves.

Repsol YPF-Harvard Kennedy School Fellows Research Papers

226

Figure 6

(3) The following covariates x nt are endogenous: GDP in the Middle East and North Africa, which I assume affects the supply of oil but not its demand. (4) The following covariates are exogenous market controls x ct that affect both supply and demand: an indicator variable for year of the 1973 Arab oil embargo and an indicator variable for all years from 1973 onwards. All the exogenous covariates zt = (x dt ,x st ,x ct ) will be used as instruments in my instrumental variables estimations. Table 5 presents the estimates of the reducedform relationships (4) and (5) between price and quantity, respectively, and all the covariates. In the

price equations, GDP in the Middle East and North Africa has a significant positive coefficient.13 Moreover, population and electricity production from oil both have a significant negative coefficient in the log price regression, as does the dummy variable for the year of the Arab oil embargo. Log prices were higher after the oil embargo. For the reduced-form quantity regressions, world GDP, population, electricity production from oil and from gas, and the Arab oil embargo all had a significant positive effect on the quantity of oil consumed. None of the coefficients in the regressions of petroleum production were individually significant, but altogether they were jointly significant (p-value = 0.00 in both the level and log regressions).

C.-Y. Cynthia Lin

227

Figure 7

To test whether Assumption 2 that the instruments are correlated with price appears reasonable, I regress price on the instruments zt = (x dt ,x st ,x ct ). The results are provided in Table 6. The difference between the regressions in Table 6 and the analogous price regressions in Table 5 is that the former no longer includes the endogenous covariate GDP in the Middle East and North Africa as a regressor. In the regression of oil price level, none of the instruments has a significant coefficient, although altogether they are jointly significant (p-value = 0.00) and the supply shifters are jointly significant at a 10% level (p-value = 0.06). When log oil price is used, only the market controls are individually significant, although all

the instruments together are jointly significant (pvalue = 0.00) and the supply shifters are jointly significant at a 10% level (p-value = 0.07). Because the supply shifters are jointly significant while the demand shifters are not, one expects the demand equation to be better identified than the supply equation. However, because neither set of shifters is jointly significant at a 5% level, identification of either equation may be weak at best.14 Table 7 presents the results of the structural estimation of the demand and supply equations using oil consumption as a measure of quantity. As explained above, assuming that my instruments are valid, the OLS results are neither identified nor efficient, the SUR results are efficient but

Repsol YPF-Harvard Kennedy School Fellows Research Papers

228

Table 5 Reduced form estimates for (1) annual world oil price and (2) annual world oil or petroleum quantity consumed

world GDP (trillion 1982–1984$) GDP in Middle East and North Africa (trillion 1982–1984$) population (billions) electricity production (trillion kwh) electricity production from oil (%) electricity production from gas (%) world oil reserves (billion barrels) OPEC oil reserves (billion barrels) year = 1973 year >= 1973 constant p-value (Prob > F) adj. R squared # observations

Dependent variable is: Quantity consumed Price (1.b) (2.a) (2.b) (2.c) (2.d) (1.a) oil price ln(oil price) oil qty ln(oil qty) pet. qty ln(pet. qty) –0.31 –0.05 1.10** 0.02** 2.12 0.05 (1.02) (0.05) (0.30) (0.01) (1.41) (0.04) 183.2*** 6.76*** 10.18 0.23 53.13 1.34 (27.75) (1.22) (8.09) (0.18) (38.19) (0.95) –24.63 –2.66* 18.60* 0.61*** 53.46 1.50 (23.12) (1.02) (6.74) (0.15) (31.81) (0.79) –0.74 0.31 –1.94 –0.09** –9.04 –0.28 (4.79) (0.21) (1.40) (0.03) (6.59) (0.16) –0.75 –0.05** 0.79*** 0.02*** 0.87 0.02 (0.37) (0.02) (0.11) (0.00) (0.51) (0.01) –0.38 –0.01 0.78*** 0.01** –0.02 –0.01 (0.56) (0.02) (0.16) (0.00) (0.77) (0.02) 0.04 0.00 –0.00 0.00 –0.06 –0.00 (0.07) (0.00) (0.02) (0.00) (0.09) (0.00) 0.01 –0.00 0.00 –0.00 0.06 0.00 (0.08) (0.00) (0.02) (0.00) (0.11) (0.00) –6.11 –0.86** 4.44** 0.09* 14.96 0.35 (5.30) (0.23) (1.55) (0.03) (7.30) (0.18) –2.59 0.98** –2.84 –0.08 –15.70 –0.40 (6.13) (0.27) (1.79) (0.04) (8.43) (0.21) 71.60 9.15** –50.32* 1.14 –145.68 –1.28 (66.61) (2.94) (19.42) (0.44) (91.65) (2.29) 0.00*** 0.00*** 0.00*** 0.00*** 0.00*** 0.00*** 0.86 0.94 0.99 0.98 0.81 0.71 36 36 36 36 36 36

Notes: Standard errors are in parentheses. Significance codes:*5% level, **1% level, and ***0.1% level. Prob>F is the p-value from F-tests on all the coefficients. Oil price is in 1982-1984$/barrel; oil and petroleum consumption are in million barrels/day.

not identified, the equation-by-equation 2SLS results are identified but not efficient, and the 3SLS results are both identified and efficient. Table 8 presents analogous results using logs for oil consumption and for oil price instead of levels. There are several main results to be gleaned from the estimates in Tables 7 and 8 of annual demand and supply functions using oil consumption as quantity in both levels and logarithmic form, respectively. First, the price coefficients are only significant for the SUR estimations, in which case they are negative for both the demand and the supply equations. In contrast, economic theory suggests that the coefficient should be negative in the demand equation but positive in the supply equation. Second, while the significance and magnitudes vary, the signs on the coefficients in the

levels estimations are generally the same as those for the corresponding log estimations. Third, GDP in the Middle East and North Africa has a positive effect on supply that is also significant in most specifications. This result accords well with anecdotal evidence that, for countries in the Middle East, oil production is closely tied with national economic policy (Bob Tippee, personal communication, January 23, 2004). Fourth, the signs of the coefficients in the SUR estimation are often the opposite of the signs for the corresponding coefficients in the other three regressions. Lastly, for the efficient and consistent 3SLS estimates, the price coefficient is not significant for either supply or demand in either the levels or log regressions. Tables 9 and 10 present analogous results for the supply and demand estimation in levels and

C.-Y. Cynthia Lin

229

Table 6 Effects of instruments on annual world oil price Dependent variable is: (1) oil price Demand shifters world GDP (trillion 1982–1984$) 0.76 (1.64) population (billions) –31.59 (37.51) electricity production (trillion kwh) 7.34 (7.52) electricity production from oil (%) 0.17 (0.56) electricity production from gas (%) 1.08 (0.83) p-value from joint test of all demand shifters [0.61] Supply shifters world oil reserves (billion barrels) 0.05 (0.11) OPEC oil reserves (billion barrels) –0.12 (0.12) p-value from joint test of all supply shifters [0.06] year = 1973 year >= 1973 constant

Market controls –15.27 (8.32) 17.68 (8.62) 88.82 (109.10)

p-value from joint test of all coefficients (Prob > F) adj. R squared # observations

0.00*** 0.64 36

(2) ln(oil price) –0.01 (0.07) –2.92 (1.49) 0.61 (0.30) –0.01 (0.02) 0.04 (0.03)

that are consistent with economic theory if my econometric and theoretical assumptions hold, it does not. One possible reason is that my instruments are too weak because they are not adequately correlated with price. Another is that my maintained assumption of a static and perfectly competitive world oil market is incorrect. Because the results of the regressions of price on instruments in Table 6 show a weak correlation, the first explanation seems plausible. To determine whether better instruments can yield estimates of supply and demand curves consistent with a static and perfectly competitive world market, I now run analogous analyses with my more comprehensive monthly data. 5.2 Monthly Supply and Demand (1981–2000)

[0.41]

For my monthly analyses, I make the following exclusion restrictions:

0.01 (0.00) –0.01 (0.00)

(1) The following covariates are exogenous demand shifters x dt that affect the demand for oil but not its supply: world GDP; world population; world commercial energy use; world electricity production; electricity production from oil in the world, in high-income OECD countries, in high-income non-OECD countries, and in the Middle East and North Africa; electricity production from gas in the world, in high-income OECD countries, in highincome non-OECD countries, and in the Middle East and North Africa; and world natural gas reserves. (2) The following covariates are exogenous supply shifters x st that affect the supply of oil but not its demand: total world rig count and world oil reserves. (3) The following covariates x nt are endogenous: GDP in the Middle East and North Africa, which I assume affects the supply of oil but not its demand. (4) The following covariates are exogenous market controls x ct that affect both supply and demand: an indicator variable for the summer months (June, July and August), and an indicator variable for the winter months (December, January and February).15

[0.07] –1.20** (0.33) 1.72*** (0.24) 9.79* (4.29) 0.00*** 0.99 36

Notes: Standard errors are in parentheses. Significance codes: *5% level, **1% level, and ***0.1% level. Prob > F is the p-value from F-tests on all the coefficients. F-tests are also conducted for all the demand shifters and for all the supply shifters. Oil price is in 1982–1984$.

logarithmic form, respectively, using petroleum consumption rather than oil consumption as the measure of quantity. For this case, the coefficient on price is not significant for either supply or demand in any of the specifications. Thus, for my annual analyses, although 3SLS should yield efficiently identified price coefficients

All the exogenous covariates zt = (x dt ,x st ,x ct ) will be used as instruments in my instrumental variables estimations.

Repsol YPF-Harvard Kennedy School Fellows Research Papers

230

Tables 11a and 11b present the estimates of the reduced-form relationships (4) and (5) between price of oil and quantity of oil production, respectively, and all the covariates. For the reduced-form

price regressions, the signs of the significant coefficients appear to be robust to whether or not the price is the OPEC price or the non-OPEC price, and to whether or not the price is logged. Among the

Table 7 Annual supply and demand using oil consumption for quantity OLS (1)

SUR (2)

2SLS (3)

3SLS (4)

Demand equation: Dependent variable is quantity of oil consumption (million barrels/day) oil price (1982–1984$/barrel) 0.04 –0.35* 0.08 (0.03) (0.17) (0.07) world GDP (trillion 1982–1984$) 1.05*** –0.61* 1.11*** (0.24) (0.26) (0.27) population (billions) 18.51** –12.18 19.82** (5.65) (6.17) (6.19) electricity production (trillion kwh) –1.66 2.18 –1.93 (1.22) (1.25) (1.34) electricity production from oil (%) 0.82*** –0.48*** 0.81*** (0.06) (0.10) (0.07) electricity production from gas (%) 0.79*** –0.42** 0.78*** (0.13) (0.14) (0.14) year = 1973 4.31* –14.52 5.14* (1.56) (8.73) (2.09) year >= 1973 2.23 36.4*** –3.22 (1.62) (5.83) (2.30) constant –49.9** 96.0*** –54.0** (16.70) (19.33) (18.40) adj. R squared 0.99 0.52 0.99

0.08 (0.07) 1.01*** (0.24) 16.40** (5.74) –1.15 (1.22) 0.77*** (0.06) 0.77*** (0.13) 5.32* (2.06) –3.00 (2.24) –42.92* (16.90) 0.99

Supply equation: Dependent variable is quantity of oil consumption (million barrels/day) oil price (1982–1984$/barrel) –0.11 –0.25** 0.45 (0.08) (0.09) (0.45) 58.1*** 22.7*** 49.76 GDP in Middle East and North Africa (trillion 1982–1984$) (12.02) (5.84) (29.09) world oil reserves (billion barrels) 0.06* 0.00 0.04 (0.01) (0.04) (0.02) OPEC oil reserves (billion barrels) –0.03 0.01 0.02 (0.02) (0.01) (0.05) year = 1973 5.03 –6.16 14.26 (2.81) (4.73) (7.12) year >= 1973 –3.76 18.0*** –15.03 (2.98) (3.30) (8.67) 17.3*** 35.5*** 10.37 constant (2.67) (2.42) (6.71) 0.86 0.86 adj. R squared 0.96 # observations 36 36 36

0.44 (0.44) 51.11 (28.42) 0.03 (0.04) 0.02 (0.05) 14.84 (7.66) –14.84 (8.57) 10.77 (6.48) 0.87 36

Notes: Standard errors are in parentheses. Significance codes: *5% level, **1% level, and ***0.1% level. For 2SLS and 3SLS estimations, the instruments used are: world GDP, population, electricity production, electricity production from oil, electricity production from gas, world oil reserves, OPEC oil reserves, year = 1973, and year >= 1973. For the SUR and 3SLS estimations, the maximum number of iterations was 1e6 and the toleration level for each iteration was 1e–5.

