Aggregate Shocks and Income Distribution: Can

0 downloads 0 Views 526KB Size Report
Apr 8, 2003 - 2 The exchange rate losses quoted above refer to the change in the price of ... “fundamentals” like monetary expansion and take action at some point, i.e. rush pre-emptively to convert their own .... Rural Tradable Formal (RTF).
rd

3 DRAFT

th

This Version April 8 2003

Aggregate Shocks and Income Distribution: Can Macro-Micro Models Help Identify Winners and Losers from a Devaluation?

Francisco H. G. Ferreira (World Bank), Philippe G. Leite (World Bank) Luiz A. Pereira da Silva (World Bank) and Paulo Picchetti (USP-FIPE)1 [email protected], [email protected] [email protected], [email protected]

Abstract: The macroeconomic crises of the late 1990s in Mexico, East Asia, Russia, Brazil and Argentina rekindled interest in an old subject: can policymakers predict the distributional impacts of macroeconomic shocks and policies? Such an ability would have obvious implications for both social and macroeconomic policymaking. In this paper, we test a novel approach, based on linking an applied general equilibrium model with a microeconomic model of household income determination. The general equilibrium model uses parameters econometrically estimated on macroeconomic time-series. The household income model is econometrically estimated on a household survey cross-section. The two are linked by vectors of changes in prices, wages and employment levels per household groups. We applied this method to the Brazilian devaluation of 1999, when the Real lost 56% of its value. We find that the approach does a reasonable job of predicting aggregate poverty and inequality changes, and changes in the employment structure. It performs poorly in predicting disaggregated changes in the income distribution.

Key Words: Distributional impacts of macroeconomic shocks JEL: D31, D58

1

Picchetti is with FIPE and the Department of Economics, FEA / USP. Ferreira, Leite and Pereira da Silva are with the World Bank. We are grateful, without implication, to François Bourguignon and Anne-Sophie Robilliard for their guidance and inspiration. We acknowledge useful comments received by Armando Castelar Pinheiro on a preliminar version and by Alexandre Samy de Castro. Remaining errors are ours.

Page 1

rd

3 DRAFT

Introduction

In 1995, the Mexican Peso lost 30% of its face value (against the US dollar). It was followed by the Thai Bath and the Malaysian Ringgit in 1997 (46% and 35% respectively), and then by the Indonesian Rupiah and the Russian Rouble in 1998 (42% and 71% respectively). In 1999, it was the turn of the Brazilian 2

Real, which lost 32% of its value as a crawling peg was hastily abandoned, in favor of a floating exchange rate regime. In 2001-2, came Argentina’s and Uruguay’s crises. These episodes have been exhaustively analyzed in their macro-financial dimensions. A large and still growing body of literature (see for instance Glick, Moreno and Spiegel (2001) and Agenor et al. (1999)) has reviewed the common 3

elements of financial crises in emerging markets . The literature has also proposed interpretations of the origin and spread of the crises ranging from a “fundamentalist” view (i.e. the crises resulted from weak macroeconomic and financial fundamentals) to a “financial panic” view (i.e. the crises were self-fulfilling 4

due to investors’ behavior unrelated to economic conditions) .

Regardless of which macroeconomic interpretation prevails, there is widespread agreement that these shocks have had a substantial impact on poverty and on the distribution of incomes in the countries affected. See, for instance, Lokshin and Ravallion (2000) on the case of Russia, and Kakwani (1998) on Thailand. There has also been comparative work on the impact of financial crises in Asia and Latin America on labor markets and household incomes, such as Fallon and Lucas (2002). They conclude that employment fell by much less than production in crisis-hit countries but that there were considerable changes in employment status, location and sectoral composition. They also show that cuts in real wages (due to real depreciation of the currency) were accompanied by small rises in unemployment and that families smoothed their incomes through increased participation and private transfers. Baldacci, Mello and Inchauste (2002) use cross-country analysis to show – with a differences-in-differences methodology - that financial crisis affects negatively income distribution and increase poverty.

