Choice of environmental policy in the presence of learning ... - CiteSeerX

Energy Economics 28 (2006) 223 – 242 www.elsevier.com/locate/eneco

Choice of environmental policy in the presence of learning by doing Nic Rivers, Mark Jaccard *,1 School of Resource and Environmental Management, Simon Fraser University, Vancouver, British Columbia, Canada V5A 1S6 Received 20 July 2005; received in revised form 22 December 2005; accepted 2 January 2006 Available online 9 February 2006

Abstract Many clean energy technologies have experienced rapid cost declines recently as a result of accumulating experience with their production and use. It has been argued that if this relationship between experience and cost continues in the future, then it may make sense to foster market penetration of clean energy technologies on a large scale today, despite their current high costs, in order to drive down their costs in the future. If the magnitude of cost declines is sufficient, then rapid large-scale diffusion of clean energy technologies could even provide discounted social benefits that exceed discounted social costs. However, private firms may not make these investments even when socially beneficial since knowledge gained from investments spills over to other firms, and also because private firms have imperfect foresight about cost declines resulting from learning by doing investments. In this case where diffusion of clean energy technologies would provide a social net benefit, but where private firms are unwilling to invest, it may make sense for a government to adopt a regulatory approach aimed at forcing diffusion of clean energy technologies. This paper tests this hypothesis using the criteria of economic efficiency and finds that such a conclusion is unwarranted in most circumstances—instead of forcing diffusion through regulatory instruments, government should seek to correct externalities using market-based instruments. In practice however, government chooses policy based on a broader set of criteria, including especially political acceptability. Our analysis shows that in many cases, regulatory instruments are only slightly more expensive than market-based instruments. For this reason, even though regulatory instruments may be more

* Corresponding author. Tel.: +1 604 291 4219. E-mail address: [email protected] (M. Jaccard). 1 The paper benefited from comments by Chris Bataille, Jillian Mallory, Katherine Muncaster, and Karsten Neuhoff. Remaining errors are the responsibility of the authors. 0140-9883/$ - see front matter D 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.eneco.2006.01.002

224

N. Rivers, M. Jaccard / Energy Economics 28 (2006) 223–242

expensive than comparable market-based instruments, they may still be superior when evaluated using multiple attributes like political acceptability. D 2006 Elsevier B.V. All rights reserved. JEL classification: C60; D62 Keywords: Learning by doing; Renewable energy; Environmental policy; Spillover; Foresight

1. Introduction The generation and use of electricity is associated with environmental externalities that distort the market. For example, absent a cost or regulation for emissions of carbon dioxide, sulfur oxides, or nitrogen oxides, electricity generators will tend to use technologies that produce more emissions than is socially optimal. The deviation in this outcome from the social optimum points to a possible role for policy makers in correcting the externalities that distort the market. Policy makers have at their disposal several different policy tools that can be used to this end. Aside from voluntary, moral suasion, and information provision policies, which have met with relatively little success (Harrison, 1999; Khanna, 2001; OECD, 1999), these policy instruments can be grouped into two general categories: market-based instruments like taxes and subsidies, and command and control instruments like efficiency standards and Best Available Control Technology (BACT) regulations. The economics literature has formed a general consensus suggesting that market-based instruments (as opposed to command and control regulations) are the optimal choice of policy instrument for reaching an environmental goal (Jaffe et al., 2002). With heterogeneous firms (with corresponding heterogeneous pollution abatement cost curves), command and control regulations that impose the same technology requirement on all firms in the industry do not balance marginal costs of abatement amongst firms. As a result, the cost of pollution abatement is higher than for a market-based instrument that sends the same marginal cost signal to all firms. In addition, command and control technology standards have been criticized because they do not provide incentives for firms to go beyond the level of abatement stipulated by the regulation. In contrast, market-based standards continue to provide incentive for pollution abatement up to the point where the marginal cost of abatement equals the level of the financial incentive or disincentive. However, much of the literature comparing alternative polices for pollution abatement has taken a somewhat simplistic perspective on the actual technologies by which pollution is abated—in particular, by assuming that attributes of technologies for pollution abatement are fixed over time. In reality, this is likely not the case. Indeed, there is substantial empirical evidence that suggests that as a firm gains experience with producing a technology, the amount of labor required to produce each unit of that technology decreases (Wright, 1936). Economists have extended this concept to suggest that experience with producing a technology increases not only labor productivity, but productivity of all factor inputs (Arrow, 1962). The result of these factor productivity improvements is that cumulative experience with producing a technology tends to lower its cost of production.2 This has been called learning by doing. 2

We use the term btechnologyQ to refer to both physical goods like wind generators or cars, as well as other products like electricity produced by wind generators. In this latter case, learning would lower the cost of the physical wind turbine as well as the techniques for siting, installing, connecting, and operating the wind turbine, all of which would lower the cost of delivered electricity.


225

Fig. 1. Costs and benefits of investments in clean energy technologies.

Learning by doing has been observed for a variety of different technologies (Argote and Epple, 1990 provides a summary of empirical studies). Recently, energy policy researchers have uncovered empirical evidence of learning by doing for many different types of energy supply technologies (McDonald and Schrattenholzer, 2001 provides a summary of some of this evidence). Of particular importance to the question of policies to address environmental externalities in the energy system could be the presence of learning by doing for clean electricity generation technologies, like wind or solar power, or thermal generation with carbon sequestration. Empirical evidence suggests that costs for some of these technologies have fallen dramatically in past decades, and continue to fall as a result of increasing cumulative experience. Because of the potential for cost reductions, there could be societal benefits from a concerted investment today in clean energy technologies. Fig. 1 provides an illustration of this benefit.3 The clean energy technology is currently much more expensive than the conventional technology. When externalities (e.g., CO2) are accounted for, the cost of the conventional technology increases, but the relative immaturity of the clean energy technology means that it is still more expensive. However, if an effort is made to deploy the clean energy technology in the market despite its initial high cost, the effect of learning may drop the cost of the clean energy technology in the future to below the cost of the conventional technology when externality damages are included in its production cost. At this point, society starts to receive a benefit from its investment (area B). If the total discounted benefit from this strategy exceeds the total additional costs of early deployment of the clean energy technology (area A), then this early deployment strategy provides net benefits to society and should be adopted. Neuhoff (2004), Margolis (2003), Neij (1997), Wene (1999) and others have advocated for this type of bstrategic deploymentQ or btechnology buy-downQ strategy. This article explores the policy instruments required to support such an initiative. Several characteristics of the process of learning may affect the choice of policy for pollution abatement in the energy system. First, cost reductions from learning by doing accrue as a result of accumulated experience. For example, the cost of wind power has fallen dramatically in recent years because manufacturers have learned how to build larger wind turbines and capture economies of scale, to use less and lighter materials, and to design rotor blades with improved 3 We have depicted the cost of the conventional technology as static. In reality, costs for mature technologies continue to decline as cumulative production increases. However, this decline is slower than for emerging technologies, so the general results of the figure would remain the same even if these declines were included.