C.-Y. Cynthia Lin

231

covariates with a positive effect on price are the total world rig count, GDP in the Middle East and North Africa, 16 world electricity production, electricity production from natural gas in high-income

OECD countries, and world oil reserves. Among the covariates with a negative effect on price are world population, world commercial energy use, electricity production from oil in the Middle East

Table 8 Annual supply and demand using ln(oil consumption) for quantity OLS (1)

SUR (2)

2SLS (3)

3SLS (4)

Demand equation: Dependent variable is ln(quantity of oil consumption) 0.03 –0.16* (0.02) (0.07) world GDP (trillion 1982–1984$) 0.02*** –0.01* (0.01) (0.00) population (billions) 0.75*** –0.37** (0.13) (0.12) electricity production (trillion kwh) –0.11*** 0.07** (0.03) (0.02) electricity production from oil (%) 0.02*** –0.01*** (0.00) (0.00) electricity production from gas (%) 0.01*** –0.01* (0.00) (0.00) year = 1973 0.11** –0.33 (0.04) (0.18) year >= 1973 –0.10 0.87*** (0.05) (0.16) constant 0.71 5.37*** (0.39) (0.41) adj. R squared 0.99 0.53

0.02 (0.04) 0.02** (0.01) 0.73*** (0.15) –0.11** (0.03) 0.02*** (0.00) 0.02*** (0.00) 0.10 (0.06) –0.08 (0.09) 0.76 (0.50) 0.99

0.06 (0.04) 0.02** (0.01) 0.73*** (0.17) –0.10* (0.04) 0.02*** (0.00) 0.01* (0.00) 0.17* (0.07) –0.14 (0.10) 0.85 (0.55) 0.98

Supply equation: Dependent variable is ln(quantity of oil consumption) –0.05 –0.10* (0.04) (0.04) GDP in Middle East and North Africa (trillion 1982–1984$) 0.96*** 0.26* (0.10) (0.24) world oil reserves (billion barrels) 0.002*** 0.00 (0.000) (0.00) OPEC oil reserves (billion barrels) –0.002** –0.00 (0.00) (0.00) year = 1973 0.07 –0.16 (0.10) (0.07) 0.47*** –0.04 year >= 1973 (0.08) (0.09) 3.72*** 3.24*** constant (0.08) (0.08) adj. R squared 0.95 0.84 # observations 36 36

–0.06 (0.09) 1.21*** (0.27) 0.002*** (0.001) –0.002* (0.001) 0.09 (0.12) –0.07 (0.16) 3.24*** (0.16) 0.95 36

–0.11 (0.09) 1.48*** (0.26) 0.002** (0.001) –0.00 (0.00) 0.06 (0.13) –0.00 (0.17) 3.36*** (0.17) 0.94 36

ln(oil price)

ln(price)

Notes: Standard errors are in parentheses. Significance codes: *5% level, **1% level, and ***0.1% level. For 2SLS and 3SLS estimations, the instruments used are: world GDP, population, electricity production, electricity production from oil, electricity production from gas, world oil reserves, OPEC oil reserves, year = 1973, and year >= 1973. For the SUR and 3SLS estimations, the maximum number of iterations was 1e6 and the toleration level for each iteration was 1e-5. Oil price is in 1982-1984$/barrel; oil consumption is in million barrels/day.

Repsol YPF-Harvard Kennedy School Fellows Research Papers

232

and North Africa, and electricity production from natural gas in high-income non-OECD countries. For the reduced-form quantity regressions, more coefficients are significant in the regressions of non-

OPEC oil production than in those of world or OPEC oil production. For world production, world commercial energy use, electricity production from oil in high-income OECD countries, and electricity from

Table 9 Annual supply and demand using petroleum consumption for quantity OLS (1)

SUR (2)

2SLS (3)

3SLS (4)

Demand equation: Dependent variable is quantity of petroleum consumption oil price (1982–1984$/barrel) 0.28 –0.32 0.62 (0.15) (0.19) (0.38) world GDP (trillion 1982–1984$) 1.73 –0.19 2.24 (1.17) (0.12) (1.37) population (billions) 41.85 –5.70 52.65 (27.47) (2.83) (31.79) electricity production (trillion kwh) –6.06 1.11 –8.36 (5.94) (0.57) (6.86) electricity production from oil (%) 0.71* –0.08 0.64 (0.31) (0.05) (0.34) electricity production from gas (%) –0.05 0.03 –0.16 (0.65) (0.07) (0.72) year = 1973 14.33 –11.84 21.22 (7.59) (10.47) (10.70) year >= 1973 –14.02 29.6*** –22.23 (7.86) (5.47) (11.78) constant –122.3 61.6*** –145.9 (81.14) (9.54) (94.45) adj. R squared 0.80 0.42 0.76

0.40 (0.29) 0.42 (0.70) 20.19 (22.46) –0.29 (4.38) 0.51* (0.22) 0.21 (0.53) 18.34 (9.63) –15.46 (10.28) –45.00 (63.36) 0.78

Supply equation: Dependent variable is quantity of petroleum consumption –0.20 –0.29 –0.43 (0.21) (0.17) (0.64) 114.3*** 8.95 138.5** GDP in Middle East and North Africa (trillion 1982–1984$) (30.68) (4.85) (41.16) world oil reserves (billion barrels) 0.00 –0.01 0.00 (0.01) (0.06) (0.05) OPEC oil reserves (billion barrels) 0.04 0.01 0.03 (0.06) (0.01) (0.07) year = 1973 13.60 –9.18 13.10 (7.17) (9.41) (10.92) year >= 1973 –14.30 24.9*** –14.67 (7.61) (4.98) (12.26) 16.11* 41.6*** 17.19 constant (6.85) (3.57) (9.49) 0.79 0.53 adj. R squared 0.80 # observations 36 36 36

–0.36 (0.62) 129.7** (37.92) 0.02 (0.04) 0.01 (0.06) 13.31 (10.82) –15.92 (11.97) 14.62 (8.31) 0.79 36

oil price (1982–1984$/barrel)

Notes: Standard errors are in parentheses. Significance codes: *5% level, **1% level, and ***0.1% level. For 2SLS and 3SLS estimations, the instruments used are: world GDP, population, electricity production, electricity production from oil, electricity production from gas, world oil reserves, OPEC oil reserves, year = 1973, and year >= 1973. For the SUR and 3SLS estimations, the maximum number of iterations was 1e6 and the toleration level for each iteration was 1e-5. Oil price is in 1982-1984$/barrel; petroleum consumption is in million barrels/day.

C.-Y. Cynthia Lin

233

gas in high-income non-OECD countries all have a significant positive effect on world oil production, while electricity production from gas in high-income OECD countries has a significant negative effect.

To test whether Assumption 2 that the instruments are correlated with price appears reasonable, I regress price on the instruments zt = (x dt ,x st ,x ct ). The results are provided in Table 12.

Table 10 Annual Supply and Demand using ln(Petroleum Consumption) for quantity OLS (1)

SUR (2)

2SLS (3)

3SLS (4)

Demand equation: Dependent variable is ln(quantity of petroleum consumption) ln(price) 0.18 –0.15 0.27 (0.09) (0.09) (0.22) world GDP (trillion 1982–1984$) 0.05 –0.00 0.01 (0.03) (0.00) (0.04) population (billions) 1.51* –0.15* 1.70 (0.69) (0.05) (0.83) electricity production (trillion kwh) –0.26 0.03* –0.30 (0.15) (0.01) (0.18) electricity production from oil (%) 0.02* –0.00* 0.02* (0.01) (0.00) (0.01) electricity production from gas (%) –0.01 0.00 –0.01 (0.02) (0.00) (0.02) year = 1973 0.46* –0.28 0.57 (0.21) (0.24) (0.35) year >= 1973 –0.57* 0.72*** –0.73 (0.25) (0.17) (0.48) constant –1.60 4.40*** –2.29 (2.08) (0.23) (2.69) adj. R squared 0.72 0.34 0.71

0.20 (0.17) 0.03 (0.02) 1.49 (0.73) –0.24 (0.15) 0.02** (0.1) –0.01 (0.01) 0.48 (0.30) –0.57 (0.41) –1.51 (2.30) 0.71

Supply equation: Dependent variable is ln(quantity of petroleum consumption) –0.08 –0.14 –0.46 (0.11) (0.08) (0.30) 2.23** 0.09 2.84** GDP in Middle East and North Africa (trillion 1982–1984$) (0.69) (0.08) (0.93) world oil reserves (billion barrels) 0.00 –0.00 0.00 (0.00) (0.00) (0.00) OPEC oil reserves (billion barrels) –0.00 0.00 –0.00 (0.00) (0.00) (0.00) year = 1973 0.27 –0.22 –0.13 (0.20) (0.22) (0.41) year >= 1973 –0.26 0.63*** 0.30 (0.23) (0.15) (0.55) 3.25*** 3.89*** 3.88*** constant (0.24) (0.14) (0.56) 0.57 0.44 adj. R squared 0.69 # observations 36 36 36

–0.35 (0.27) 2.29** (0.76) 0.00 (0.00) –0.00 (0.00) –0.06 (0.39) 0.16 (0.51) 3.63*** (0.48) 0.61 36

ln(price)

Notes: Standard errors are in parentheses. Significance codes: *5% level, **1% level, and ***0.1% level. For 2SLS and 3SLS estimations, the instruments used are: world GDP, population, electricity production, electricity production from oil, electricity production from gas, world oil reserves, OPEC oil reserves, year = 1973, and year >= 1973. For the SUR and 3SLS estimations, the maximum number of iterations was 1e6 and the toleration level for each iteration was 1e-5. Oil price is in 1982-1984$/barrel; petroleum consumption is in million barrels/day.

234

Repsol YPF-Harvard Kennedy School Fellows Research Papers

Table 11a Reduced form estimates for monthly oil price (1981–2000) Dependent variable is: Price (1) OPEC total world rig count (100 rigs) summer dummy (Jun, Jul, Aug) winter dummy (Dec, Jan, Feb)

Monthly Covariates 0.23*** (0.05) –0.40 (0.35) –0.10 (0.37)

Annual Covariates 57.22 (25.21) GDP, Mid East & N.Afr. (trillion 1982–1984$) 57.22* (25.21) population, world (billions) –72.01*** (13.55) commercial energy use, world (mill. kt of oil equivalent) –40.14*** (7.04) electricity production, world (trillion kwh) 23.86*** (5.00) electricity prodn from oil, world (%) –2.48 (1.37) electricity prodn from oil, high-inc OECD (%) –2.02 (1.16) electricity prodn from oil, high-inc non-OECD (%) –0.16 (0.16) electricity prodn from oil, Mid East & N.Afr. (%) –2.42*** (0.69) electricity prodn from natural gas, world (%) –4.93** (1.83) electricity prodn from natural gas, high-inc OECD (%) 6.68*** (1.08) electricity prodn from natural gas, high-inc non-OECD (%) –1.81*** (0.39) electricity prodn from natural gas, Mid East & N.Afr. (%) –1.78* (0.73) 0.05*** crude oil reserves, world (billion barrels) (0.01) –6.41 natural gas reserves, world (1015 cubic ft) (3.75) constant 673.5*** (106.4) p-value (Prob > F) 0.00 *** adj. R squared 0.94 # observations 240 GDP, world (trillion 1982–1984$)

ln(Price)

(2) non-OPEC

(3) OPEC

(4) non-OPEC

0.19*** (0.05) –0.48 (0.35) 0.09 (0.37)

0.015*** (0.004) –0.02 (0.03) –0.02 (0.03)

0.015*** (0.004) –0.02 (0.02) –0.03 (0.03)

–0.02 (0.77) 36.99 (25.51) –87.3*** (13.71) –34.7*** (7.12) 26.62*** (5.06) –3.68** (1.39) –0.44 (1.18) –0.20 (0.17) –2.63*** (0.70) –6.00** (1.86) 7.13*** (1.09) –1.48*** (0.40) –2.12** (0.74) 0.04** (0.01) –8.32* (3.79) 107.6*** (107.6) 0.00 *** 0.94 240

0.15** (0.06) 5.76** (1.84) –5.92*** (0.99) –2.84*** (0.51) 1.72*** (0.36) –0.04 (0.10) –0.22** (0.08) –0.02 (0.01) –0.12* (0.05) –0.16 (0.13) 0.37*** (0.08) –0.15*** (0.03) –0.06 (0.05) 0.004*** (0.001) –0.18 (0.03) 42.64*** (7.76) 0.00 *** 0.89 240

0.11* (0.05) 4.57* (1.81) –6.25*** (0.97) –2.78*** (0.51) 1.89*** (0.36) –0.10 (0.10) –0.16 (0.08) –0.02* (0.01) –0.14** (0.05) –0.24 (0.13) 0.39*** (0.08) –0.14*** (0.03) –0.09 (0.05) 0.004*** (0.001) –0.30 (0.27) 46.82*** (7.63) 0.00*** 0.88 240

Notes: Standard errors are in parentheses. Significance codes: *5% level, **1% level, and ***0.1% level. Prob > F is the p-value from F-tests on all the coefficients. Oil price is in 1982-1984$/barrel.

C.-Y. Cynthia Lin

235

Table 11b Reduced form estimates for monthly oil quantity produced (1981–2000) Dependent variable is: Quantity (1) world total world rig count (100 rigs) summer dummy (Jun, Jul, Aug) winter dummy (Dec, Jan, Feb)

GDP, world (trillion 1982–1984$) GDP, Mid East & N.Afr. (tr. 1982–1984$) population, world (billions) commercial energy use, world (m. kt of oil equiv) electricity production, world (trillion kwh) elec. prodn from oil, world (%) elec. prodn from oil, high-inc OECD (%) elec. prodn from oil, high-inc non-OECD (%) elec. prodn from oil, Mid East & N. Afr. (%) elec. prodn from gas, world (%) elec. prodn from gas, high-inc OECD (%) elec. prodn from gas, high-inc non-OECD (%) elec. prodn from gas, Mid East & N.Afr. (%) crude oil reserves, world (billion barrels) natural gas reserves, world (1015 cubic ft) constant p-value (Prob > F) adj. R squared # observations

(2) OPEC

ln(Quantity) (3) non-OPEC (4) world (5) OPEC (6) non-OPEC

Monthly Covariates 0.04 0.01 (0.03) (0.03) –0.33 –0.03 (0.21) (0.19) 0.08 –0.04 (0.22) (0.20)

0.03* (0.01) –0.30*** (0.08) 0.12 (0.08)

0.00 (0.00) –0.01 (0.00) 0.00 (0.00)

0.00 (0.00) 0.00 (0.01) –0.00 (0.01)

6e–4 (3e-4) –8e–3*** (2e-3) 0.00 (0.00)

Annual Covariates –0.74 –0.24 (0.46) (0.41) –12.28 13.99 (15.19) (13.53) 0.73 –17.79* (8.16) (7.27) 8.47* 1.58 (4.24) (3.78) 2.56 5.19 (3.01) (2.68) –0.31 5.19 (0.83) (2.68) 1.98** 1.51* (0.70) (0.62) 0.02 –0.15 (0.10) (0.09) 0.30 –0.50 (0.42) (0.37) 0.20 1.18 (0.99) (1.11) –1.82** –1.43* (0.65) (0.58) 0.76** 0.42* (0.24) (0.21) 0.22 –0.46 (0.39) (0.44) –0.01 0.00 (0.01) (0.01) –2.42 3.59 (2.26) (2.01) –41.94 45.47 (64.08) (57.07) 0.00*** 0.00*** 0.91 0.92 240 240

–0.50** (0.17) –26.3*** (4.65) 18.5*** (3.04) 6.88*** (1.58) –2.64* (1.12) –0.66* (0.31) 0.48 (0.26) 0.17*** (0.04) 0.79*** (0.16) –0.98* (0.41) –0.40 (0.24) 0.34*** (0.09) 0.68*** (0.16) –0.02*** (0.00) –6.02*** (0.84) –87.4*** (23.8) 0.00*** 0.90 240

–0.01 (0.01) –0.15 (0.26) –0.01 (0.14) 0.17* (0.07) 0.03 (0.05) –0.00 (0.01) 0.03** (0.01) –0.00 (0.00) 0.01 (0.01) 0.01 (0.02) –0.03** (0.01) 0.01** (0.00) 0.00 (0.01) –0.00 (0.00) –0.03 (0.04) 2.17 (1.12) 0.00*** 0.91 240

0.01 (0.02) 1.08 (0.68) –0.86* (0.37) 0.13 (0.19) 0.17 (0.14) 0.05 (0.04) 0.06 (0.03) –0.010* (0.004) –0.02 (0.02) 0.10* (05) –0.08** (0.03) 0.02 (0.01) –0.02 (0.02) 0.00 (0.00) 0.25* (0.10) 2.67 (2.88) 0.00*** 0.90 240

–0.013** (0.005) –0.70*** (0.15) 0.51*** (0.08) 0.18*** (0.04) –0.07* (0.03) –0.02* (0.01) 0.02 (0.01) 0.005*** (0.001) 0.02*** (0.00) –0.03* (0.01) 0.01 (0.01) 0.009*** (0.002) 0.02*** (0.00) –5e-4*** (0.1e-4) –0.17*** (0.00) 0.19 (0.64) 0.00*** 0.90 240

Notes: Standard errors are in parentheses. Significance codes: *5% level, **1% level, and ***0.1% level. Prob>F is the p-value from F-tests on all the coefficients. Oil production is in million barrels/day.