2

The exchange rate losses quoted above refer to the change in the price of one unit of the local currency in USD terms, calculated using end-of-period, nominal values, see IMF-IFS line ..AG.ZF 3

All of these financial crisis in emerging markets (a) occurred after significant financial liberalization under rigid exchange rate regimes; (b) were preceded by massive capital inflows that allowed the accumulation of significant unhedged foreign currency liabilities by domestic agents that became illiquid or insolvent when these capital flows suddenly reversed; and (c) tended to cause contagion and spread to other countries.

4

The first-generation explanation of currency crises (Krugman (1979)) stressed the role of inconsistencies between a government actual policies (i.e. leading to an excessive expansion of all sorts of claims in local currency over the country’s stock of foreign exchange reserves) and a government’s commitment to a fixed or pegged exchange rate regime. Agents holding domestic currency would observe the situation with increasing nervousness by looking at “fundamentals” like monetary expansion and take action at some point, i.e. rush pre-emptively to convert their own stock of local currency into reserves and thus trigger the crisis. A second-generation explanation of currency crises, Obstfeld (1994) proposed the idea of governments rationally abandoning an unsustainable exchange rate regime after conducting a cost-benefit analysis of defending it at the expense of growth and employment. These new models of currency crises allow multiple equilibria due to agents’ different expectations, see Masson (1999) and thus gave a theoretical backing for the possibility of “self-fulfilling crisis” where “financial panic” among investors due to herd behavior, imperfect information trigger capital outflows rather than policy inconsistencies.

Page 2

rd

3 DRAFT

However fundamental such descriptive analysis is for establishing the facts and understanding their nature, It falls short of providing a model of the transmission mechanisms through which aggregate shocks in general – and devaluations in particular – affect individual incomes and occupations across the economy. Given its general equilibrium nature, this latter problem has traditionally been approached through computable general equilibrium models (CGEs), where households are represented by representative household groups (RHGs). See, e.g. Adelman and Robinson (1988).

While the literature applying CGEs to developing countries has generated a number of useful insights, its use for addressing distributional questions has been particularly problematic. In addition to the usual problems of lack of transparency and robustness (see Stern, 1989), CGE models have suffered from the fact that changes in individual occupations and earnings can be very heterogeneous within the sectors of economic activity and skill levels traditionally used to construct representative household groups. Bourguignon, Robilliard and Robinson (2002) have shown that, at least in the case of Indonesia in 19981999, the poverty and distributional effects using RHGs can be very different from the effects when real households are used.

In this paper, we follow Bourguignon, Robilliard and Robinson (2002) in combining micro-simulation models running on real household data with a general equilibrium model of the economy. The model operates on two levels: at the “top”, there is a general equilibrium model of the economy, based on a sectorally disaggregated IS-LM framework. This model is described in Section 1. At the “bottom”, there is a reduced-form, three-equation household income determination model, which is described in Section 3. Linking the two are forty-eight linkage aggregated variables (LAVs), which represent the price, wage and employment vectors generated by the macro-model. These are used to adjust constant terms in the equations of the micro-model, which are then run to simulate counterfactual distributions of occupations, earnings, and household income. Counterfactual poverty and inequality measures can be calculated for these counterfactuals, and compared with the corresponding actuals.

We innovate in two respects: first, we do not calibrate the macroeconomic model with ad-hoc parameters. Instead, we obtain as many of them as possible from macro-econometric models estimated on timeseries national accounts and aggregated household survey data from Brazil, 1981-2000. Second, unlike Bourguignon, Robilliard and Robinson (2002), we compare our counterfactual results for 1999 with the actual changes revealed by the 1999 household data available for the country.

Our findings are mixed. The combined macro-micro model predicts the changes in aggregate inequality and poverty indicators reasonably well, and does a reasonable job in predicting the direction and magnitude of changes in the occupational and employment structure of the economy as a result of the

Page 3

rd

3 DRAFT

shock. The model prediction errors for individual earnings and household incomes, however, remain unacceptably large.