226


aerodynamics for greater efficiency. Operators of wind turbines have simultaneously learned how to improve wind turbine availability, have developed high-resolution wind maps to aid in locating optimal sites, and have improved site engineering practices. The result of these many incremental improvements has been a dramatic fall in the cost of wind-produced electricity over time (Neij, 1997). Each of these improvements is in many ways a non-excludable good (Romer, 1990). When, for example, one wind turbine manufacturer developed improved turbine blade pitch controls, it was not able (or only partially able) to prevent other manufacturers from appropriating this knowledge and producing similar innovations of their own. Knowledge bspills overQ from one firm to another (Arrow, 1962). Because of this positive externality, there may be less investment by private firms in technology development than is optimal from a social perspective. Second, investment by firms in the development of improved technology and processes is different from other types of investments. While the return from any investment is to some degree uncertain, the variance in returns on blearningQ investments is much larger than for other investments, and much of the value is associated with very low probability but very high value outcomes (Scherer et al., 2000). Firms may therefore choose to invest less in learning than is optimal from a social perspective, in order to hedge against losses. This underinvestment in learning is a result of imperfect foresight (due to uncertainty) about the cost reductions that follow a learning investment. These characteristics of learning by doing suggest that investments in learning may be lower than socially optimal if left to the private sector. Even if bstrategic deploymentQ investments provide net benefits to society, as illustrated in Fig. 1, private firms may not anticipate or be able to capture that benefit, and choose an investment path that is socially sub-optimal. This fact could have an important consequence for selection of policy instruments to encourage clean energy supply. A tax or other market-based instrument may correct for the environmental externality, but not provide sufficient incentive for widespread adoption of clean energy technologies by private firms. A regulation, on the other hand, can explicitly require investment in clean energy technologies, and may help society to capture benefits from learning in clean energy technologies. However, a regulatory approach still requires identical technology choices by all firms being regulated, which can impose high costs. The aim of this paper is to identify an optimal policy approach for the energy sector when learning by doing occurs. Previous studies have attempted to determine the optimal policy instrument for stimulating investment in a particular technology. These studies have generally assumed that the attributes of the various technologies that can be chosen to make up new stock are frozen over time. Jung et al. (1996) use comparative statics to show that market-based instruments provide greater industry-level incentive for new technology adoption than command and control performance standards. These findings echo those of Milliman and Prince (1989), who find that firm-level incentives for adoption of a new technology are greater for market-based instruments than for command and control regulations.4 There has, however, been little attention paid to the choice of policy instrument when the attributes (in particular, the costs) of new technologies are dynamic, despite the focus in the 4 All of these studies also compare different classes of market-based instruments, and find that in most cases auctioned emission permits offer the most incentive for technology adoption, followed by taxes and subsidies, and finally freely allocated emission permits. Other studies (e.g., Keohane, 1999) find that if the auctioned permit price is endogenously determined, then the adoption of a new technology lowers this permit price and results in lower emissions abatement than under an emission tax.


227

literature on models incorporating endogenous technical change (e.g., Loeschel, 2002; Jaffe et al., 2000; Manne and Richels, 2004).5 With substantial evidence that the cost of a technology is a function of accumulated experience with producing the technology, a re-examination of the optimal ranking of policy instruments is required. Ficher and Newell (2004) conduct this type of analysis, comparing six different market-based policy instruments for supporting renewable energy while the cost of renewable energy is endogenous. They find that the assumption of endogenous costs does not significantly affect the ranking of optimal policy instruments, with an emissions tax being the most effective instrument for inducing emissions abatement. However, the Fischer and Newell study ignored the effect of knowledge spillover, and assumed that firms had perfect information about future costs of technologies. This study compares a simple command and control standard that forces identical choices of technology for all firms in the model to a generic market-based instrument that allows flexibility of technology choices. It includes knowledge spillover between firms and models firms as profitmaximizing agents with imperfect foresight. In Section 2 we develop a series of models designed to compare policies for pollution abatement. In Section 3 we use these models to numerically compare market-based and command and control instruments for emissions abatement for the case of electricity generators. Finally, in Section 4 we conclude with a discussion of policy implications of our models. 2. Model In this section, we start with a simple theoretical model, and progressively refine it by relaxing restrictive assumptions so that subsequent versions of the model more closely approximate the real world. Each model is used to generate insights relating to the choice of environmental policy when technological learning is induced by policy. Throughout the Results section that follows, we use the models to compare the relative cost effectiveness (in cost per unit of pollution reduced) of a generic market-based approach to emissions abatement (tax, tradable emissions permits) versus a command and control approach focused on the requirement for certain forms of technology. Each policy aims to achieve the same amount of emissions reduction from throughout the economy. Our model has three periods—past (period 0), present (period 1), and future (period 2).6 For simplicity, we assume that no discounting takes place within the periods, but that costs are discounted from period 2 to period 1 by a discount rate d. Firms in the model are electricity generators and have three technologies available for generating electricity, as illustrated in Fig. 2: ! A baseline technology, which is relatively cheap, but which emits a high amount of pollution. Standard coal-combustion electricity generators are an example of the baseline technology. Throughout the paper, we refer to the baseline technology with a subscript b. ! An alternative technology, which is somewhat more expensive, but emits less pollution than the baseline technology. Natural gas fired generation is an example of the alternative technology. Throughout the paper, we refer to the alternative technology with a subscript f. 5

Most exploration of endogenous technical change has taken place in integrated energy-economy models, which generally lack the resolution required to understand important interactions within and between firms that influence the optimal choice of policy for stimulating investment in a technology. 6 The past (period 0) is only relevant in setting the starting stock of knowledge for each technology.