Repsol YPF-Harvard Kennedy School Fellows Research Papers

236

Table 12 Effects of instruments on monthly oil price Dependent variable is: Price (1) OPEC

ln(Price)

(2) non-OPEC

(3) OPEC

(4) non-OPEC

Demand shifters –0.50 (0.58) population, world (billions) –69.63*** (13.63) commercial energy use, world (mill. kt of oil equivalent) –39.86*** (7.11) electricity production, world (trillion kwh) 28.25*** (4.67) electricity prodn from oil, world (%) –5.02*** (0.80) electricity prodn from oil, high-inc OECD (%) –0.17 (0.83) electricity prodn from oil, high-inc non-OECD (%) 0.13 (0.10) electricity prodn from oil, Mid East & N. Afr. (%) –1.99** (0.67) electricity prodn from natural gas, world (%) –7.96*** (1.28) electricity prodn from natural gas, high-inc OECD (%) 6.85*** (1.08) electricity prodn from natural gas, high-inc non-OECD (%) –1.24*** (0.31) electricity prodn from natural gas, Mid East & N.Afr. (%) –1.56* (0.74) natural gas reserves, world (1015 cubic ft) –13.54*** (2.07) p-value from joint test of all demand shifters [0.00] ***

–0.75 (0.58) –85.76*** (13.70) –34.50*** (7.14) 29.46*** (4.68) –5.33*** (0.80) 0.76 (0.84) –0.01 (0.10) –2.35*** (0.68) –7.96*** (1.28) 7.24*** (1.09) –1.11*** (0.31) –2.01** (0.74) –12.93*** (2.08) [0.00] ***

0.03 (0.04) –5.67*** (1.00) –2.82*** (0.52) 2.17*** (0.34) –0.29*** (0.06) –0.04 (0.06) 0.01 (0.01) –0.07 (0.05) –0.47*** (0.09) 0.39*** (0.08) –0.09*** (0.02) –0.04 (0.05) –0.89*** (0.15) [0.00] ***

0.02 (0.04) –6.06*** (0.98) –2.76*** (0.51) 2.24*** (0.34) –0.30*** (0.06) –0.01 (0.06) –0.00 (0.01) –0.10* (0.05) –0.48*** (0.09) 0.40*** (0.08) –0.09*** (0.02) –0.07 (0.05) –0.87*** (0.15) [0.00]***

Supply shifters 0.22*** (0.05) 0.03** (0.01) [0.00] ***

0.19*** (0.05) 0.03** (0.01) [0.00] ***

0.014*** (0.004) 0.0012* (0.0006) [0.00] ***

0.014*** (0.004) 0.0015* (0.0006) [0.00]***

Market controls –0.40 (0.35) –0.08 (0.37) 708.1*** (106.2) 0.00 *** 0.94 240

–0.48 (0.35) –0.07 (0.37) 764.1*** (106.8) 0.00 *** 0.94 240

–0.03 (0.03) –0.02 (0.03) 46.13*** (7.83) 0.00 *** 0.88 240

–0.03 (0.03) –0.03 (0.03) 49.59*** (7.65) 0.00*** 0.88 240

GDP, world (trillion 1982–1984$)

total world rig count (100 rigs) crude oil reserves, world (billion barrels) p-value from joint test of all supply shifters summer dummy (Jun, Jul, Aug) winter dummy (Dec, Jan, Feb) constant p-value from joint test of all coefficients (Prob > F) adj. R squared # observations

Notes: Standard errors are in parentheses. Significance codes: *5% level, **1% level, and ***0.1% level. Prob>F is the p-value from F-tests on all the coefficients. F-tests are also conducted for all the demand shifters and for all the supply shifters. Oil price is in 1982-1984$/barrel. All covariates are annual values except the total world rig count, the summer dummy and the winter dummy.

C.-Y. Cynthia Lin

The difference between the regressions in Table 12 and the analogous price regressions in Table 11a is that the former no longer includes the endogenous covariate GDP in the Middle East and North Africa as a regressor. Unlike in the annual analyses, the instruments used in my monthly analyses appear to be highly correlated with price. Not only are all the instruments together jointly significant (p-value = 0.00 in all regressions), but the demand shifters and supply shifters are significant as well. Thus, my instruments appear not only credible, but also strong as well. The demand shifters that have a significant positive effect on price are world electricity production and electricity production from natural gas in high-income OECD countries. The demand shifters that have a significant negative effect on price are population, commercial energy use, electricity production from oil in the world and in the Middle East and North Africa, and electricity production from natural gas in the world, in high-income nonOECD countries, and in the Middle East and North Africa. The demand shifters are jointly significant (p-value = 0.00 in all regressions). Because many demand shifters are individually significant, and because they are together jointly significant, the supply equation should be identified when these shifters are used as instruments. For the supply shifters, the rig count and world oil reserves both have significant positive effects on price. These signs seem reasonable, as rig counts and world oil reserves should both shift the supply curve upward. The supply shifters are jointly significant (p-value = 0.00 in all regressions). Because the supply shifters are individually and jointly significant, the demand equation should be identified when these shifters are used as instruments. Thus, Assumption 2 that the instruments are correlated with price appears to hold, and the use of these instruments should yield identification.17 As a consequence, if the estimates of supply and demand arising from instrumental variables techniques are not consistent with a simple theoretical model of a static and perfectly competitive world oil market, the fault is likely to lie in the theoretical assumptions themselves rather than in its econometric estimation. Tables 13a and 13b present the estimates for demand and supply, respectively, when the price variable is the OPEC oil price and the quantity

237

variable is world oil production. Analogous results are presented for OPEC oil price and OPEC oil production in Tables 14a-b; non-OPEC oil price and world oil production in Tables 15a-b; and nonOPEC oil price and non-OPEC oil production in Tables 16a-b. For the estimates of demand, economic theory predicts that price should have a negative effect on demand, and econometric theory predicts that, if the theoretical model is correct, properly instrumenting for price will yield identified price coefficients of the appropriate sign. However, as seen in Tables 13a-16a, while the price coefficient is significantly negative in all of the (non-instrumented) OLS and SUR specifications for all the price-quantity combinations used, once instruments are added in 2SLS, the coefficients, while still negative, are no longer significant. Moreover, for the (instrumented) 3SLS estimations, which should yield coefficients that are both identified and efficient, the price coefficient is significantly positive in the regression of OPEC oil demand on OPEC price, although it is significantly negative in the regression of non-OPEC oil demand on non-OPEC price and not significant at a 5% level in the regressions of world oil demand on either OPEC or non-OPEC price. World oil demand therefore appears inelastic to oil price, OPEC or otherwise. Moreover, while world oil demand and non-OPEC oil demand are consistent with a static and perfectly competitive world oil market, OPEC oil demand is not. For the estimates of supply, on the other hand, economic theory predicts that price should have a positive effect on demand, and econometric theory predicts that, if the theoretical model is correct, properly instrumenting for price will yield identified price coefficients of the appropriate sign. As expected, the price coefficient has the wrong sign in the (non-instrumented) OLS and SUR specifications for all the price-quantity combinations in Tables 13b-16b. Using instruments for price and for GDP in the Middle East and North Africa does not yield a significantly positive price coefficient in any of the 2SLS or 3SLS specifications, although in some cases it yields coefficients that are no longer significantly negative. In particular, the use of instruments yields an OPEC supply curve that is inelastic to OPEC price. Thus, while OPEC supply is consistent with a static and perfectly competitive oil market, both world supply and nonOPEC supply are not.

Repsol YPF-Harvard Kennedy School Fellows Research Papers

238

Table 13a Monthly demand using OPEC oil price and quantity of world oil production Dependent variable is quantity of world oil production (million barrels/day)

OPEC oil price (1982–1984$/barrel)

summer dummy (Jun, Jul, Aug) winter dummy (Dec, Jan, Feb)

OLS (1)

SUR (2)

2SLS (3)

3SLS (4)

–0.12*** (0.03)

–0.32*** (0.03)

–0.00 (0.11)

0.03 (0.08)

–0.49 (0.50) 0.30 (0.50)

–0.39 (0.22) 0.16 (0.22)

–0.32 (0.28) 0.15 (0.28)

–0.12* (0.05) 2.49* (1.03) 0.95 (0.54) –0.55 (0.38) –0.29** (0.09) 0.21** (0.08) 0.06*** (0.01) 0.21*** (0.05) –0.31* (0.13) –0.08 (0.08) 0.09*** (0.03) 0.16** (0.05) –0.58* (0.26) 41.21*** (8.61) 0.49 240

–0.67 (0.37) –0.80 (12.2) 6.15 (4.16) 2.33 (3.80) –0.23 (0.83) 1.37** (0.47) –0.07 (0.07) –0.30 (0.26) 0.06 (1.22) –0.91 (0.84) 0.47** (0.15) –0.34 (0.28) –1.86 (1.96) 24.90 (79.56) 0.91 240

–0.90*** (0.17) 16.14* (6.94) 8.06* (3.33) –1.81 (2.73) –1.26* (0.58) 1.64*** (0.21) 0.23*** (0.04) 0.75*** (0.15) –1.13 (0.85) –1.15* (0.58) 0.74*** (0.09) 0.45** (0.15) –3.61* (1.52) –88.95 (59.98) 0.85 240

Monthly Covariates –0.42* (0.20) 0.24 (0.20)

Annual Covariates GDP, world (trillion 1982–1984$) –0.98** (0.30) population, world (billions) –14.77 (8.21) commercial energy use, world (million kt of oil equivalent) 2.62 (3.37) electricity production, world (trillion kwh) 6.39* (2.72) electricity production from oil, world (%) –1.35** (0.43) electricity production from oil, high-inc OECD (%) 1.82*** (0.36) electricity production from oil, high-inc non-OECD (%) –0.00 (0.05) electricity production from oil, Mid East & N. Afr. (%) –0.38 (0.24) electricity production from natural gas, world (%) –1.53* (0.69) electricity production from natural gas, high-inc OECD (%) 0.21 (0.47) 0.38** electricity production from natural gas, high-inc non-OECD (%) (0.14) electricity production from natural gas, Mid East & N.Afr. (%) –0.31 (0.27) natural gas reserves, world (1015 cubic ft) –4.30*** (1.19) 121.7* constant (49.95) adj. R squared 0.92 240 # observations

Notes: Standard errors are in parentheses. Significance codes: *5% level, **1% level, and ***0.1% level. For 2SLS and 3SLS estimations, the instruments used are: total world rig count; summer dummy; winter dummy; world GDP; world population; world commercial energy use; world electricity production; electricity production from oil for world, high-income OECD, high-income non-OECD, and Middle East and North Africa; electricity production from natural gas for world, high-income OECD, high-income non-OECD, and Middle East and North Africa; world crude oil reserves; and world natural gas reserves. For the SUR and 3SLS estimations, demand and supply equations corresponding to the same choice of price and quantity variables were estimated jointly, the maximum number of iterations was 1e6, and the toleration level for each iteration was 1e-5.

C.-Y. Cynthia Lin

239

Table 13b Monthly supply using OPEC oil price and quantity of world oil production Dependent variable is quantity of world oil production (million barrels/day) OLS (1)

SUR (2)

2SLS (3)

3SLS (4)

–0.21*** (0.04)

–0.31*** (0.02)

–0.21*** (0.05)

–0.17*** (0.03)

Monthly Covariates 0.09** (0.03) –0.41 (0.26) 0.06 (0.27)

0.012*** (0.004) –0.48 (0.45) 0.27 (0.45)

0.08* (0.04) –0.41 (0.26) 0.07 (0.27)

0.06** (0.02) –0.40 (0.26) 0.11 (0.27)

Annual Covariates 48.85*** (3.49) 0.028*** (0.001) 18.68*** (1.84) 0.86 240

8.77*** (1.49) 0.005*** (0.001) 56.56*** (1.41) 0.60 240

50.52*** (3.55) 0.028*** (0.001) 18.05*** (1.88) 0.86 240

49.96*** (3.49) 0.029*** (0.001) 18.02*** (1.88) 0.86 240

OPEC oil price (1982–1984$/barrel)

total world rig count (100 rigs) summer dummy (Jun, Jul, Aug) winter dummy (Dec, Jan, Feb)

GDP, Mid East & N.Afr. (trillion 1982–1984$) crude oil reserves, world (billion barrels) constant adj. R squared # observations

Notes: Standard errors are in parentheses. Significance codes: *5% level, **1% level, and ***0.1% level. For 2SLS and 3SLS estimations, the instruments used are: total world rig count; summer dummy; winter dummy; world GDP; world population; world commercial energy use; world electricity production; electricity production from oil for world, high-income OECD, high-income non-OECD, and Middle East and North Africa; electricity production from natural gas for world, high-income OECD, high-income non-OECD, and Middle East and North Africa; world crude oil reserves; and world natural gas reserves. For the SUR and 3SLS estimations, demand and supply equations corresponding to the same choice of price and quantity variables were estimated jointly, the maximum number of iterations was 1e6, and the toleration level for each iteration was 1e-5.