The remainder of the paper is structured as follows. The next section describes the structure of macroeconomic model used for Brazil and the event that we are seeking to model: the 1998-99 devaluation of the Real. Section 2 presents the procedure to generate LAVs, the scenario of financial crisis that is simulated and its associated LAVs. Section 3 describes the household-level model of income determination the results of the micro-simulation based on them. The predicted changes in the earnings, income and employment distributions are compared to the actual changes between 1998 and 1999. Section 4 concludes.

1. The macro model: a conventional IS-LM macro-econometric model 5

with a disaggregated labor market and a financial sector .

At the top of our two-level

framework, there is a conventional IS-LM macroeconometric with a

disaggregated labor market and a financial sector. By construction, factor remuneration – by skill and by sector – will be affected by policies and shocks (e.g., interest rate, exchange rate, various taxes affecting the different types of labor costs, factor substitutions, etc).

The model is estimated with 1981-2000 annual (or quarterly) data (national account –NA-- and historical household survey – PNAD – data). The econometric results of the main equations are in Annex 1. The real economy comprises 6 sectors (Urban Formal Tradable; Urban Formal Non-Tradable; Urban Informal; Rural Formal Tradable; Rural Formal Non-Tradable; and Rural Informal); the labor market comprises 3 skills, skilled, semi-skilled and unskilled; labor supply and demand are modeled by sector and skill, yielding an endogenous unemployment level. On the supply side, the model features sector specific production functions with three-level nested CES that follow Fernandes and Menezes-Filho (2001) in defining substitution between capital and labor; as well as across unskilled and semi-skilled workers, and semi-skilled and skilled workers. The financial sector, is modeled after Bourguignon, Branson and de Melo (1989) with 8 financial assets: domestic currency, domestic bank deposits, domestic Government bonds, domestic loans (debt) and equity shares; foreign currency, foreign loans to residents and Government bonds in foreign currency. There are 6 agents: households, private firms, commercial banks, Government, central bank and foreigners. The transmission of financial volatility –external and domestic-to the rest of the model occurs through the changes in agents’ portfolio choices, (e.g., holding of domestic debt, demand for foreign currency, etc). Capital outflows for example can be countervailed by a higher

5

See details in Picchetti, Pereira da Silva and Castro (2003)

Page 4

rd

3 DRAFT

interest rate policy affecting the allocation of wealth and savings, but eventually that also impacts the level of private investment and consumption.

1.1. The IS curve and the real economy model

We use a disaggregated but still standard format that most macroeconomic models are built around: a core neo-keynesian framework (Klein-Goldberger, MPS in the US or DMS, Metric in France), see Artus, Deleau and Malgrange (1986). The functioning of these large models, albeit complex, can be reduced to the interaction of three basic blocks. (1) A real economy block, determining production, components of aggregate demand and factor demand (see below 1.2). (2) A wage-prices block that determines the price level and wage rates. And (3) a financial and monetary block that determines the interest rate. Following the tradition, by making the basic neo-keynesian core functions of blocks (2) and (3) (i.e. prices, wages and interest rate) endogenous, we get the standard IS-LM structure for macroeconomic models where aggregate demand and supply curves can be related to the general price level. Due to its modular structure, the macro model can function under various configurations. In the most simple case where the financial asset market is reduced to one local currency market, we have a standard IS-LM.

The key transmission mechanism in the model between the real economy and the financial markets comes from the linkage of real private consumption and investment to the endogenous domestic interest rate that is determined by equilibrium in the financial sector. In particular: (a) real private consumption is a standard function of disposable income, the general price index and the real deposit interest rate (to account for a “wealth” or portfolio effect); (b) real private investment is decomposed into its building and construction / machinery components; both components follow a standard specification including aggregate demand (an accelerator) and the price of capital --decomposed into the real exchange rate (given the importance of imported equipments) and the domestic working capital interest rate--.