228


Fig. 2. Attributes of technologies.

! A clean energy technology, which is much more expensive, but emits much less pollution than the baseline technology. Wind or solar generation or thermal generation with sequestration of carbon dioxide emissions are examples of clean energy technologies. Throughout the paper, we refer to the clean energy technology with a subscript r. Both the alternative and the clean energy technologies experience declining costs through learning by doing as per Eq. (1) (Wright, 1936; Arrow, 1962):7 Nj;t bj cj;tV ¼ cj;t¼0 V ð1Þ Nj;t¼0 where c jV,t is the cost of producing energy from the first unit of technology j installed in period t, N j,t is the total cumulative production energy from technology j up to but not including period t, and b j is a parameter that defines how fast the cost of producing energy from technology j falls as cumulative production increases, based on Eq. (2): PRj ¼ 2bj

ð2Þ

where PRj is the progress ratio, defined as the percentage cost reductions obtained through a doubling of production of technology j.8 A PR of 80%, for example, means that as cumulative production doubles, the cost to produce one unit falls to 80% of its initial value. Many studies have attempted to empirically measure the progress ratio of both environmental technologies and other technologies in general. Although there is significant variation in the results, progress ratios between 75% and 95% are generally found, with both Dutton and Thomas (1984) and McDonald and Schrattenholzer (2001) reporting a median PR of about 80– 84% for the technologies in the studies they surveyed. Some studies have found that a lower PR 7

In this paper, we assume that the baseline technology is mature and experiences no cost declines to simplify our paper. Relaxing this assumption would not change our conclusions (provided the learning for the baseline technology is slower than for the two alternative technologies) but would add complexity to the paper. 8 Some literature defines the dlearning rateT (LR) as LR = 1 PR. Although this formulation is more intuitive (higher LR means faster learning, as opposed to lower PR means faster learning), we use the PR in this paper, because most literature follows this convention.


229

(indicating faster cost reductions) is associated with relatively immature technologies, while a higher PR is associated with relatively mature technologies (Boston Consulting Group, 1968). In our model, the initial production of the alternative technology (N f,t=0) is much higher than that of the clean energy technology (N r,t=0), so absolute costs fall faster with each incremental unit of production for the clean energy technology. In the absence of policy, firms would choose the baseline technology, because it is the cheapest option. The marginal cost of pollution abatement, a, measured in dollars per tonne of pollution abated, can be calculated for the clean energy and alternative technologies (compared to the base technology) as in Eq. (3): aj;t ¼

cj;t cb eb ej

ð3Þ

where c j,t is the marginal cost of technology j in period t in $/kWh and e j is the emission rate of technology j in t/kWh. We assume that both c f and c r are increasing and concave with quantity produced in a period, to reflect the fact that the best sites for each technology are the first ones exploited. The supply curve for technology j in period t is quadratic as in Eq. (4): 2 cj;t ¼ cj;tV þ kj Nj;t

ð4Þ

where c jV,t is the cost in $/kWh for the first unit of energy produced from technology j in period t, k j is a constant that reflects the steepness of the supply curve for technology j (i.e., how fast perunit costs increase in a period), and N j,t is the amount of energy produced by technology j in period t, measured in kWh. By substituting Eq. (4) into Eq. (3), we obtain the marginal emissions abatement cost curve for each technology in each time period t, which are shown representatively in Fig. 3. Our formulation for the marginal cost of pollution abatement shows that the cost of generating energy from a technology falls with increasing cumulative production over time as experience is gained with the technology, but increases with installed capacity within a period, as the best sites for the technology are exploited. Each period, the firms need to purchase some amount of new technology to make up for technology stock turnover as well as to meet incremental demand increases. Our model focuses

Fig. 3. Marginal emissions abatement cost curves.

230


on the firms’ decision of which new technologies to purchase, and the appropriate policy for affecting this decision. In period 1, government begins regulating the amount of pollution emitted by firms using either a market-based approach or a command and control approach. The market-based approach consists of a financial signal to stimulate pollution abatement (tax, emissions cap and tradable permits system, subsidy) that has a level T t in period t, in dollars per tonne of pollution abated. This financial signal encourages firms to switch from the baseline technology to the alternative or clean energy technologies in periods 1 and 2. The actual amount of emissions abatement from technology j in period t is determined by the marginal abatement cost curves for each technology in each period, and is denoted Q j,t . Thus, the total emissions abatement in period t due to the market-based signal is Q t = Q f,t + Q r,t . This is depicted in Fig. 3. We calculate these levels, and the resulting social cost of the market-based instrument, in equations shown below. Alternatively, the government may opt to use a command and control instrument. In our model, we model the command and control instrument as requiring firms to invest in the clean energy technology, rather than allowing firms the freedom to choose between the alternative and clean energy technologies, as under the market-based approach. The regulation can therefore be thought of as a type of Best Available Control Technology (BACT) policy, or else a variation on a Renewable Portfolio Standard that requires each firm to invest in renewable energy generation. We require that the total amount of emissions abatement from the command and control regulation be the same in each period as under the market-based regulation. We then calculate the social cost of emissions abatement using the command and control instrument, using equations shown below, and compare to the social cost of the market-based instrument. 2.1. Model 1—one firm with perfect foresight Our simple model begins by assuming that all electricity generation is provided by one firm. Because there is only one firm in the market, there is no knowledge spillover to other firms. In addition, we assume that the firm has perfect foresight regarding the future costs of the technologies. In this situation, given the flexibility to do so, the firm will try to find the optimal production of each technology over both time periods so that total discounted costs are minimized. This might involve producing more of a technology than would be expected in the first period (from a static efficiency standpoint) to reduce its cost in the second period. If the regulator uses a market-based instrument, the present value cost of the policy for the firm will be: C¼