For any given combination of price and quantity, the signs of the significant coefficients on the covariates tend to be robust across the different estimation methods used (OLS, SUR, 2SLS, and 3SLS). However, for the 3SLS results, the signs are not robust across the different price-quantity combinations: while the signs are similar in the 3SLS estimations using OPEC price and world quantity; OPEC price and OPEC quantity; and non-OPEC price and world quantity, the signs are often flipped in the 3SLS estimations using non-OPEC price and non-OPEC quantity. For example, world population, world commercial energy use, electricity production from oil in the Middle East and North Africa, and electricity production from natural gas in the Middle East and North Africa all have significant positive effects on demand in all price-quantity combinations except that of nonOPEC price and non-OPEC quantity, in which case the effects are significantly negative. Similarly,

electricity production from natural gas in highincome OECD countries has a negative effect in all price-quantity combinations except that of nonOPEC price and non-OPEC quantity, in which case the effect is significantly positive. For the 3SLS estimates of supply, world crude oil reserves has a significant positive effect on supply in all pricequantity combinations except that of non-OPEC price and non-OPEC quantity, in which case the effects are significantly negative. Many of the signs of the 3SLS coefficients on the covariates in the demand equations that are robust with respect to the price-quantity combination used appear realistic. For example, electricity production from oil both in high-income OECD countries and in high-income non-OECD countries has a positive effect on oil demand. This is reasonable, as the more oil is needed for electricity, the higher should be oil demand. The stock of natural gas reserves has a negative effect on demand,

Repsol YPF-Harvard Kennedy School Fellows Research Papers

240

Table 14a Monthly demand using OPEC oil price and quantity of OPEC oil production Dependent variable is quantity of OPEC oil production (million barrels/day)

OPEC oil price (1982–1984$/barrel)

summer dummy (Jun, Jul, Aug) winter dummy (Dec, Jan, Feb)

OLS (1)

SUR (2)

2SLS (3)

3SLS (4)

–0.13*** (0.03)

–0.19*** (0.03)

–0.01 (0.10)

0.66*** (0.16)

–0.12 (0.48) 0.06 (0.48)

–0.04 (0.19) –0.01 (0.19)

0.28 (0.42) –0.29 (0.42)

–0.20*** (0.06) 5.07*** (1.10) 1.96*** (0.57) –0.96* (0.40) –0.40*** (0.10) 0.29*** (0.08) 0.09*** (0.01) 0.35*** (0.05) –0.48*** (0.14) –0.12 (0.09) 0.14*** (0.03) 0.29*** (0.06) –1.47*** (0.27) –20.68* (9.11) 0.40 240

–0.57 (0.32) –17.96 (10.77) 0.76 (3.67) 6.66* (3.35) –0.44 (0.73) 1.90*** (0.42) –0.09 (0.06) –0.55* (0.23) 0.16 (1.08) –1.07 (0.75) 0.51*** (0.13) –0.58* (0.25) 1.49 (1.73) 79.36 (70.21) 0.91 240

–0.20 (0.45) 58.83*** (16.45) 30.16*** (6.42) –20.14*** (5.59) 2.24 (1.16) 0.88 (0.58) 0.13 (0.09) 2.01*** (0.35) 4.00* (1.71) –4.87*** (1.18) 1.20*** (0.20) 1.56*** (0.37) 5.08 (2.90) –562.8*** (116.6) 0.56 240

Monthly Covariates –0.09 (0.18) 0.04 (0.18)

Annual Covariates GDP, world (trillion 1982–1984$) –0.79** (0.27) population, world (billions) –27.73*** (7.38) commercial energy use, world (million kt of oil equivalent) –1.71 (3.03) electricity production, world (trillion kwh) 9.50*** (2.44) electricity production from oil, world (%) –1.22** (0.39) electricity production from oil, high-inc OECD (%) 2.22*** (0.33) electricity production from oil, high-inc non-OECD (%) –0.04 (0.05) electricity production from oil, Mid East & N. Afr. (%) –0.60** (0.22) electricity production from natural gas, world (%) –0.95 (0.62) electricity production from natural gas, high-inc OECD (%) –0.29 (0.42) 0.44*** electricity production from natural gas, high-inc non-OECD (%) (0.12) electricity production from natural gas, Mid East & N.Afr. (%) –0.55* (0.24) natural gas reserves, world (1015 cubic ft) –0.21 (1.07) 147.1** constant (44.93) adj. R squared 0.92 240 # observations

Notes: Standard errors are in parentheses. Significance codes: *5% level, **1% level, and ***0.1% level. For 2SLS and 3SLS estimations, the instruments used are: total world rig count; summer dummy; winter dummy; world GDP; world population; world commercial energy use; world electricity production; electricity production from oil for world, high-income OECD, high-income non-OECD, and Middle East and North Africa; electricity production from natural gas for world, high-income OECD, high-income non-OECD, and Middle East and North Africa; world crude oil reserves; and world natural gas reserves. For the SUR and 3SLS estimations, demand and supply equations corresponding to the same choice of price and quantity variables were estimated jointly, the maximum number of iterations was 1e6, and the toleration level for each iteration was 1e-5.

C.-Y. Cynthia Lin

241

Table 14b Monthly supply using OPEC oil price and quantity of OPEC oil production Dependent variable is quantity of OPEC oil production

OPEC oil price (1982–1984$/barrel)

total world rig count (100 rigs) summer dummy (Jun, Jul, Aug) winter dummy (Dec, Jan, Feb)

GDP, Mid East & N.Afr. (trillion 1982–1984$) crude oil reserves, world (billion barrels) constant adj. R squared # observations

OLS (1)

SUR (2)

2SLS (3)

3SLS (4)

–0.11** (0.04)

–0.19*** (0.02)

–0.06 (0.05)

–0.07 (0.04)

0.010* (0.004) –0.12 (0.42) 0.04 (0.42)

0.12*** (0.04) –0.02 (0.26) –0.26 (0.27)

0.13*** (0.03) –0.03 (0.26) –0.28 (0.26)

12.19*** (1.53) 0.008*** (0.001) 14.20*** (1.43) 0.54 240

29.87*** (3.50) 0.031*** (0.001) –17.48*** (1.85) 0.83 240

30.11*** (3.46) 0.031*** (0.001) –17.47*** (1.85) 0.83 240

Monthly Covariates 0.15*** (0.03) –0.04 (0.26) –0.31 (0.26) Annual Covariates 29.82*** (3.42) 0.031*** (0.001) –17.00*** (1.81) 0.83 240

Notes: Standard errors are in parentheses. Significance codes: *5% level, **1% level, and ***0.1% level. For 2SLS and 3SLS estimations, the instruments used are: total world rig count; summer dummy; winter dummy; world GDP; world population; world commercial energy use; world electricity production; electricity production from oil for world, high-income OECD, high-income non-OECD, and Middle East and North Africa; electricity production from natural gas for world, high-income OECD, high-income non-OECD, and Middle East and North Africa; world crude oil reserves; and world natural gas reserves. For the SUR and 3SLS estimations, demand and supply equations corresponding to the same choice of price and quantity variables were estimated jointly, the maximum number of iterations was 1e6, and the toleration level for each iteration was 1e-5.

which again is reasonable because natural gas is a substitute for oil.18 One potentially surprising result is that world GDP has a negative effect on demand, which suggests that, controlling for such covariates as energy use and electricity production, oil is an inferior good, perhaps because a richer world economy would use oil more efficiently. The signs on the 3SLS coefficients in the supply equation appear realistic as well. For example, total rig count has a positive effect on supply, since, all else equal, the more exploration and production there is that takes place, the more oil there is to supply. GDP in the Middle East and North Africa has a positive effect on supply. Except in the regression of non-OPEC supply on non-OPEC price, the stock of crude oil reserves has a positive effect on supply, as expected. Do the results change when oil prices and quantities are logged? The 3SLS results for estimates of demand and supply for the various price-

quantity combinations when prices and quantities are in logs rather than in levels are presented in Tables 17a and 17b, respectively. For the most part, the qualitative results from 3SLS appear robust to whether the equations are in levels or logarithmic form. The main exception is that for the estimates of supply, the only supply curve that has a nonnegative slope consistent with economic theory when the prices and quantities are logged is nonOPEC supply, not OPEC supply, as was the case when prices and quantities were in levels. Thus, the previous result that OPEC supply was consistent with theory is not robust to the functional form of the supply curve. The robust result, therefore, is that for most specifications of supply, the price coefficients are not consistent with the assumptions of a static and perfectly competitive market. Because the instruments used in my monthly analyses are both strong and credible, 3SLS yields efficiently identified price coefficients. Are these

Repsol YPF-Harvard Kennedy School Fellows Research Papers

242

Table 15a Monthly demand using non-OPEC oil price and quantity of world oil production Dependent variable is quantity of world oil production (million barrels/day)

non-OPEC oil price (1982–1984$/barrel)

summer dummy (Jun, Jul, Aug) winter dummy (Dec, Jan, Feb)

OLS (1)

SUR (2)

2SLS (3)

3SLS (4)

–0.12** (0.04)

–0.31*** (0.03)

–0.02 (0.13)

0.06 (0.09)

–0.51 (0.53) 0.27 (0.53)

–0.35 (0.22) 0.17 (0.21)

–0.30 (0.29) 0.14 (0.28)

–0.12* (0.06) 2.18 (1.13) 0.54 (0.56) –0.35 (0.41) –0.31** (0.10) 0.20* (0.08) 0.07*** (0.01) 0.21*** (0.05) –0.36* (0.14) –0.04 (0.09) 0.08** (0.03) 0.18** (0.06) –0.68* (0.28) 44.23*** (9.28) 0.43 240

–0.71 (0.38) –2.95 (14.35) 5.80 (3.98) 2.88 (4.11) –0.37 (0.89) 1.44* (0.56) –0.07 (0.06) –0.31 (0.27) –0.14 (1.26) –0.76 (0.92) 0.46** (0.15) –0.35 (0.28) –2.15 (1.98) 38.38 (89.11) 0.91 240

–0.82*** (0.19) 18.84* (8.75) 7.20* (3.08) –1.98 (3.01) –1.11 (0.63) 1.48*** (0.27) 0.26*** (0.03) 0.86*** (0.18) –0.99 (0.89) –1.35* (0.65) 0.73*** (0.08) 0.58*** (0.17) –3.56* (1.55) –107.5 (68.32) 0.84 240

Monthly Covariates –0.40 (0.21) 0.20 (0.21)

Annual Covariates GDP, world (trillion 1982–1984$) –0.88** (0.31) population, world (billions) –11.92 (8.66) commercial energy use, world (million kt of oil equivalent) 4.28 (3.44) electricity production, world (trillion kwh) 5.23 (2.81) electricity production from oil, world (%) –0.98* (0.45) electricity production from oil, high-inc OECD (%) 1.76*** (0.38) electricity production from oil, high-inc non-OECD (%) –0.04 (0.05) electricity production from oil, Mid East & N. Afr. (%) –0.39 (0.25) electricity production from natural gas, world (%) –0.96 (0.70) electricity production from natural gas, high-inc OECD (%) –0.15 (0.48) 0.42** electricity production from natural gas, high-inc non-OECD (%) (0.14) electricity production from natural gas, Mid East & N.Afr. (%) –0.37 (0.28) natural gas reserves, world (1015 cubic ft) –3.38 (1.21) 95.06 constant (52.11) adj. R squared 0.92 240 # observations

Notes: Standard errors are in parentheses. Significance codes: *5% level, **1% level, and ***0.1% level. For 2SLS and 3SLS estimations, the instruments used are: total world rig count; summer dummy; winter dummy; world GDP; world population; world commercial energy use; world electricity production; electricity production from oil for world, high-income OECD, high-income non-OECD, and Middle East and North Africa; electricity production from natural gas for world, high-income OECD, high-income non-OECD, and Middle East and North Africa; world crude oil reserves; and world natural gas reserves. For the SUR and 3SLS estimations, demand and supply equations corresponding to the same choice of price and quantity variables were estimated jointly, the maximum number of iterations was 1e6, and the toleration level for each iteration was 1e-5.

C.-Y. Cynthia Lin

243

Table 15b Monthly supply using non-OPEC oil price and quantity of world oil production

non-OPEC oil price (1982–1984$/barrel)

total world rig count (100 rigs) summer dummy (Jun, Jul, Aug) winter dummy (Dec, Jan, Feb)

GDP, Mid East & N..Afr. (trillion 1982–1984$) crude oil reserves, world (billion barrels) constant adj. R squared # observations

Dependent variable is quantity of world oil production (million barrels/day) OLS (1) SUR (2) 2SLS (3) 3SLS (4) –0.12** –0.30*** –0.10 –0.14*** (0.02) (0.05) (0.03) (0.04) Monthly Covariates 0.03 (0.03) –0.39 (0.27) 0.13 (0.28) Annual Covariates 47.75*** (3.64) 0.029*** (0.001) 17.79*** (1.92) 0.85 240

0.009* (0.004) –0.49 (0.47) 0.24 (0.47)

0.02 (0.04) –0.39 (0.28) 0.16 (0.28)

0.04* (0.02) –0.40 (0.27) 0.12 (0.28)

8.67*** (1.57) 0.005*** (0.001) 56.52*** (1.48) 0.56 240

49.62*** (3.71) 0.03*** (0.00) 17.02*** (1.94) 0.85 240

49.92*** (3.68) 0.030*** (0.001) 16.94*** (1.94) 0.85) 240

Notes: Standard errors are in parentheses. Significance codes: *5% level, **1% level, and ***0.1% level. For 2SLS and 3SLS estimations, the instruments used are: total world rig count; summer dummy; winter dummy; world GDP; world population; world commercial energy use; world electricity production; electricity production from oil for world, high-income OECD, high-income non-OECD, and Middle East and North Africa; electricity production from natural gas for world, high-income OECD, high-income non-OECD, and Middle East and North Africa; world crude oil reserves; and world natural gas reserves. For the SUR and 3SLS estimations, demand and supply equations corresponding to the same choice of price and quantity variables were estimated jointly, the maximum number of iterations was 1e6, and the toleration level for each iteration was 1e-5.

coefficients consistent with economic theory? Well, yes and no. Non-OPEC oil demand does indeed exhibit a negative slope with respect to non-OPEC oil price, while world oil demand is inelastic to both OPEC price and non-OPEC price. Thus, both non-OPEC and world demand functions are consistent with economic theory, which predicts that demand should be (weakly) downward-sloping. Moreover, OPEC supply in levels form appears inelastic to OPEC price, and log non-OPEC supply appears inelastic to log non-OPEC price, which are both consistent with the theoretical prediction that supply should be (weakly) upward-sloping. However, the 3SLS estimates for OPEC demand (in both levels and logs), OPEC supply (in logs), nonOPEC supply (in levels), and world supply (in both levels and logs) all yield price coefficients of the wrong sign. It thus appears that OPEC demand and most specifications for supply do not satisfy the simply theoretical assumptions of a static perfectly competitive oil market.