The balance of payments (BoP) is modeled in a fairly detailed way. Real services and real imports and exports of goods are disaggregated into sub-components (e.g., major types of commodities and types of services). The general specification for all these items make each of them dependent (respectively for debit and credit components) on domestic and external demand and relative prices, i.e. the real exchange rate. The current account balance is constructed by accounting identities. Capital movements are assumed to depend only on interest rate differential (domestic and foreign) and the expected depreciation of the exchange rate.

Our macro model is constructed in separate modules that can operate independently . As such, the model can function as a pure keynesian model i.e. when equilibrium is restricted to the goods market. Assuming fixed prices, fixed exchange rate, fixed wages and fixed interest rates and an exogenous level of public spending, the equilibrium (closure) condition will yield the traditional multiplier of public spending.

Page 5

rd

3 DRAFT

Block 1 run in sequence (on top of) Block 2 and with no interaction between wages — assumed exogenous — provides a simple pure Keynesian demand-driven model, i.e. only changes in real aggregate demand variables affect the level of production.

Similarly, the equilibrium of the BoP equates the changes in reserves as resulting from the trade balance and net capital movements. There are two possible exchange rate regimes. Under a fixed regime, the central bank intervenes in the foreign exchange market to maintain a fixed parity or a crawling peg path vis-à-vis a specific foreign currency target (say, the USD). Thus, the change in the demand for money by households is affected by this foreign component (an exogenous element in the supply of money). Under a flexible regime, the central bank does not intervene and the BoP equation determines the freely floating exchange rate. Hence, here, the exchange rate is determined in the BoP and the money market equilibrium is simply solved by the domestic component of supply and demand.

1.2. Factor Demand in the Real Economy model a) Aggregate Supply and Demand Modeling Strategies

The main motivation behind the break up of supply into different sectors is the ability of the model to differentiate between effects from macro and external shocks, from changes in legislation that potentially affect the allocation of workers between formal and informal sectors of the economy, and also from changes in production composition between agricultural and industrial products. Accordingly, the supply side in the model is divided into six sectors: • • • • • •

Urban Tradable Formal (UTF) Urban Non-Tradable Formal (UNF) Urban Non-Tradable Informal (UNI) Rural Tradable Formal (RTF) Rural Non-Tradable Formal (RNF) Rural Non-Tradable Informal (RNI)

For each of these sectors, production is modeled as value-added, leading to demand for production factors.

b) Factor demand functions

Factor demand functions determine demand for capital and labor by skill-level. In order to relate the results produced by the model at this level, in terms of employment and earnings for workers in these sectors, to the household survey data used in the micro simulation stage of the modeling strategy, the classification of workers in relation to skill-level had to be done in terms of observed characteristics for the individuals. We chose to define skill-level according to years of formal education. Low-skill level workers

Page 6

rd

3 DRAFT

have between zero and four years of education, whereas intermediate-skilled workers have between five and eleven years, and high-skilled workers have more than eleven years.

The demands for these different kinds of labor and for capital come from the application of Shephard’s Lemma to the Production function for y of each of the six sectors, which is represented by a three-level nested CES. The motivation behind this approach is to provide flexibility between rates of substitution between capital and labor, and between different kinds of labor. The first level of the CES allows for substitution between capital and a composite measure of labor. In the second level, this composite labor can be decomposed between qualified and unqualified jobs. The third level accounts for the fact that unqualified jobs can either be performed by low-skill or intermediate-skill workers, whereas qualified jobs can either be performed by intermediate-skill or high-skill workers. Therefore, as in Fernandes and Menezes-Filho (2001), we assume that there is substitution between all kinds of labor, except between high-skill and low-skill labor. The production function can then be represented as:

y = β λ K

−ρ

+ (1 − λ )La  −ρ



1

ρ

(1)

where:

La = δ LQ

−ψ

+ (1 − δ )LU  −ψ



−ε LQ = φ LiQ + (1 − φ )L−hε 



LU = ξ L−l γ + (1 − ξ )L−iUγ 



1

ψ

(2)

1

ε

(3)

1

γ

(4)

K = Capital La = Composite Labor LQ = Composite Labor for qualified jobs, can be performed either by Li (intermediate-skill workers) or by Lh (high-skill workers) Lu = Composite Labor for unqualified jobs, can be performed either by Li or by Ll (low-skill workers).   σ ka   σ QU Elasticities of substitution:  σ  ih   σ li 

= = = =

1 1 1 1

1 + ρ 1 + ψ 1 + ε 1 + γ

(5)

Page 7

rd

3 DRAFT

At the sectoral level, then demand for capital coming from our production function provides the disaggregated demand for investment, which provides the flexibility of responding differently to relative prices of final production, of relative prices of labor, and to sectoral levels of production.

Therefore, at this stage we have modeled 24 Linkage Aggregate Variables (LAVs), specifying for each of the six sectors demand for three different kinds of labor and for capital.

c) Conversion Matrix

Our modeling strategy implies that the number and definition of final demand sectors are different form those of production sectors. This creates the issue of reconciling the output and price variables in the two sides of the model. Our approach follows Fisher, Klein and Shinkai (1965). On the one hand, the output demanded by the final demand sectors must be distributed over the production sectors; on the other hand, the prices generated by the price-formation equations in the production sectors must be aggregated to obtain prices for the final demand sectors.

In our model, data are available on total output by production sector, coming from national accounts. In the case of informal sector, there is no readily available statistic on production. Therefore, value-added for these sectors was estimated based on reported incomes from informal workers in the household surveys.

The output conversion matrix was estimated with the natural restrictions (e.g., informal non-tradable sectors do not export), and the resulting weights were then used to convert sector prices into GNP deflators.

Page 8

UTF_Y

UTF

UNF_Y

UNF

RTF_K

RNF_Y

RNF

Aggregation Matrix

UNI_K

RTF_Y

RTF

UNF_K UTF_K

UNI_Y

UNI

RNI_K

RNI_Y

RNF_K

RNI

Figure 1: An overview of the main blocks of the macro model

Loans Payment s

Page 9

Imports

Loans Government Consumption Exports Payment s

Financial Sector

Labor Market

Foreign Sector

Private Consumption

Reserve s

Central Bank

Government

Taxes

Labor Income

Transfer s

- Taxes

rd

Private Consumption

Disposable Income

3 DRAFT

rd

3 DRAFT

1.3. The LM curve and the Financial sector

In textbooks the IS-LM classic framework simply adds to the pure Keynesian model above another market that has to be simultaneously in equilibrium (the money market). Hence, the impact on the real economy (the IS curve) of changes in fiscal policy depends on the money market equilibrium (the LM curve) taken as the simplest equilibrium in the local currency market with fixed exchange rate. Of course, the multiplier of public spending is lower because the LM curves intersects the IS curve “at a lower level of output” than the equilibrium obtained through the pure keynesian model, hence bringing the contractionnary effect of higher interest rates on aggregate demand components such as private consumption and investment. The LM curve is the more horizontal the more capital movements are sensitive to interest rate differential and that the propensity to import is small. In particular, the standard result (Mundell) is that the multiplier of expansionary fiscal policy is higher, the more capital movements are sensitive to interest rate and the sensitivity of the trade balance in nominal terms to domestic activity is small. In a situation of flexible exchange rates, the exchange rate adjusts to yield the equilibrium of the foreign exchange market.

In our macro model we endogenize the behavior of prices and wages, relating them to the IS-LM framework, and therefore getting the aggregate supply curve where production depends on the price level (that is itself endogenous). With a proper specification, output will have the familiar shape of a positive relation between prices and production (provided that the mark-up of wages on prices does not exceed productivity). The aggregate supply curve will have a positive slope to the price level. The slope is increasing in the mark-ups, and decreasing in productivity. The form and speed of the adjustment in the labor market will have important consequences for the way cycles will be built into the framework.