Z

Z Qr;t¼1 af ;t¼1 dQf ;t¼1 þ ar;t¼1 dQr;t¼1 þ Tt¼1 QT ;t¼1 Qf ;t¼1 Qr;t¼1 0 0 Z Qf ;t¼2 Z Qr;t¼2 1 þ af ;t¼2 dQf ;t¼2 þ ar;t¼2 dQr;t¼2 1þd 0 0 þ Tt¼2 QT;t¼2 Qf ;t¼2 Qr;t¼2 Qf ;t¼1

ð5Þ

where T t is the level of the tax in period t, Q j,t is the amount of emissions abatement achieved using technology j in period t, and Q T,t is the total business as usual emissions of the firm in


231

period t. The amount of emissions abatement using technology j is related to the total production of technology j as per Eq. (6): Qj;t ¼ Nj;t eb ej ð6Þ Substituting Eqs. (1), (3), (4), and (6) into Eq. (5) and taking partial derivatives of the total cost to the firm with respect to the amount of emissions abatement by each technology in each time period, BC / BQ f,t=1, BC / BQ r,t=1, BC / BQ f,t=2, BC / BQ r,t=2, gives four equations with four unknowns, Q f,t=1, Q r,t=1, Q f,t=2, and Q r,t=2. Setting partial derivatives equal to 0 gives the level of abatement with each technology that minimizes present value costs to the firm. The equations are included in the Appendix. We solve for these unknowns in the Results section using a numerical Newton–Raphson technique because an algebraic solution is not possible. If the regulator applies a command and control regulation to force the firm to achieve an equivalent amount of emissions abatement in each period using only the clean energy technology, then the total discounted cost to the firm will be: Z Qf ;t¼2 þQr;t¼2 Z Qf ;t¼1 þQr;t¼1 1 C¼ ar;t¼1 dQr;t¼1 þ ar;t¼2 dQr;t¼2 : ð7Þ 1þd 0 0 To determine the optimal regulatory strategy from a social perspective, we compare the social cost of the regulatory policy to the social cost of the market-based policy. These costs are the same as the private costs, but exclude any tax payments incurred under the market-based policy. We present this social cost comparison as the difference in costs between the regulatory policy and the market-based policy per tonne of pollution abated, as in Eq. (8): Cost savings from tax Z Qf ;t¼2 þQr;t¼2 0 Z Qf ;t¼1 þQr;t¼1 1 ar;t¼1 dQr;t¼1 ar;t¼2 dQr;t¼2 C B 0 1 0 C ¼B þ @ A 1þd Qf ;t¼1 þ Qr;t¼1 Qf ;t¼2 þ Qr;t¼2 20 Z 6B B 6 4@

af ;t¼1 dQf ;t¼1 þ 0

Qr;t¼1

0

Qf ;t¼1 þ Qr;t¼1

0Z B B @

Z

Qf ;t¼1

0

Qf ;t¼2

af ;t¼2 dQf ;t¼2 þ

Z

Qr;t¼2

0

Qf ;t¼2 þ Qr;t¼2

1 ar;t¼1 dQr;t¼1 C 1 Cþ A 1þd

13 ar;t¼2 dQr;t¼2 C7 C7 : A5

ð8Þ

Eq. (8) shows the common result that the market-based policy is always at least as cost effective as the command and control policy from a social perspective, and in most cases is more cost effective than the command and control policy, when there is one firm with perfect foresight. It has been shown empirically and theoretically by other researchers that imposing strict technology standards on firms can impose large private and social costs relative to marketbased approaches (Jaffe et al., 2002 provide an overview of such literature). 2.2. Model 2—one firm with imperfect foresight In reality, the firm does not have perfect foresight about the future costs of a technology. It cannot therefore predict with certainty the costs of either the alternative or the clean energy

232


technology in period 2 based on its investment in period 1. At one extreme, the firm assumes no cost reductions due to experience. At the other, the firm overestimates the cost reductions due to experience by some factor. The firm’s choice of which technology to invest in period 1 is now defined by its expectation of the cost of each technology in period 2, E(c jV,t=2). E(c jV,t=2) is based on the parameter q j , which represents the firm’s anticipation of cost reductions for technology j. If q j = 1, the firm anticipates the actual cost reductions perfectly. If q j = 0, the firm assumes no future cost reductions into its period 1 decisions. If q j = 2, the firm overestimates the cost reductions available through learning by a factor of two. Eq. (9) shows the firm’s expectation of cost reductions for the clean energy and alternative technologies. ! Nj;t¼1 bj E cj ;t¼2 V ¼ cj V;t¼1 qj cj;t¼1 V cj;t¼1 V : ð9Þ Nj;t¼0 We now substitute Eq. (9) into Eq. (5) to make a sequential decision model, rather than a simple optimization model. In period 1, the firm chooses the level of emissions abatement from each technology that it expects will lead to the cheapest net present cost of complying with emissions abatement obligations. In period 2, the firm reevaluates and chooses the optimal emissions reduction allocation with its new (now perfect) knowledge of the cost of each technology in period 2. Again, setting partial derivates equal to 0 allows us to calculate the investment path taken by a cost-minimizing firm (this time with imperfect foresight). The result of incorporating this imperfect foresight into the model is that in some cases a market-based policy is dominant, while in other cases a command and control policy is dominant. In particular, if the firm has poor foresight about potential cost reductions from the clean energy technology, a command and control policy that forces the firm to invest in the clean energy technology can be optimal. To understand this, consider the situation where investment in the clean energy technology in period 1 by the firm would stimulate cost reductions through learning by doing that would lower the cost of the clean energy technology below that of the alternative technology in period 2 enough that the net present cost of emissions abatement using the clean energy technology is lower over both periods than using the alternative technology. If the firm has perfect foresight regarding the future costs of technologies, it would invest in the clean energy technology in period 1 to stimulate cost reductions through learning by doing— thereby choosing the path that minimizes net present costs. If the firm has poor (or no) foresight regarding cost reductions through learning, and is subject to a market-based instrument, it would invest in the alternative technology in period 1, and not capture the benefits of the fast learning available for the clean energy technology. If a command and control regulation that forces the firm to invest in the clean energy technology is used, on the other hand, the cost reductions through learning by doing with the clean energy technology are captured, and net present costs may be decreased. So depending on the characteristics of technologies, a command and control regulation to force investment in the clean energy technology can have a lower social cost than a market-based instrument when the firm has imperfect foresight about future costs of technologies. Otherwise, the same results apply as in the previous section. Note that for the government to choose the effective policy in this situation (market-based vs. command and control policy) requires that the government have better information about the future costs of technologies than the firm. This may be realistic in some cases, but the government may also have worse information than the firm. For this reason, this result alone does not motivate the use of regulations over taxes in regulating environmental outcomes, unless