6.

Conclusion

Is it possible to obtain efficiently identified estimates of aggregate supply and demand curves for world oil under the assumptions of a static and perfectly competitive world oil market, or is the endeavor doomed to yield a dry hole? The answer at first blush appears mixed. When annual data spanning 1965–2000 were used, the instruments chosen, while credible, were weak, and, as a consequence, neither supply nor demand was identified. In contrast, with monthly data spanning 1981–2000, the instruments chosen were both strong and credible. However, although the efficiently identified monthly supply and demand curves were consistent with economic theory in the cases of world demand, non-OPEC demand and two specifications for supply, this was not the case for either OPEC demand or for most specifications for supply. That even the use of strong and credible instruments and of joint estimation did not yield

Repsol YPF-Harvard Kennedy School Fellows Research Papers

244

Table 16a Monthly demand using non-OPEC oil price and quantity of non-OPEC oil production

non-OPEC oil price (1982–1984$/barrel)

summer dummy (Jun, Jul, Aug) winter dummy (Dec, Jan, Feb)

Dependent variable is quantity of non-OPEC oil production (million barrels/day) OLS (1) SUR (2) 2SLS (3) 3SLS (4) –0.04** –0.14*** –0.01 –1.25*** (0.01) (0.05) (0.26) (0.01) Monthly Covariates –0.33*** (0.08) 0.19* (0.08)

Annual Covariates GDP, world (trillion 1982–1984$) –0.18 (0.12) population, world (billions) 12.67*** (3.45) commercial energy use, world (million kt of oil equivalent) 4.60*** (1.37) electricity production, world (trillion kwh) –3.14** (1.12) electricity production from oil, world (%) –0.10 (0.18) electricity production from oil, high-inc OECD (%) –0.37* (0.15) electricity production from oil, high-inc non-OECD (%) 0.03 (0.02) electricity production from oil, Mid East & N. Afr. (%) 0.21* (0.10) electricity production from natural gas, world (%) –0.53 (0.28) electricity production from natural gas, high-inc OECD (%) 0.48* (0.19) –0.06 electricity production from natural gas, high-inc non-OECD (%) (0.06) electricity production from natural gas, Mid East & N.Afr. (%) 0.23* (0.11) natural gas reserves, world (1015 cubic ft) –3.99*** (0.48) constant –25.79 (20.76) adj. R squared 0.88 # observations 240

–0.38 (0.23) 0.23 (0.23)

–0.31*** (0.09) 0.18* (0.08)

–0.97 (0.60) 0.60 (0.59)

0.06* (0.03) –2.52*** (0.52) –0.99*** (0.26) 0.41* (0.19) 0.07 (0.05) –0.03 (0.04) –0.020*** (0.005) –0.11*** (0.02) 0.12 (0.06) 0.03 (0.04) –0.03* (0.01) –0.10*** (0.03) 0.90*** (0.13) 60.09*** (4.28) 0.07 240

–0.12 (0.15) 15.78** (5.70) 5.12** (1.58) –3.96* (1.63) 0.11 (0.35) –0.48* (0.22) 0.02 (0.02) 0.24* (0.11) –0.24 (0.50) 0.27 (0.36) –0.05 (0.06) 0.23* (0.11) 0.23* (0.11) –45.45 (35.38) 0.88 240

–1.75* (0.69) –98.74*** (28.58) –40.12*** (8.66) 36.64*** (8.76) –7.77*** (1.81) 3.01** (1.01) 0.22* (0.11) –2.61*** (0.53) –10.30*** (2.55) 7.10*** (1.86) –0.52 (0.28) –2.58*** (0.54) –12.96** (4.19) 947.4*** (189.1) –4.82 240

Notes: Standard errors are in parentheses. Significance codes: *5% level, **1% level, and ***0.1% level. For 2SLS and 3SLS estimations, the instruments used are: total world rig count; summer dummy; winter dummy; world GDP; world population; world commercial energy use; world electricity production; electricity production from oil for world, high-income OECD, high-income non-OECD, and Middle East and North Africa; electricity production from natural gas for world, high-income OECD, high-income non-OECD, and Middle East and North Africa; world crude oil reserves; and world natural gas reserves. For the SUR and 3SLS estimations, demand and supply equations corresponding to the same choice of price and quantity variables were estimated jointly, the maximum number of iterations was 1e6, and the toleration level for each iteration was 1e-5.

C.-Y. Cynthia Lin

245

Table 16b Monthly supply using non-OPEC oil price and quantity of non-OPEC oil production Dependent variable is non-OPEC oil production (million barrels/day) OLS (1)

SUR (2)

2SLS (3)

3SLS (4)

–0.10*** (0.02)

–0.12*** (0.01)

–0.16*** (0.03)

–0.15*** (0.03)

Monthly Covariates –0.06*** (0.02) –0.38* (0.17) 0.36* (0.17)

0.005** (0.002) –0.37 (0.20) 0.21 (0.20)

–0.03 (0.02) –0.40* (0.17) 0.30 (0.17)

–0.03 (0.02) –0.40* (0.17) 0.30 (0.17)

Annual Covariates 18.65*** (2.20) –0.0020** (0.0007) 35.41*** (1.16) 0.52 240

–2.28** (0.69) –0.0017*** (0.0003) 41.29*** (0.65) 0.32 240

20.19*** (2.27) –0.0021** (0.0008) 35.12*** (1.19) 0.51 240

20.15*** (2.26) –0.0021** (0.0008) 35.13*** (1.19) 0.51 240

non-OPEC oil price (1982–1984$/barrel)

total world rig count (100 rigs) summer dummy (Jun, Jul, Aug) winter dummy (Dec, Jan, Feb)

GDP, Mid East & N.Afr. (trillion 1982–1984$) crude oil reserves, world (billion barrels) constant adj. R squared # observations

Notes: Standard errors are in parentheses. Significance codes: *5% level, **1% level, and ***0.1% level. For 2SLS and 3SLS estimations, the instruments used are: total world rig count; summer dummy; winter dummy; world GDP; world population; world commercial energy use; world electricity production; electricity production from oil for world, high-income OECD, high-income non-OECD, and Middle East and North Africa; electricity production from natural gas for world, high-income OECD, high-income non-OECD, and Middle East and North Africa; world crude oil reserves; and world natural gas reserves. For the SUR and 3SLS estimations, demand and supply equations corresponding to the same choice of price and quantity variables were estimated jointly, the maximum number of iterations was 1e6, and the toleration level for each iteration was 1e-5.

price coefficients of the expected sign for OPEC demand or for most specifications for supply suggests that either my econometric specification or the underlying economic theory is incorrect. Is the econometric specification to blame? In addition to my Assumptions 1–3 on the instruments, which appear to be satisfied for my monthly analyses, another underlying assumption of my econometric model is that both demand and supply are linear with fixed coefficients and additive errors. This assumption could be relaxed in future work using methods such as those developed by Angrist et al. (2000), Manski, (1997), and by Newey, Powell and Vella (1999). To a first-order approximation, however, one would expect that imposing linearity and additivity should not affect the sign of the price coefficients. A potentially more devastating culprit for my counter-intuitive results, in addition to the underlying econometric assumptions of linearity and

additivity, is the theory itself. My model of the world oil market assumed that it was both static and perfectly competitive. However, the oil market is unlikely to be either. The first problematic theoretical assumption is that the oil market consists of static markets isolated in time. Because oil production is a capitalintensive process involving irreversible investments, and because oil itself is a nonrenewable resource whose extraction costs are likely to increase over time, the amount of oil supplied at any point in time is unlikely to be independent of the amount of oil supplied at any other point in time. Indeed, the Hotelling model of nonrenewable resource extraction predicts that, even if the market were perfectly competitive, market price would exceed marginal costs, with the difference reflecting the scarcity rent of the resource (Hotelling, 1931). Thus, oil supply is unlikely to be static. To better estimate the supply for oil, a dynamic model is needed.

Repsol YPF-Harvard Kennedy School Fellows Research Papers

246

Table 17a 3SLS estimates of monthly demand using ln(oil price) & ln(quantity of oil production) Dependent variable is log quantity of oil production (million barrels/day) for:

log OPEC oil price (1982–1984$/barrel)

(1) world

(2) OPEC

(3) world

(4) non-OPEC

–0.01 (0.02)

0.44** (0.15) 0.00 (0.02)

–0.19*** (0.05)

0.01 (0.02) –0.01 (0.02)

–0.01 (0.00) 0.00 (0.00)

–0.014* (0.006) 0.01 (0.01)

–0.04 (0.02) 3.16** (1.09) 1.63*** (0.39) –1.13** (0.35) 0.12 (0.06) 0.06* (0.03) 0.00 (0.00) 0.06** (0.02) 0.22* (0.10) –0.23*** (0.07) 0.06*** (0.01) 0.04 (0.02) 0.40* (0.17) –24.6*** (7.06) 0.59 240

–0.016*** (0.002) 0.34* (0.17) 0.16* (0.07) –0.06 (0.06) –0.02 (0.01) 0.027*** (0.004) 0.0040*** (0.0004) 0.013*** (0.003) –0.01 (0.02) –0.02* (0.01) 0.013*** (0.002) 0.008** (0.002) –0.05 (0.03) 1.11 (1.25) 0.84 240

–0.00 (0.01) –1.01** (0.36) –0.47*** (0.12) 0.42*** (0.11) –0.07*** (0.02) 0.01 (0.01) 0.00 (0.00) –0.012* (0.006) –0.11*** (0.03) 0.07*** (0.02) –0.013** (0.004) –0.01 (0.01) –0.23*** (0.05) 12.17*** (2.26) 0.17 240

non-OPEC oil price (1982–1984$/barrel)

summer dummy (Jun, Jul, Aug) winter dummy (Dec, Jan, Feb)

Monthly Covariates –0.01 (0.00) 0.00 (0.00)

Annual Covariates –0.017*** (0.002) population, world (billions) 0.32 (0.17) commercial energy use, world (million kt of oil equivalent) 0.16* (0.07) electricity production, world (trillion kwh) –0.05 (0.06) electricity production from oil, world (%) –0.02 (0.01) electricity production from oil, high-inc OECD (%) 0.028*** (0.003) electricity production from oil, high-inc non-OECD (%) 0.003*** (0.001) electricity production from oil, Mid East & N. Afr. (%) 0.011*** (0.002) electricity production from natural gas, world (%) –0.01 (0.02) –0.02 electricity production from natural gas, high-inc OECD (%) (0.01) electricity production from natural gas, high-inc non-OECD (%) 0.013*** (0.002) electricity production from natural gas, Mid East & N.Afr. (%) 0.006* (0.002) –0.05 natural gas reserves, world (1015 cubic ft) (0.03) constant 1.43 (1.25) adj. R squared 0.85 # observations 240 GDP, world (trillion 1982–1984$)

Notes: Standard errors are in parentheses. Significance codes: *5% level, **1% level, and ***0.1% level. The instruments used are: total world rig count; summer dummy; winter dummy; world GDP; world population; world commercial energy use; world electricity production; electricity production from oil for world, high-income OECD, high-income non-OECD, and Middle East and North Africa; electricity production from natural gas for world, high-income OECD, high-income non-OECD, and Middle East and North Africa; world crude oil reserves; and world natural gas reserves. Demand and supply equations corresponding to the same choice of price and quantity variables were estimated jointly, the maximum number of iterations was 1e6, and the toleration level for each iteration was 1e-5.

C.-Y. Cynthia Lin

247

Table 17b 3SLS estimates of monthly supply using ln(oil price) & ln(quantity of oil production) Dependent variable is log quantity of oil production (million barrels/day) for:

log OPEC oil price (1982–1984$/barrel)

(1) world

(2) OPEC

–0.06*** (0.01)

–0.12*** (0.04)

non-OPEC oil price (1982–1984$/barrel)

total world rig count (100 rigs) summer dummy (Jun, Jul, Aug) winter dummy (Dec, Jan, Feb)

GDP, Mid East & N.Afr. (trillion 1982–1984$) crude oil reserves, world (billion barrels) constant adj. R squared # observations

(3) world

(3) non-OPEC

–0.05*** (0.01)

–0.01 (0.01)

Monthly Covariates 0.0008* (0.0004) –0.01 (0.00) 0.00 (0.00)

0.008*** (0.001) –0.00 (0.01) –0.02 (0.01)

0.00 (0.00) –0.01 (0.00) 0.00 (0.00)

–3.5e-3*** (0.5e-3) –0.010* (0.005) 0.012* (0.005)

Annual Covariates 0.78*** (0.06) 4.6e-4*** (0.2e-4) 3.54*** (0.04) 0.86 240

1.06*** (0.17) 0.0014*** (0.0001) 1.65*** (0.13) 0.81 240

0.77*** (0.06) 4.8e-4*** (0.2e-4) 3.51*** (0.04) 0.85 240

0.53*** (0.06) –4.9e-5* (2.2e-5) 3.57*** (0.05) 0.49 240

Notes: Standard errors are in parentheses. Significance codes: *5% level, **1% level, and ***0.1% level. The instruments used are: total world rig count; summer dummy; winter dummy; world GDP; world population; world commercial energy use; world electricity production; electricity production from oil for world, high-income OECD, high-income non-OECD, and Middle East and North Africa; electricity production from natural gas for world, high-income OECD, high-income non-OECD, and Middle East and North Africa; world crude oil reserves; and world natural gas reserves. Demand and supply equations corresponding to the same choice of price and quantity variables were estimated jointly, the maximum number of iterations was 1e6, and the toleration level for each iteration was 1e-5.