233

new evidence arises that governments are more aware than firms of cost savings from emerging environmental technologies. 2.3. Model 3—many identical firms with imperfect foresight In this section, we relax the single-firm assumption, and introduce many identical firms into the economy. This is more realistic except in those jurisdictions where electricity monopolies still control all generation investment decisions. Each firm has imperfect foresight, and chooses technologies in period 1 in order to minimize the expected net present cost for both periods. In period 2, each firm is then able to choose again the technology that is optimal in that period, with its now perfect knowledge about the costs of technologies. So far, this model refinement will produce exactly the same results as the model in the previous section (i.e., many small firms with exactly the same behaviour patterns and combined characteristics as one big firm will all act identically, so that their combined actions are the same as the one big firm). However, there is another variable to introduce when many firms are present. Several researchers have noted that investments in learning will be suboptimal in the case where more than one firm is present (Arrow, 1962; Griliches, 1979; Jaffe, 1986). Knowledge and information (the products of research and development, as well as of production experience) are never fully appropriable to a firm. In other words, if firm A makes a learning investment, firm B will learn something from firm A’s investment (through, for example, poaching of employees or reverse engineering of products, or through more benign avenues like the gradual accumulation of knowledge in the industry). This is a case of positive externality, and implies that private firms in a competitive market will provide a lower investment in learning than is appropriate from a social perspective. As a consequence, although the monopoly firm in model 2 (which is able to appropriate all benefits of its experience) would invest in the clean energy technology in certain circumstances (where that investment provides greater expected net present benefits than the alternative), the many firms in a competitive market might not in the same circumstance, because they cannot each appropriate the full benefits of their learning. In this case, a more socially desirable outcome may be obtained through mandating the use of the clean energy technology for all firms than through using a market-based instrument like a tax or a tradable permit system that allows firms more flexibility in choosing privately optimal technologies. We represent knowledge spillover in the economy by assuming that cost reductions for the alternative and clean energy technologies accrue in two ways. First, the total experience of all firms in the economy lowers the cost of a technology at some rate, and second, the experience of any particular firm d with a technology lowers the cost of the technology for that firm. The parameter a shows how much of the cost reductions due to cumulative experience spill over from a firm making a learning investment to other firms in the economy. If a = 1, all of the learning is global, and if a = 0, all of the learning is firm-specific. Some empirical evidence exists that about 30–50% of accumulated knowledge disseminates to the wider economy as spillover (Irwin and Klenow, 1994). The cost of technology j to firm d in period 2 is then: Nj;t¼1 bj a nj;t¼1;d bj ð1aÞ cj;t¼2;d V ¼ cj;t¼1 V Nj;t¼0 nj;t¼0;d

ð10Þ

234


where n j,t,d is the cumulative production of technology j in period t by firm d. The global P investment in technology j in period t is N j,t = d n j,t,d . In period 1, each firm has an expected cost for technology j in period 2, determined by combining Eq. (10) with Eq. (9): ! Nj;t¼1 bj a nj;t¼1;d bj ð1aÞ V ¼ cj;t¼1 V qj cj;t¼1 V cj;t¼1 V E cj;t¼2;d : ð11Þ Nj;t¼0 nj;t¼0;d At this point there is no way for a firm d to know the value of N j,t=1 in period 1. In our model we assume that firms make decisions sequentially in each period. Firm 1 invests first in period 1, and makes no assumptions about what subsequent firms may invest in, so calculates E(c j,t=2,d V ) based only on its own investment. In other words, it assumes that N j,t=1 = N j,t=0 + n j,t=1,1. Firm 2 has seen what firm 1 does in period 1, and factors this information into its decision. In other words, it assumes that N j,t=1 = N j,t=0 + n j,t=1,1 n j,t=0,1 +n j,t=1,2 n j,t=1,2. When it comes time for the last firm to invest, this firm has observed technology choices by other firms in period 1, so knows the value of N j,t=1 with certainty. This type of sequential decision model has been used in the literature to show how technologies can become dlocked inT by decisions made by the first decision makers (e.g., Arthur, 1989; Banerjee, 1992). For example, when firm 1 makes its decision, it decreases the cost of the technology it invests in (because of accumulated global experience), which makes that technology more attractive to firm 2, and so on down the line. In other words, the decision of firm 1 influences decisions made by all subsequent firms. We also test the results assuming that firms’ decisions are made with no expectations of decisions of other firms, and with complete knowledge of decisions of other firms. Changing this assumption does not significantly change the model results. 2.4. Model 4—many heterogeneous firms with imperfect foresight In reality, firms in the economy are not identical. In this final refinement of the model, we introduce heterogeneity amongst the firms by allowing n j,t=0,d , the initial investment in technology j in period 0, to be different for each firm d. This not only affects the rate of the firmspecific learning for each technology, but also affects the size of each firm. In addition, we allow heterogeneity amongst the initial costs of technologies for each firm. The heterogeneity amongst the firms in this model provides additional reason for using the market-based instrument instead of the command and control regulation. Because of the heterogeneity amongst the firms, abatement costs are different for the two competing technologies. The market-based instrument allows firms to optimally allocate emissions reductions between these technologies, and allows emissions reductions to be optimally allocated between firms, while the command and control standard requires that all firms use the clean energy technology, and undertake an amount of emissions reductions proportional to their emissions, rather than commensurate with their abatement cost. To mathematically generate the required heterogeneity in the model, we assume that n j,0,d are gamma distributed throughout the firms for each technology, and assume that c j,t=1,d V are normally distributed.9 We then find the savings from using a market-based instrument instead of a command and control instrument in a similar way as for model 3. 9

A gamma distribution prevents negative values and is skewed to the right, and is therefore appropriate for the distribution of firm size.


235

3. Results In this section, we present quantitative results based on the four iterations of the model developed in the previous section. In each case, we base the evaluation of policy instruments on a comparison of the social cost of emissions abatement from the market-based and command and control policy instruments (Eq. (8)). In order to make this quantitative comparison, we require numerical values for the parameters in the model. We choose parameters by assuming that firms are electricity generators, that the baseline technology is coal-fired generation, that the alternative technology is natural gas-fired CCGT generation, and that the clean energy technology is wind turbine generation. Table 1 shows the values used for the simulations that follow, which are generally representative of a developed-country economy. We simulate a $100/t CO2 tax in periods 1 and 2. This section proceeds by outlining four main lessons for policy based on the models developed. Lesson 1: When regulating emissions from a monopoly firm with perfect foresight, marketbased instruments are always more cost effective than command and control instruments. Fig. 4 shows the relative cost effectiveness of the market-based instrument compared to the regulatory instrument when only one firm with perfect foresight exists, and with the parameters in Table 1 (the result of Eq. (8)). The figure shows that over a reasonable range of progress ratios and starting costs for the clean energy technology, the market-based instrument reduces carbon dioxide emissions more cost effectively than the command and control regulation. In no situation is the command and control regulation superior to the market-based instrument. Cost savings from using the market-based instrument instead of the command and control instrument are in the range of 14–18 $/t CO2, for the numerical assumptions in Table 1. On a relative basis, the regulatory instrument is 15–39% more expensive than the market-based instrument over the range of parameters tested. This result is not novel. Even in the presence of endogenous technological learning, monopoly firms with perfect foresight are able to dynamically optimize to choose the technology allocation that minimizes total discounted costs over both periods. We present it only in order that results from subsequent models may be compared to those from this model. Table 1 Numerical model parameters Parameter

Value

Units

Description

eb ea er cb c aV c Vr PRa PRr ka kr N f,0 N r,0 a d –

0.0003203 0.0001811 0 0.03 0.04 0.06 96% 80% 1E 24 5E 24 20 0.5 0.3 0.03 10

t/kWh t/kWh t/kWh $/kWh $/kWh $/kWh – – $/kWh3 $/kWh3 TWh TWh – – –

Emissions of carbon dioxide from coal-fired generation Emissions of carbon dioxide from gas-fired generation Emissions of carbon dioxide from wind generation Cost of coal-fired generation Cost of first unit of gas-fired generation in period 1 Cost of first unit of wind generation in period 1 Progress ratio of natural gas-fired generation Progress ratio of wind generation Cost curve steepness for natural gas-fired generation Cost curve steepness for wind generation Production of gas-fired generation in period 0 Production of wind generation in period 0 Spillover rate of learning (models 3 and 4 only) Discount rate Number of firms in the economy (models 3 and 4 only)

236


Fig. 4. Comparison of market-based and command and control instruments for the case with one monopoly firm with perfect foresight.

Lesson 2: When regulating emissions from a monopoly firm with imperfect foresight, regulatory instruments that stipulate the use of a clean energy technology are generally less cost effective than market-based instruments. Fig. 5 shows the relative cost effectiveness of the market-based instrument compared to a regulatory instrument when the market consists of one firm with imperfect foresight. Results are shown for a range of assumptions about the firm’s foresight about future technology costs (q r from 0 to 2 on the horizontal axis of the figure). When q r = 0, the monopoly firm in this model bases period 1 decisions on the assumption that costs for technologies are static, even though the model includes a representation of endogenous costs. When q r = 1, the firm perfectly predicts the

Fig. 5. Comparison of market-based and command and control instruments for the case of one monopoly firm with imperfect foresight.


237

amount of cost reductions available from its investment in the clean energy technology (and to a lesser extent the alternative technology), so the results are exactly the same as in the previous section. When q r = 2, the firm overestimates the amount of learning-induced cost reductions that will occur, and consequently over-invests in the clean energy technology in period 1. The figure shows that in some situations, a regulatory instrument that stipulates the use of the clean energy technology by the firm can be more cost effective than a market-based instrument that allows the firm the flexibility to choose between the two available technologies. This result is only observed when the firm underestimates the amount of learning available from the clean energy technology, and when the rate of learning for the clean energy technology is high. For the parameters tested, this model showed that a market-based instrument can offer savings of up to $22/t CO2 when learning occurs quickly and when the firm is optimistic about cost reductions, but that the regulatory instrument can save up to $3/t CO2 compared to the market-based instrument when the firm does not factor learning by doing into its investment decisions. On a relative basis, the market-based instrument is between 4% worse and 29% better than the regulatory instrument over the range of parameters. For the government to choose the optimal policy in these situations requires that the government have better information on the future costs of technologies than the firm. Although this situation is possible, we have not found evidence to suggest that it is likely, and it is equally possible that the government has worse information on future technology costs than the firm. As a result, although this model shows that social welfare benefits from command and control regulations are possible, it is not enough on its own to justify the use of command and control regulations by the government. Nonetheless, it is important to recognize that multi-period doptimizingT decisions by firms are not based on true future costs of technologies, but on expected future costs, and that these expectations can be erroneous. Lesson 3: When regulating emissions from an economy with many identical firms, regulatory instruments that stipulate the use of a clean energy technology are generally less cost effective than market-based instruments. Fig. 6 shows the relative cost effectiveness of the market-based instrument compared to the regulatory instrument for the case where there are 10 firms in the economy with imperfect foresight. On the horizontal axis, the figure shows different values for the a parameter, which

Fig. 6. Comparison of market-based and command and control instruments for the case of many identical firms.