The second problematic theoretical assumption is that the oil market is perfectly competitive. A more realistic model would account for the substantial market power exerted by the OPEC oil cartel. It thus appears that the theoretical assumptions of a static and perfectly competitive market may be unrealistic, especially in modeling the supply of oil. Indeed, the identified but inefficient 2SLS estimates for monthly demand all exhibited the appropriate negative sign; the sign for OPEC demand only flipped when OPEC demand was estimated jointly with OPEC supply in effort to obtain estimates that were not only identified but also efficient. Had the supply side been more realistically modeled, then joint estimation of demand and supply may not only have increased the efficiency and significance of the already-negative

and identified price coefficients for demand, but also yielded significant positive price coefficients for supply as well. Thus, attempting to efficiently identify aggregate oil supply and demand market in the context of a static and perfectly competitive oil market may indeed be a dry hole. It is a dry hole not because of the non-existence of either econometric methods or instruments to enable efficient identification, but rather because of the non-plausibility of the static perfect competition assumptions in the first place. An econometric model that incorporates either the dynamic or oligopolistic aspects of the oil market, or both, appears to be a more promising prospect for exploration and development, and one from which richer and more realistic results are likely to be extracted.

Repsol YPF-Harvard Kennedy School Fellows Research Papers

248

7.

Appendix A. Countries used in regional aggregates

The following lists the countries used in each of the regional aggregates for the World Bank Group World Development Indicators data. High-income economies are those in which 2002 GNI per capita was $9,076 or more. High-income OECD: Australia Austria Belgium Canada Denmark Finland France Germany Greece Iceland Ireland Italy

Japan Korea, Rep. Luxembourg Netherlands New Zealand Norway Portugal Spain Sweden Switzerland United Kingdom United States

High-income non OECD Andorra Antigua and Barbuda Aruba Bahamas, The Bahrain Barbados Bermuda Brunei Cayman Islands Channel Islands Cyprus Faeroe Islands French Polynesia Greenland Guam Hong Kong, China Isle of Man

Israel Kuwait Liechtenstein Macao, China Malta Monaco Netherlands Antilles New Caledonia Puerto Rico Qatar San Marino Singapore Slovenia Taiwan, China United Arab Emirates Virgin Islands (U.S.)

Middle East & North Africa (does not include highincome economies): Djibouti Egypt, Arab Rep. Iran, Islamic Rep. Iraq Jordan Lebanon Libya

Morocco Oman Saudi Arabia Syrian Arab Republic Tunisia West Bank and Gaza Yemen, Rep.

Endnotes 1. [email protected] 2. Angrist et al. (2000) investigate the consequences of relaxing both the linearity and additivity assumptions for the interpretation of linear instrumental variables estimators, and apply their approach to estimating the demand for fish. 3. Throughout this paper, identification and consistency, though technically not equivalent, will be used interchangeably. 4. Although the 2SLS estimates are consistent, they are still biased. No method for obtaining unbiased estimates of the structural parameters exists (Goldberger, 1991, p. 343). 5. Since 2SLS does not fully use all the available information to potentially enhance the efficiency of its estimates, it is sometimes referred to as “limited information estimation” (Ruud, 2000). 6. For this reason, 3SLS is sometimes referred to as “full information estimation” (Ruud, 2000). 7. In addition to oil supply and demand, the trendless nature of oil prices is yet another aspect of the oil market that continues to puzzle economists. See Lin (2004a) and Lin (2004b) for theoretical expositions of the latter empirical puzzle and attempts to rationalize the puzzle with theory. 8. I did obtain data on annual world petroleum production, but the data only spans 1973-1991. Because this data had so few observations, I did not use it for my annual analyses. 9. Averaging over the months for each year is more appropriate than summing because monthly production is reported as a rate: million barrels per day. 10. I chose the period length to optimally trade off the number of observations I could use with the number of variables with data available for the entire length of that period. 11. For most of the annual variables (e.g., population), it made more sense to use the actual annual observation for each month rather than dividing it by 12 to convert it to an average month’s share. Moreover, because using an average value for each month rather than the actual annual value only changes the

C.-Y. Cynthia Lin

12.

13.

14.

15.

16.

17.

18.

scale of the corresponding coefficient, for simplicity I chose to use the annual value for each month for all the variables. See Appendix A for a list of the countries included in each aggregate. I also collected data for GDP for not only the world aggregate, but for the other three regional aggregates of high-income OECD, highincome non-OECD, and Middle East and North Africa as well. However, because these series were highly collinear, I used only the world GDP in my estimations. Similarly, world population is highly correlated with population in high-income OECD and in high-income non-OECD. Likewise, world electricity production is highly correlated with than in high-income OECD and high-income non-OECD. World commercial energy use is highly correlated with commercial energy use in Middle East and North Africa, the only other of the four regional aggregates for which data for this variable was available. Lastly, world crude oil reserves and natural gas reserves are highly correlated with their respective variables for OPEC only. Because GDP in the Middle East and North Africa may be endogenous, I instrument for it in my estimations of the structural demand and supply equations. In the reduced-form regressions, however, I treat it as exogenous. Similarly, regressions could be run of the endogenous GDP in the Middle East and North Africa on the instruments to see if the instruments can also be appropriately used for this regressor as well. The year of the monthly market was too highly correlated with some of the annual covariates to be included as an additional market control. Because GDP in the Middle East and North Africa may be endogenous, I instrument for it in my estimations of the structural demand and supply equations. In the reduced-form regressions, however, I treat it as exogenous. Similarly, regressions could be run of the endogenous GDP in the Middle East and North Africa on the instruments to see if the instruments can also be appropriately used for this regressor as well. Ideally, natural gas price should be used as a regressor in the demand equation. As an extension to my work, I can look for appropriate monthly natural gas price data to include in my demand estimation.

References Adelman, M.A. (1962). Natural gas and the world petroleum market. The Journal of Industrial Economics, 10, 76–112. Angrist, J., Graddy, K., & Imbens, G.W. (2000). The inter-

249

pretation of instrumental variables estimators in simultaneous equations models with an application to the demand for fish. The Review of Economic Studies, 67 (3), 499–527. Berndt, E.R., & Wood, D.O. (1975). Technology, prices, and the derived demand for energy. The Review of Economics and Statistics, 57 (3), 259–268. Gately, D. (1984). A ten-year retrospective: OPEC and the world oil market. Journal of Economic Literature, 22 (3), 1100–1114. Goldberger, A.S. (1991). A course in econometrics. Cambridge, MA: Harvard University Press. Hausman, J.A. (1975). Project independence report: an appraisal of U.S. energy needs up to 1985. The Bell Journal of Economics, 6 (2), 517–551. Hotelling, H. The economics of exhaustible resources. The Journal of Political Economy, 39 (2), 137–175. Kennedy, M. An economic model of the world oil market. The Bell Journal of Economics and Management, 5 (2), 540–577. Krautkraemer, J. (1998). Nonrenewable resource scarcity. Journal of Economic Literature, 36 (4), 2065–2107. Lin, C.-Y.C. (2004a). Hotelling revisited: Oil prices and endogenous technological progress. Mimeo. Harvard University. Lin, C.-Y.C. (2004b). Steady-state growth in a Hotelling model of resource extraction. Mimeo. Harvard University. Mankiw, N.G. (1998). Principles of economics. Fort Worth, TX: Dryden Press. Manski, C.F. (1995). Identification problems in the social sciences. Cambridge, MA: Harvard University Press. Manski, C.F. (1997). Monotone treatment response. Econometrica, 65 (6), 1311–1334. Newey, W.K., Powell, J.L., &Vella, F. (1999). Nonparametric estimation of triangular simultaneous equations models. Econometrica, 67 (3), 565–603. Nordhaus, W.D., Houthakker, H.S., & Sachs, J.D. (1980). Oil and economic performance in industrial countries. Brookings Papers on Economic Activity, 1980 (2), 341–399. Oettinger, G.S. (1999). An empirical analysis of the daily labor supply of stadium vendors. The Journal of Political Economy, 107 (2), 360–392. R Development Core Team. (2003). R: A language and environment for statistical computing [Computer programming software]. Vienna, Austria: R Foundation for Statistical Computing. URL: http://www.R-project.org. Ruud, P.A. (2000). An introduction to classical econometric theory. Oxford: Oxford University Press. Yergin, D. (1992). The prize: the epic quest for oil, money, and power. New York: Free Press.

13

Optimal World Oil Extraction: Calibrating and Simulating the Hotelling Model C.-Y. Cynthia Lin1 Department of Economics, Harvard University

exhibit stock effects. Numerical solutions are generated for various specifications of the elasticity of demand for both isoelastic demand and linear demand under each of two possible market structures: perfect competition and monopoly. My simulations enable me to examine the following questions. First, how well does the simple Hotelling model appear to explain historical data? And second, how are future world oil prices and extraction rates predicted to evolve? From among the various specifications I tried, the model best fits actual data when the oil market is perfectly competitive and when demand is inelastic. However, even the best-fit specification fails to adequately explain the data, which suggests that a richer theoretical model may be needed.

Abstract This paper uses data on world oil price and consumption to calibrate a Hotelling model of optimal resource extraction with unlimited potential reserves when costs This paper was prepared in residence under a Repsol YPFHarvard Kennedy School Fellowship. I would like to thank Gary Chamberlain for his advice and guidance throughout this project. This paper also benefited from discussions with Martin Weitzman, William Hogan and Howard Stone. The time series data used in this study were acquired with the help of Brian Greene and with funds from the Littauer Library at Harvard University. I am indebted to Bijan Mossavar-Rahmani (Chairman, Mondoil Corporation) and William Hogan for arranging for me to visit Apache Corporation’s headquarters in Houston and a drilling rig and production platform offshore of Louisiana, and for their support of my research, and I thank the Repsol-YPF-Harvard Kennedy School Energy Fellows Program for providing travel funds. I thank Mark Bauer (Reservoir Engineering Manager, Apache), Robert Dye (VP, Apache), Steve Farris (President, CEO & COO, Apache), Richard Gould (VP, Wells Fargo), Paul Griesedieck (Manager, Apache), Thomas Halsey (Corporate Strategic Research, ExxonMobil Research and Engineering), Becky Harden (Land Manager, Apache), David Higgins (Director, Apache), Jon Jeppesen (Sr. VP, Apache), Adrian Lajous (President, Oxford Institute for Energy Studies), Kenneth McMinn (Offshore District Production Manager, Apache), Kregg Olson (Director, Apache), and Bob Tippee (Editor, Oil and Gas Journal) for enlightening discussions about the petroleum industry. I thank Derrick Martin for flying me offshore by helicopter, Mike Thibodeaux for giving me a tour of the production platform, Joey Bridges for giving me a tour of the drilling rig, and especially Billy Ebarb (Production Superintendent, Apache) for accompanying me throughout the entire offshore trip. William Horvath (Energy Information Administration) provided data that might be useful for future versions of this paper. Aloulou Fawzi (Energy Information Administration) and Jeff Obermiller (American Petroleum Institute) provided leads for possible data sources. I received financial support from an EPA Science to Achieve Results (STAR) graduate fellowship and a National Science Foundation (NSF) graduate research fellowship. All errors are my own.

1.

Introduction

The problem of optimal nonrenewable resource extraction was first examined by Hotelling (1931), whose basic model predicted that the shadow price of the resource stock, which is an economic measure of the scarcity of the resource, should grow at the rate of interest. Since then, economists have expanded Hotelling’s basic theoretical framework to allow for more realistic features such as increasing extraction costs (Hanson, 1980; Solow & Wan, 1976), unlimited potential reserves (Pindyck, 1978), market imperfections (Khalatbari, 1977; Stiglitz, 1976; Sweeney, 1977), technological progress (Lin, 2004a,b), and uncertainty (Hoel, 1978; Pindyck, 1980). In addition to enriching the theory, economists have also taken the Hotelling model to data. For example, much work has also been done in attempt to measure the shadow price of the resource (see e.g., Devarajan & Fisher, 1982; Halvorsen & Smith, 1984; Lasserre, 1985). Some studies have also attempted to use data to test Hotelling’s theory. Empirical tests of the dynamic efficiency conditions of the Hotelling model under 251

252

perfect competition have been applied to data on a hard rock mining firm (Farrow, 1985), on Canadian metal mining firms (Halvorsen & Smith, 1991), and on Canadian copper mining firms (Young, 1992). However, owing in part to limitations on the cost data needed, such tests have yet to be applied to the world oil industry.2 Moreover, an empirical test of dynamic efficiency of world oil extraction may also need to relax the assumption of perfect competition.3 In this paper, I use data on world oil price and consumption to calibrate a Hotelling model of optimal oil extraction when reserves are unlimited and when costs depend on the cumulative stock extracted. I use the calibrated model to simulate solution trajectories under both perfect competition and monopoly for various specifications of the elasticity of demand for both isoelastic demand and linear demand. My paper follows most closely the work of Pindyck (1978), who first develops a theoretical model that allows for unlimited potential reserves and that requires resource producers to simultaneously determine optimal rates of exploration and production, and who then examines numerically the characteristics of the competitive and monopoly solutions to his model using data for oil in the Permian region of Texas over the period 1965–1974. Though this paper also examines both competitive and monopoly solutions and allows for unlimited reserves, it differs from that of Pindyck (1978) in several ways. First, the data I use to calibrate my model is for world oil over the period 1965–2001, which spans both a wider geographic area and a longer period of time than Pindyck’s data does. Second, instead of Pindyck’s more complex—-and perhaps more realistic—-model that allows producers to choose the rates of both exploration and extraction, I use Hotelling’s more basic model of allowing producers to choose extraction rates only. I thus can gauge whether Hotelling’s simpler model is sufficient to explain the historical data. Third, while Pindyck assumes the demand is linear, I examine the both the linear demand case and the isoelastic demand case. The fourth way in which I expand Pindyck’s work is that I present my results under a range of values of the demand elasticity instead of just choosing one.4 A fifth difference is that, while this paper compares the predicted trajectories for both price and extraction

Repsol YPF-Harvard Kennedy School Fellows Research Papers

with actual data and varies the parameters in attempt to match the data, Pindyck only compares the optimal values of well drilling and price to their historical values for the one set of parameters he used.5 My simulations enable me to examine the following questions. First, how well does the simple Hotelling model appear to explain historical data? And second, how are future world oil prices and extraction rates predicted to evolve? From among the various specifications I tried, the model best fits actual data when the oil market is perfectly competitive and when demand is inelastic. Under these assumptions, real oil price should fall in the range $117–160/barrel over the years 2000–2010 and extraction should be roughly constant at 59.8 million barrels/day over the entire simulation period 1857–2300. However, even the best-fit specification fails to adequately explain the data, which suggests that a richer theoretical model may be needed. The balance of this paper proceeds as follows. In Section 2, I present my basic Hotelling model. In Section 3, I describe my data. In Section 4, I explain the functional form assumptions and calibration methods used for my simulations. My results are presented in Section 5. Section 6 concludes. 2.