238


reflects knowledge spillover in the economy. The three lines in the graph denote different values for the q r parameter—the degree of foresight by firms for cost reductions in the clean energy technology. One of the hypotheses that this paper attempts to test is that because of knowledge spillovers in the economy, regulatory instruments promoting renewable energy technologies could be more cost effective than market-based instruments. Fig. 6 shows that this is generally not the case—in almost all cases, a market-based instrument is preferable to a command and control regulation. For the range of parameters tested, at one end of the range the market-based instrument saved up to $17/t CO2 compared to the regulatory instrument, while at the other end of the range the regulatory instrument saved up to $4/t CO2 compared to the market-based instrument. In relative terms, the market-based instrument is between 5% worse to 22% more cost effective than the regulatory instrument over the range of parameters tested. Lesson 4: When regulating emissions from an economy with many heterogeneous firms, market-based instruments are almost always more cost effective than regulatory instruments. Model 4 differs from model 3 in that firms are heterogeneous, which implies that abatement cost curves for technologies are heterogeneous, meaning that the market-based instrument should offer increased flexibility over the command and control instrument, which specifies levels of investment by each firm in the clean energy technology. Qualitatively, the results are the same as lesson 3—even though learning spillover in the economy and imperfect foresight by firms are explicitly modeled, a market-based instrument is almost always more cost effective than a regulatory instrument. In addition, the presence of heterogeneity amongst firms reduces the degree of possible benefits to be gained from command and control regulation, and increases its average cost in relation to the market-based instrument. As firms become increasingly heterogeneous, the cost savings from the use of the market-based instrument become more pronounced. We do not present quantitative results here, and instead present results from this analysis combined with an uncertainty analysis in the following section. 3.1. Uncertainty analysis There are many studies that attempt to empirically estimate progress ratios for technologies based on historical prices and installations. Even for papers looking at the same technology, the estimated progress ratios can differ widely. Even with a consistent data set, Papineau (2005) shows that changing analysis method can change the resulting estimated progress ratio significantly, with her estimates of the progress ratio for solar PV ranging from 0.77 to 0.95. Because of the uncertainty in this critical parameter, many researchers advocate a stochastic representation of learning rates in energy-economy models (e.g., McDonald and Schrattenholzer, 2001; Dutton and Thomas, 1984; Papineau, 2005). In this section, we introduce stochasticity into model 4 for the key parameters, and test the robustness of our results to this incorporation of uncertainty. Table 2 shows the distributions assumed for the stochastic parameters in the model. Table 2 shows assumed distributions for both dGlobalT and dLocalT parameters. Sampling Local parameters represents the heterogeneity between firms discussed in the previous section, while sampling Global parameters accounts for uncertainty in key parameters. With uncertain parameters defined as in Table 2, we use a Monte Carlo sampling process with 1000 runs to generate a probability distribution in the output. Fig. 7 shows the results of these simulations presented as a cumulative distribution function that shows the cost savings from using a market-based instrument instead of a command and control instrument. We present


239

Table 2 Distribution of uncertain parameters in stochastic model Parameter

Distribution

Mean

Std. dev. (LV)a

Std. dev. (HV)a

Global/localb

c b,0,d V c a,0,d V V c r,0,d PRf PRr n b, 0,d n a,0,d n r,0,d a kf kr qf qr

Normal Normal Normal Normal Normal Gamma Gamma Gamma Betac Gamma Gamma – Betac

0.03 $/kWh 0.04 $/kWh 0.06 $/kWh 96% 80% 200 TWh 20 TWh 2 TWh 0.5 1E 24 5E 24 1 0.6

0.001 0.001 0.001 0.75% 5% 20 TWh 2 TWh 0.2 TWh 0.3 1E 25 5E 25 – 0.1

0.002 0.002 0.002 1.5% 5% 40 TWh 4 TWh 2 TWh 0.3 2E 25 1E 24 – 0.1

Local Local Local Global Global Local Local Local Global Local Local Global Global

a There are two values presented for the standard deviation of parameters labeled LV for dLow VarianceT and HH for dHigh VarianceT. These correspond to the scenarios in Fig. 7. b Global parameters are ones for which the distribution is sampled from once for each run, and applied on a global basis. For local parameters, the distribution is sampled from for each firm in each run, so that the parameter is allowed to have different values for different firms. c The distribution for the global spillover rate and the foresight parameter is a beta distribution with a maximum value of 1 and a minimum value of 0.

results for two cases: a dLow VarianceT case where parameters are more certain and firms are more similar, and a dHigh VarianceT case where firms are assumed to be more heterogeneous and parameters are less certain. Average cost savings for the market-based instrument are higher for the High Variance case, because this scenario takes greater advantage of the flexibility of the market-based instrument. The regulatory instrument was inferior to the market-based instrument in each of the runs for both the high- and low-variance cases by an average of $17–20/t. 4. Discussion and conclusions Several researchers have noted that because of the potential for cost reductions through learning for clean energy technologies, it may make sense for society to begin a process of

Fig. 7. Stochastic comparison of market-based and command and control instruments.

240


bstrategic deploymentQ of clean energy technologies today in order to lower their costs in the future, even though their costs today may be high relative to conventional energy technologies (even when externalities are fully accounted for). Some evidence suggests that the discounted benefits of such an approach outweigh the costs. However, if left to the private sector, it is unclear whether such an investment will occur, even after externalities from pollution have been corrected for, because of the high initial cost of clean energy technologies. This is due to several characteristics of learning investments—notably that firms do not have perfect foresight about the effects of their learning investments, and because knowledge gained during a firm’s learning process spills over to other firms. If strategic deployment of clean energy technologies is socially desirable, but private markets do not face adequate incentives to pursue that strategy, it may make sense to use a regulatory approach to encourage the penetration of clean energy technologies, instead of a market-based approach. This article tested this hypothesis using a numerical model. The results of the testing showed that a regulatory approach is almost always inferior to a market-based approach to emissions abatement in terms of economic efficiency, even when knowledge spillover and imperfect foresight are accounted for. When developing energy policy, economic theory therefore suggests that governments should focus on correcting for environmental externalities using a market-based instrument, but should not focus on additional bstrategic deploymentQ of clean energy technologies using a regulatory instrument. In practice however, governments choose policies using a wider set of criteria than simple economic efficiency, including effectiveness, political acceptability, and administrative simplicity. Using these broader criteria often pushes governments towards sector-specific and regulatory policies. Our analysis, which includes the combined effects of knowledge spillover and imperfect foresight, has shown that the cost of these theoretically suboptimal policies may only be 10–20% greater than more efficient market-based policies. This may be a small price to pay for governments interested in satisfying a broader set of criteria than simply economic efficiency. Our analysis focused on comparing a strict market-based approach for promoting clean energy technologies with a strict regulatory approach, and found that a market-based approach is almost always superior. In practice however, new policies like the Renewable Portfolio Standard (RPS) have been developed that combine aspects of both policy approaches. These new policies may be superior to the strict market-based and regulatory policies examined in this paper in simultaneously addressing the two externalities examined here—a negative externality from pollution and a positive externality from knowledge spillover. This question is left as a possible avenue for future research. Appendix A For model 1, the partial derivative equations are: bf Q cfV;t¼1 Qff ;t¼1 bf Qf ;t¼2 kf Q2f;t¼1 BC cfV;t¼1 cb ;t¼0 ¼0 ¼ þ Tt¼1 þ BQf ;t¼1 eb ef eb ef ð1 þ drÞQf ;t¼1 ðeb ef Þ BC ¼ BQf ;t¼2