The Basic Hotelling Model

In this section, I present my theoretical model of optimal nonrenewable resource extraction under both perfect competition and monopoly. The notation follows closely that used by Weitzman (2003). 2.1

The General Framework

Suppose there are t oil markets indexed by t = 1,...,T. For each time t, the supply of oil is given by E(t), the total extraction flow in units of oil per unit time at time t.6 Let X(t) denote the total cumulative stock of oil extracted at time t: (1) where the initial stock X(0) is taken as given. The market price of oil at time t is P(t). The demand for oil when the market price is P is given by the demand function D(P). Markets are assumed to clear, which means that, at each time t, the price P(t) acts to equate supply and demand:

C.-Y. Cynthia Lin

253

(2) The cost of extracting E units of oil when the total stock already extracted is X is given by C(X,E). I use the term ‘’stock effects’’ to refer to the dependence of extraction cost on the stock X of reserve extracted. There are several possible reasons why this dependence is positive. First, extraction costs may increase with the cumulative stock extracted if the resource needed to be extracted from greater depths as it was being depleted. Second, costs may increase if well pressure declined as more of the reserve was depleted. Third, since different grades of oil may differ in their extraction costs, and since the cheaper grades are likely to be mined to exhaustion before the more expensive grades are mined, the cost of extraction may increase as the cheaper grades are exhausted, and therefore as the total stock already extracted increased. Let p(t) denote the non-negative current-value shadow price measuring the value of a unit of reserve at time t. This shadow price is known by a variety of terms, including ‘’marginal user cost’’, because it measures the opportunity cost of extracting the resource; ‘’in situ value’’, because it measures the marginal value of leaving an additional unit of resource in the ground; ‘’scarcity rent’’, because it is an economic measure of scarcity, and ‘’dynamic rent’’, to reflect the difference between price and marginal extraction cost (Krautkraemer, 1998; Weitzman, 2003). As the reader will soon see, the shadow price plays a crucial role in the Hotelling model. The competitive interest rate is ρ. 2.2

Perfect Competition

When the oil market is perfectly competitive, extraction E(t) represents the total amount of oil extracted by all the firms in the market at a given point in time.7 The total benefits p(·) from oil at time t is given by the area under the demand curve: (3) This area measures the gross consumer surplus, and is a measure of the consumers’ total moneymetricized willingness-to-pay. As shown in Weitzman (2003), using the area under the

demand curve in place of revenue yields the same outcome as a perfectly competitive market.8 For mathematical simplicity, I thus choose to model the perfectly competitive firm’s maximization problem using the area under the demand curve. The per-period net benefit G(X,E) from extracting E units of oil when the total stock already extracted is X is given by total benefits minus total costs: (4) The social planner’s optimal control problem, which yields the same solution as would arise in perfect competition, is to choose her extraction profile {E(t)} to maximize the present discounted value of her entire stream of net benefits, given her initial stock X(0) and given the relationship between her extraction E(t) and the cumulative stock extracted X(t), and subject to the constraints that both extraction and stock are nonnegative. Her problem is thus given by:

(5) where q(t) < 0 is the multiplier associated with the equation of motion for the total stock X(t) of oil extracted. The absolute value of this multiplier is precisely the shadow price p(t) of the reserve: (6) From the Maximum Principle, the first-order necessary conditions for a feasible trajectory {X*(t), E*(t)} to be optimal are:9 (7) (8) (9) Condition [#1] states that, at each time t, the shadow price p(t) must equal the competitive market price p(t) minus the marginal cost of extraction ; this condition is needed to ensure static optimality at each point in time. Condition

Repsol YPF-Harvard Kennedy School Fellows Research Papers

254

[#2] governs how the shadow price p(t) must evolve over time; conditions [#1] and [#2] combined are needed to ensure intertemporal optimality over all finite subperiods. Condition [#3], the transversality condition, is required for the solution to be dynamically optimal over the entire infinite horizon (Weitzman, 2003). One can use the constraints, market clearing condition and first-order conditions to reformulate the Hotelling problem into the following ordinary differential equation boundary value problem:

which yields the following first-order conditions: (14) (15) (16) In order to ensure static optimality, condition [#1] requires that in monopoly, unlike in perfect competition, the shadow price p(t) must equal marginal revenue Φ ′(E(t)) minus marginal cost at every time t. Conditions [#2] and [#3] are the same in monopoly as in perfect competition. As before, one can use the constraints, market clearing condition and first-order conditions to reformulate the Hotelling problem into an ordinary differential equation boundary value problem, where now the problem is given by:

(10) The solution to the boundary value problem (10) is equivalent to that of the optimal control problem (5); I thus derive solutions to the latter problem by solving the former. 2.3

Monopoly

When oil is produced by a single monopolist rather than a multitude of perfectly competitive firms, the total benefits of oil production no longer equal the area under the demand curve, but rather total revenue instead. Total revenue Φ (·) at time t is given by: (11) As a consequence, the monopolist’s per-period profit G(X,E) is given by: (12) and his optimal control problem is given by:

(13)

(17) The solution to the boundary value problem (17) is equivalent to that of the optimal control problem (13). Having explained my theoretical model, I now describe the data I use both for calibrating my model and for assessing its validity.

C.-Y. Cynthia Lin

3.

255

Table 2 Estimates of the demand elasticity ε

Data

In order to calibrate my theoretical model and assess its validity, I use annual data spanning the years 1965–2001. For oil price P, I use the real annual spot price for crude oil, averaged over the Brent, Dubai, and West Texas Intermediate (WTI) prices. This average price time series was obtained from the World Bank and deflated to 1982–1984 U.S. dollars using the consumer price index (CPI).10 For oil quantity E, I use world oil consumption as reported by BP.11 Table 1 provides summary statistics for my price and quantity data. Table 1 Summary statistics Variable Real world oil price (1982–1984$/barrel) World oil consumption (million barrels/day)

mean

s.d.

16.00 11.03

min

max

3.12

44.75

trend

0.13 (0.17) 59.33 11.54 31.23 75.45 0.99* (0.07)

Notes: The trend is the coefficient on year when the variable is regressed on year and a constant. Significance codes: * 0.1% level.

As can be seen from Table 1, real world oil price has no significant trend over 1965–2001. The trendless nature of oil price over time is in accordance with many empirical studies (see Krautkraemer, 1998, & references therein).12 In contrast, world oil consumption is increasing over 1965–2001 by nearly 1 million barrels/day each year. 4.

Calibration

In this section, I describe the functional form and parameter assumptions I use to calibrate my model. For my cost function C(X,E), I use a cost function estimated by Chakravorty, Roumasset and Tse (1997). Their estimate was derived from world data on proven and estimated reserves and on extraction costs compiled by the East-West Center Energy Program. After trying a variety of functional forms, the marginal cost function they found to fit the data was of the form: (18) where, when X is in units of 1015 British thermal units (Btu) and costs are in units of dollars per million Btu, the parameter values are given by c1 = 0.1774 and c2 = 0.000217.13 I therefore use as my cost function:14

Estimate of ε

Source

–0.49 to –0.45 –0.5 0.0 –0.3 –0.5 to –0.1

Bernt & Wood (1975, p. 265) Edmonson (1975, p. 172) Lin (2005)15 Nordhaus (1980, p. 347) Pindyck (1978, p. 857)

(19) For demand, a crucial parameter is the elasticity of demand ε. Possible values for the elasticity, according to various empirical studies and surveys, are shown in Table 2; they range from –0.5 to –0.0. I run my model under different values of ε, mostly in the range ε ∈ [–2,0], except as noted below. I use two different forms for the demand function D(·). In the first form, I assume that demand is isoelastic: (20) where the parameter d2 is the absolute value of the elasticity of demand: (21) Isoelastic demand functions are common in theoretical Hotelling models, as they can lend themselves to analytic solutions (see e.g., Lin 2004b; Stiglitz, 1978). For the monopoly case, the shadow price is positive only if ε < –1, which is less elastic than the range of the estimates in Table 2. Thus, in order for any extraction to occur under monopoly when demand is isoelastic, the demand function needs to be more elastic than reported in previous studies. Since d1 satisfies: (22) I calibrate d1 by using data on world oil consumption and world oil price for a given base year τ for E and P, respectively. The second form of the demand function that I use is linear demand: (23) when the slope d2 as a function of the elasticity of demand is given by: (24)

Repsol YPF-Harvard Kennedy School Fellows Research Papers

256

and the intercept d1 as a function of the slope is given by: (25) Pindyck (1978) uses a linear demand function in his simulations. I calibrate d1 and d2 by using data on world oil consumption and world oil price for a given base year τ for E and P, respectively. I use ρ = 0.05 as my market interest rate. According to Zimmermann (1951, p. 512), the first year in which world petroleum production data is available is 1857, when 2000 barrels of oil were produced. I therefore use 1857 as my initial year, corresponding to t = 0, and set my initial cumulative stock of extracted resource, Xo, to be 2000 barrels.15 I assume that the final year, corresponding to t = y, occurs in 2300. For calibrating the demand function, I use 1981 as my base year τ.16 5.

Simulation Results

In this section, I present my results under the two market structures of perfect competition and monopoly for both isoelastic demand and linear demand, and compare my results to the actual data. Although my model generates a solution for the years 1857–2300, I use my model’s simulated solution for the years 1965–2001, the period spanned by my actual data, when comparing my model to data.

Figure 1

5.1

Isoelastic Demand

5.1.1 Results under perfect competition when demand is isoelastic In analyzing the results under perfect competition when demand is isoelastic, I first attempt to discern which elasticity in the range ε ∈ [–2,0] generates trajectories for market price and for extraction that best fit the data. I use three different measures to examine the fit of the model. My first measure of fit is based on the summary statistics. A solution fits the data well if its summary statistics are similar to those of the actual data. Figures 1 and 2 plot the means, standard deviations, maxima, and minima over the years 1965–2001 of the optimal values for market price P(t) and extraction E(t), respectively, as a function of the demand elasticity ε. Error bars denote the standard deviation. For comparison, the dotted lines in the two figures show the same summary statistics over the same years for the actual world oil price and world oil consumption, respectively. For all values of the demand elasticity considered, the model’s optimal market price lies above the actual price levels. The model’s optimal extraction falls in the range of actual data when demand is less elastic, but is lower than actual demand when demand is more elastic. When demand is inelastic, the model matches the mean extraction well, but fails to capture the variance in actual extraction. Thus, when a comparison of summary statistics is used as a measure of fit, ε = 0 appears to best fit the data.

Figure 2

C.-Y. Cynthia Lin

257

Figure 3

Figure 4

In addition to a summary statistic comparison, my second measure of fit is the mean squared error (MSE). A lower MSE indicates a better fit. Figure 3 presents the MSE between the model’s solution market price and actual market price for the period 1965–2001. The MSE is lowest at ε = 0 and at ε = –2. Figure 4 presents the MSE between the model’s solution extraction and actual world oil consumption for the period 1965–2001. Unlike for market price, the MSE for extraction is lower when demand is less elastic. Using MSE as a measure of fit, either ε = 0 or ε = –2 appears to best fit the data. My third measure of fit is correlation. A higher correlation indicates a better fit. Figure 5 presents the correlation coefficient between the my model’s solution and actual data over 1965–2001 for both market price and extraction as a function of the demand elasticity. The price correlation is low and decreases slightly as demand becomes more inelastic. Predicted extraction is highly negatively correlated with actual extraction and invariant to elasticity except when ε = 0, in which case the correlation is negligible (correlation = –0.00). Using correlation as a measure of fit, ε = 0 appears to best fit the data. Based on my three measures of fit, it appears that perfect competition with isoelastic demand yields results that best fit the data when the demand elasticity ε is either 0 or –2. Figures 6 and 7 plot the solution trajectories for market price P(t), cumulative extraction X(t), shadow price p(t), and extraction E(t) when the demand elasticity ε is 0 and 2, respectively. For both elasticities, market price is increasing while

extraction is weakly decreasing, and both trajectories are monotonic. In contrast, actual world oil price is trendless but highly volatile, actual world oil consumption is increasing, and neither trajectory is strictly monotonic. The model appears to best fit the data when ε = 0. For both elasticities, the shadow price eventually goes to zero so that at terminal time T, the resource is no longer scarce. How much oil will eventually be extracted? Figure 8 plots the total stock extracted X(T) as a function of demand elasticity. The less elastic the demand, the more oil will be extracted in total. If ε = 0, a total of 9.67 trillion barrels of oil will eventually be extraction; if ε = –2, a total of 4.07 trillion barrels will be extracted. I now compare my results from perfect competition when demand is isoelastic to those from monopoly when demand is isoelastic. Figure 5

Repsol YPF-Harvard Kennedy School Fellows Research Papers

258

Figure 6

Figure 7

Results under monopoly when demand is isoelastic As explained above, when demand is isoelastic, a monopolist would only extract from a reserve when ε < –1. The results in this section are thus generated for ε ∈ [–2.0,–1.1]. I first find the elasticities that yield results that best fit the data under my three measures of fit: summary statistic comparison, MSE and correlation. As seen in Figure 9, the model yields market price profiles that best match the summary statistics of actual data for the years 1965–2001 when ε = –2.0. As seen in Figure 10, extraction best matches the summary statistics of actual data when ε = –1.1. If MSE is the measure of fit, then market price is best fit to data when ε = –2.0 (Figure 11), and extraction is best fit to data when ε = –1.2 (Figure 12). If correlations are used as a measure of fit, all elasticities yield results that fit the data equally well (Figure 13), with predicted market price weakly correlated with actual data and predicted extraction strongly anti-correlated with actual data. Thus, the model appears to best fit the data when ε = –1.1, ε = –1.2 and ε = –2.0, though the fit is poor in all cases. Figures 14, 15 and 16 present the results under the best-fit elasticities of ε = –1.1, ε = –1.2 and ε = –2.0, respectively. As with perfect competition under isoelastic demand, monopoly under isoelastic demand yields increasing market price and decreasing extraction. As in perfect competition, the shadow price eventually goes to zero so that at terminal time T, the resource is no longer scarce.

Figure 17 plots the total stock extracted as a function of elasticity. In contrast to perfect competition, the total stock extracted increases as demand becomes more elastic. Moreover, for each elasticity, the monopolist extracts a smaller amount in total than does a perfectly competitive industry. Just as in a standard static market, a monopolist produces less for each time-t market than would a perfectly competitive industry in order to increase prices and thus his revenue.17 In comparing the best-fit scenarios for perfect competition with those for monopoly, both when demand is isoelastic, it appears that the model is best fit by perfect competition with inelastic demand. I now examine the results when demand is linear rather than isoelastic.

5.1.2

Figure 8

C.-Y. Cynthia Lin

259

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Repsol YPF-Harvard Kennedy School Fellows Research Papers

260

Figure 14

Figure 15

Figure 16

Figure 17

Figure 18

Figure 19

C.-Y. Cynthia Lin

261

Figure 20

5.2 5.2.1

Linear Demand

Results under perfect competition when demand is linear How well does perfect competition fit the data when demand is linear, and under which demand elasticity is the data best fit? Figures 18 and 19 plot the summary statistics for the predicted price and extraction, respectively, for the years 1965–2001, and compares them to the analogous statistics for the actual data. The predicted market price is too high for all the elasticities, while extraction best fits the range of the actual data when demand is inelastic (i.e., ε = 0). When MSE is used to measure fit, market price is best matched to data when ε = –2 (Figure 20), while extraction is best matched to data when ε = 0 (Figure 21). When correlation is used to measure fit, the fit of market price to data decreases slightly as demand becomes more inelastic, while extraction is best fit to data when ε = 0 (Figure 22). Thus, perfect competition under linear demand appears to best fit the data when ε = 0 or ε = –2. Figures 23 and 24 plot the solution trajectories for the best-fit scenarios of ε = 0 and ε = –2, respectively. In both scenarios, market price is weakly increasing and extraction is weakly increasing. As with isoelastic demand, perfect competition best fits the data when ε = 0. Also as before, the shadow price goes to zero. Figure 25 plots the total stock extracted as a function of elasticity. As before, more total stock is extracted when demand is less elastic. When compared with the analogous isoelastic demand

Figure 21

results, the total stock extracted is weakly lower when demand is linear than when it is isoelastic for any given elasticity. 5.2.2 Results under monopoly when demand is linear I now examine the fit of the model under monopoly and linear demand.18 According to the summary statistics, model price best fits the data when ε = –2 (Figure 26). Model extraction best fits the mean of the data when ε = –0.7 and the variance of the data when ε = –2 (Figure 27). According to MSE, model price best fits data when ε = –2 (Figure 28), while model extraction best fits the data when ε = –0.7 (Figure 29). According to correlations, all elasticities yield roughly the same fit to data (Figure 30). The model thus appears to best fit data when ε = –0.7 or ε = –2. Figure 22

Repsol YPF-Harvard Kennedy School Fellows Research Papers

262

Figure 23

Figure 24

Figure 25

Figure 26

Figure 27

Figure 28

C.-Y. Cynthia Lin

263

Figure 29

Figure 30

Figure 31

Figure 32

Figures 31 and 32 plot the solution trajectories for the best-fit scenarios of ε = –0.7 and ε = –2, respectively. For both scenarios, market price is weakly increasing while extraction is weakly decreasing. As with the previous specifications, the shadow price eventually goes to zero so that at terminal time T, the resource is no longer scarce. Figure 33 plots the total stock extracted as a function of elasticity. As before, total stock extracted increases as demand becomes more inelastic, and the monopolist extracts less than the perfectly competitive industry for any given elasticity under linear demand. From among the various specifications I tried, the model best fits actual data when the oil market is perfectly competitive and when demand is inelastic. For perfect competition and inelastic

demand, the model predictions for price and extraction for the years of actual data are compared with actual data in Figures 34 and 35 for isoelastic and linear demand, respectively. Under these assumptions, real oil price should fall in the range $117–160/barrel over the years 2000–2010 and extraction should be roughly constant at 59.8 million barrels/day over the entire simulation period 1857–2300. Even for the best-fit specification, however, the simple Hotelling model fails to adequately explain the data. 6.

Conclusion

This paper uses data on world oil price and consumption to calibrate a Hotelling model of optimal resource extraction with unlimited potential

Repsol YPF-Harvard Kennedy School Fellows Research Papers

264

Figure 33

reserves when costs exhibit stock effects. Which specification for demand (isoelastic or linear), market structure (perfect competition or monopoly) and demand elasticity yields results that best fit the historical data? From among the various specifications I tried, the model best fits actual data when the oil market is perfectly competitive and when demand is inelastic. Under these assumptions, real oil price should fall in the range $117–160/barrel over the years 2000–2010 and extraction should be roughly constant at 59.8 million barrels/day over the entire simulation period 1857–2300. So, how well does the simple Hotelling model presented appear to explain historical data? Figure 34

Unfortunately, the answer is somewhat disappointing. Based on my three measures of fit (summary statistic comparison, mean squared error and correlation), none of the simulations appear to adequately explain the historical data. In particular, the theoretical model fails to capture two salient features of data. The first feature that is not captured by the theoretical model is the highly volatile but roughly trendless nature of market prices. The second feature that is not captured by the model is the increasing trend in extraction. Thus, it seems that a basic Hotelling model with unlimited potential reserves and costs that exhibit stock effects fails to explain the historical data on world oil prices and world oil consumption. In order to better reconcile the theory with data, one may therefore need to augment the basic model. Possible modifications include a different estimate of the cost function, technological progress, oligopoly, time-varying demand, and uncertainty; such models will be the subject of future work. Endnotes 1. [email protected] 2. Miller and Upton (1985) use data on U.S. domestic oil- and gas-producing companies to test another reduced-form implication of a Hotelling model. This reduced-form implication, which they term the ‘’Hotelling Valuation Principle’’, is that the value of a unit of reserves in the ground is the same as its current value above the ground less the marginal costs of extracting it.

Figure 35

C.-Y. Cynthia Lin

3. In future work, I hope to develop an analogous test for world oil. 4. Possible extensions of my work would vary either the base year on which some parameters are calibrated or the discount rate, or the terminal date of the simulations. 5. Pindyck also compares the optimal and actual values to the myopic values that would occur if future depletion were ignored but the reserve-production ratio were maintained at its optimal level. 6. I assume that, at any given time t, all the oil extracted at time t is sold on the market at time t. 7. I ignore any common access problems that may arise in perfect competition. In other words, I assume, as does Pindyck (1978), that there is a large number of identical firms that all ignore each other, or, equivalently, that a social planner or a stateowned company has sole production rights and sets a competitive price. 8. This is because P(t) = U ′(E(t)), so that the first-order conditions for the social planner’s problem are the same as those that arise in perfect competition. 9. If then per-period net benefit function G(X,E) is concave in both X and E, then, since the control set {E | E ≥ 0} is convex, the first-order conditions are both necessary and sufficient for an optimum (Weitzman, 2003). 10. I use a U.S. deflator rather than a world deflator because the original nominal time series was in current U.S. dollars. 11. As a possible extension to my paper, I could construct a longer time series by combining my 1965–2001 world oil consumption data with 18601948 world oil production data from Zimmermann (1951, p. 513), or I could use the 1951–2001 world oil production data generously given to me by William Horvath of the Energy Information Administration (EIA). 12. Economists have found the trendless nature of oil prices puzzling. See Lin (2004b) and Lin (2004c) for theoretical expositions of this puzzle and attempts to reconcile the puzzle with theory. 13. Because my data is in terms of barrels rather than mmBtu, I convert these parameters using the conversion factor: 5.8004 mmBtu = 1 barrel (USGS, 2004). 14. I ignore any possible constant term when integrating the marginal cost function to yield the total cost function. 15. These are from the 2SLS results in Lin (2005). 16. I also run simulations in which I instead pin down the initial price P(0) to equal to actual real world oil price at time t. Since the first year of price data in my data set in 1965, for these simulations t = 0 corresponds to 1965, when the real world oil price was

265

$4.51/barrel. However, because pinning down the initial price rather than initial stock often leads to solutions with negative stocks, and because the qualitative features of the results are robust to the type of initial condition chosen, the results of the simulations in which an initial condition was imposed on price are not reported here. 17. This particular base year was chosen because I wanted to use a year after the 1973 Arab oil embargo and becuase 1981 corresponds to the first year of a monthly data set on world oil price and quantity that I have and might later use. For the most part, however, the choice of 1981 for a base year was fairly arbitrary. 18. In general, a monopolist would extract more slowly than would a perfectly competitive industry. One exception is that in the case of isoelastic demand with zero extraction costs, monopoly and perfect competition yield identical solutions for both price and extraction (Stiglitz, 1976). 19. For these simulations, I vary ε from –0.1 to –2.0 since I was unable to generate a solution for the case ε = 0.

References Berndt, E.R., & Wood, D.O. (1975). Technology, prices, and the derived demand for energy. The Review of Economics and Statistics, 57 (3), 259–268. Chakravorty, U., Roumasset, J., & Tse, K. (1997). Endogenous substitution among energy resources and global warming. The Journal of Political Economy, 105 (6), 1201–1234. Deffeyes, K.S. (2001). Hubbert’s peak: the impending world oil shortage. Princeton, NJ: Princeton University Press. Devarajan, S., & Fisher, A.C. (1982). Exploration and scarcity. The Journal of Political Economy, 90 (6), 1279–1290. Edmonson, N. (1975). Real price and the consumption of mineral energy in the United States, 1901–1968. The Journal of Industrial Economics, 23 (3), 161–174. Farrow, S. (1985). Testing the efficiency of extraction from a stock resource. The Journal of Political Economy, 93 (3), 452–487. Halvorsen, R., & Smith, T.R. (1984). On measuring natural resource scarcity. The Journal of Political Economy, 92 (5), 954–964. Halvorsen, R., & Smith, T.R. (1991). A test of the theory of exhaustible resources. The Quarterly Journal of Economics, 106 (1), 123–140. Hanson, D.A. (1980). Increasing extraction costs and resource prices: Some further results. The Bell Journal of Economics, 11 (1), 335–342.

266

Hoel, M. (1978). Resource extraction, uncertainty, and learning. The Bell Journal of Economics, 9 (2), 642–645. Hotelling, H. (1931). The economics of exhaustible resources. The Journal of Political Economy, 39 (2), 137–175. Khalatbari, F. (1977). Market imperfections and the optimum rate of depletion of natural resources. Economica, 44 (176), 409–414. Krautkraemer, J.A. (1998). Nonrenewable resource scarcity. Journal of Economic Literature, 36 (4), 2065–2107. Lasserre, P. (1985). Discovery costs as a measure of rent. The Canadian Journal of Economics, 18 (3), 474–483. Lin, C.-Y.C. (2005). Efficient and consistent estimation of static supply and demand in the world oil market: A dry hole? Mimeo (distributed via SSRN). Harvard University. URL: http://papers.ssrn.com/ sol3/papers.cfm?abstract_id=616961 Lin, C.-Y.C. (2004a). Hotelling revisited: Oil prices and endogenous technological progress. Mimeo. Harvard University. Lin, C.-Y.C. (2004b). Steady-state growth in a Hotelling model of resource extraction. Mimeo. Harvard University. Miller, M.H., & Upton, C.W. (1985). A test of the Hotelling valuation principle. The Journal of Political Economy, 93 (1), 1–25. Nordhaus, W.D. (1980). Oil and economic performance in industrial countries. Brookings Papers on Economic Activity, 1980 (2), 341–399. Pindyck, R.S. (1978). The optimal exploration and production of nonrenewable resources. The Journal of Political Economy, 86 (5), 841–861.

Repsol YPF-Harvard Kennedy School Fellows Research Papers

Pindyck, R.S. (1980). Uncertainty and exhaustible resource markets. The Journal of Political Economy, 88 (6), 1203–1225. Solow, R.M., & Wan, F.Y. (1976). Extraction costs in the theory of exhaustible resources. The Bell Journal of Economics, 7 (2), 359–370. Stiglitz, J.E. (1976). Monopoly and the rate of extraction of exhaustible resources. The American Economic Review, 66 (4), 655–661. Sweeney, J.L. (1977). Economics of depletable resources: Market forces and intertemporal bias. The Review of Economic Studies, 44 (1), 124–141. Tietenberg, T.H. (1996). Environmental and natural resource economics (4th ed.). New York: HarperCollins. U.S. Geological Survey [USGS]. (2004). United States energy and world energy production and consumption statistics. [Online: web]. Cited 7 April 2004. URL: http://energy.cr.usgs.gov/energy/ stats_ctry/Stat1.html Weitzman, M.L. (2003). Income, wealth, and the maximum principle. Cambridge, MA: Harvard University Press. Young, D. (1992). Cost specification and firm behaviour in a Hotelling model of resource extraction. The Canadian Journal of Economics, 25 (1), 41–59. Zimmermann, E.W. (1951). World resources and industries: A functional appraisal of the availability of agricultural and industrial materials. New York: Harper and Brothers, Publishers.

Center for Business and Government John F. Kennedy School of Government Harvard University

Repsol YPF-Harvard Kennedy School Fellows 2003–2004 Research Papers

Repsol YPF-Harvard Kennedy School Fellows 2003–2004 Research Papers William W. Hogan Editor

Cambridge, MA April 2005