kf Q2f;t¼2 eb ef

þ

cfV;t¼1

Qf ;t¼1 Qf ;t¼0

bf

eb ef

1 þ dr

cb

Tt¼2

¼0

ðA1Þ

ðA2Þ


br Qr;t¼1 cr;t¼1 V br Qr;t¼2 kr Q2r;t¼1 Qr;t¼0 BC cr;t¼1 V cb ¼0 ¼ þ Tt¼1 þ BQr;t¼1 eb er eb er ð1 þ drÞQr;t¼1 ðeb er Þ BC ¼ BQr;t¼2

kr Q2r;t¼2 eb er

þ

cr;t¼1 V

Qr;t¼1 Qr;t¼0

br

eb er

1 þ dr

cr

Tt¼2

¼0

241

ðA3Þ

ðA4Þ

Partial derivative equations for model 2, 3, and 4 are similar to those presented here and are available upon request from the authors. References Argote, L., Epple, D., 1990. Learning curves in manufacturing. Science 247 (4945), 920 – 924. Arrow, K., 1962. The economic implications of learning by doing. Review of Economic Studies 29, 155 – 173. Arthur, B., 1989. Competing technologies, increasing returns, and lock-in by historical events. Economic Journal 99, 116 – 131. Banerjee, A., 1992. A simple model of herd behavior. The Quarterly Journal of Economics 107 (3), 797 – 817. Boston Consulting Group, 1968. Perspectives on Experience. Boston Consulting Group, Inc. Dutton, J., Thomas, A., 1984. Treating progress functions as a managerial opportunity. Academy of Management Review 9 (2), 235 – 247. Ficher, C., Newell, R., 2004. Environmental and technology policies for climate change and renewable energy. Resources for the Future Discussion Paper, pp. 04 – 05. Griliches, Z., 1979. Issues in assessing the contribution of research and development to productivity growth. Bell Journal of Economics 10 (1), 92 – 116. Harrison, K., 1999. Talking the donkey: cooperative approaches to environmental protection. Journal of Industrial Ecology 2 (3), 51 – 72. Irwin, D., Klenow, P., 1994. Learning-by-doing spillovers in the semiconductor industry. Journal of Political Economy 102, 1200 – 1227. Jaffe, A., 1986. Technological opportunity and spillovers of R and D: evidence from firms’ patents, profits, and market value. American Economic Review 76 (5), 984 – 1001. Jaffe, A., Newell, R., Stavins, R., 2000. Technological change and the environment. In: Ma¨ler, Karl-Go¨ran, Vincent, Jeffrey (Eds.), Handbook of Environmental Economics. North-Holland/Elsevier Science, Amsterdam. Jaffe, A., Newell, R., Stavins, R., 2002. Environmental policy and technological change. Environment and Resource Economics 22, 41 – 69. Jung, C., Krutilla, K., Boyd, R., 1996. Incentives for advanced pollution abatement technology at the industry level: an evaluation of policy alternatives. Journal of Environmental Economics and Management 30 (1), 95 – 111. Keohane, N., 1999. Policy instruments and the diffusion of pollution abatement technology. Harvard University Discussion Paper. Khanna, M., 2001. Non-mandatory approaches to environmental protection. Journal of Economic Surveys 15 (3), 291 – 324. Loeschel, A., 2002. Technical change in economic models of environmental policy: a survey. Ecological Economics 43, 105 – 126. Manne, A., Richels, R., 2004. The impact of learning-by-doing on the timing and costs of CO2 abatement. Energy Economics 26, 603 – 619. Margolis, R., 2003. Photovoltaic technology experience curves and markets. Presentation at NCPV and Solar Program Review Meeting, Denver CO, March 24, 2003. McDonald, A., Schrattenholzer, L., 2001. Learning rates for energy technologies. Energy Policy 29, 255 – 261. Milliman, S., Prince, R., 1989. Firm incentives to promote technological change in pollution control. Journal of Environmental Economics and Management 17, 247 – 265. Neij, L., 1997. Use of experience curves to analyze the prospects for diffusion and adoption of renewable energy technology. Energy Policy 23 (13), 1099 – 1107.

242


Neuhoff, K., 2004. Large-scale deployment of renewables for electricity generation. CMI Working Paper 59. Cambridge University. Organization for Economic Cooperation and Development, 1999. Voluntary Approaches to Environmental Policy: An Assessment. Paris. Papineau, M., 2005. An economic perspective on experience curves and dynamic economies in renewable energy technologies. Energy Policy 34 (4), 422 – 432. Romer, P., 1990. Endogenous technical change. Journal of Political Economy 98 (5), 71 – 102. Scherer, F., Harhoff, D., Kukies, J., 2000. Uncertainty and the size distribution of rewards from technological innovation. Journal of Evolutionary Economics 10, 175 – 200. Wene, C.-O., 1999. Simulating learning investments for renewable energy technology. EMF/IEA/IEW Workshop, Stanford University CA, June 20–22, 2000. Wright, T., 1936. Factors affecting the cost of airplanes. Journal of Aeronautical Sciences 3, 122 – 128.