Labor supply is modeled by skill-level, relating the participation decision to the price level and nominal wages. Therefore, labor market functioning is obtained at the aggregate level comparing supply with the aggregation of demands at the sectoral level. Some parameters relating wages paid by the employers and actually received by workers are aimed at capturing different institutional and fiscal scenarios in terms of social contributions and income taxes. Future exercises with this modeling strategy aim at relating these parameter choices to the endogenously determined levels of employment and earnings. In addition to the standard characteristics of the LM curve above, we modeled a more complete financial sector with several agents and markets. Our financial sector model follows roughly the “maquette” established by Bourguignon, Branson and de Melo (1989). Modeling specifically a financial sector was thought to be necessary to refine the transmission of financial crises – external and domestic – to the rest of the model, in particular to the disaggregated sectoral demand for labor and the real macro variables (real private consumption and investment) by including “portfolio choices” by agents.

Page 10

rd

3 DRAFT

A more complete financial sector model can capture the following transmission channels of financial crises (see details in Picchetti, Pereira da Silva and Castro (2003)). •

First, the demand pressure on the post-collapsed exchange rate that constitutes an essential element of the recent financial crises (e.g., agents shifting preferences from local to foreign currency holding, etc.). Devaluation and then depreciation transmit into domestic prices, the changes in the real exchange rate affects demand for exports and imports, private consumption and investment and the re-allocates financial assets in the portfolio of households, firms and banks.



Second, the worsening of the crowding-out effect – on private investment – of large public sector deficit financing through the issuance of domestic Government bonds. Under financial repression, bank credit to private firms by risk averse domestic commercial banks – the major holders of domestic Government bonds – is a residual of the headroom in their balance sheets – given existing domestic regulations – once they have purchased the supply of (less risky) domestic Government bonds. A financial crisis reduces this headroom further and usually triggers credit crunches because of (a) the currency mismatches in banks’ balance sheets – growing liabilities due to the devaluation – and (b) the usual increase in non-performing loans calling for provisioning and/or write-offs.



Third, the short-term effect of hikes in the country’s base policy rate that can result from either (a) the need under a pegged-fixed exchange rate to defend the regime by matching the rise in the country risk premium and the expected devaluation; or (b) the need under a floating exchange rate regime with inflation targeting to establish the credibility of the anti-inflation stance of the central bank. In any event, the resulting policy interest rate determined by the central bank (the Selic), through the spread structure of domestic interest rates, eventually impacts the real economy –real private consumption and investment-- and asset holdings.

The following characteristics of the financial sector are worth mentioning in more detail:

a) Agents and financial assets

There are 8 types of financial instruments in the financial sector of the macro model. The first 5 are denominated in local currency and are: (1) currency, (2) household time deposits, (3) Government bonds, (4) loans (debt) to households and to firms, and (5) equity shares issued by firms. In addition, there are 3 other instruments denominated in foreign currency: (6) cash or foreign currency itself, (7) foreign loans to residents in foreign currency and (8) Government bonds issued in foreign currency. We assume that there are no domestic real asset markets (e.g., real estate, gold, etc.). The 8 instruments can be issued / held by 6 categories of agents, 5 are domestic i.e. households, private sector firms, commercial banks,

Page 11

rd

3 DRAFT

the Government, the central bank and one agent regroups all foreigners or the rest of the World. Bonds are issued by the Government only. Debt instruments are issued by firms only.

b) Portfolio choices by agent and by layers

Each of the agents is represented by its savings account. Most of the action comes from the way agents allocate their net savings and financing their borrowing requirements. Following Bourguignon, Robilliard and Robinson et al (2002). the model chosen is a two stage process. For example, for creditors agents such as households, the first stage is a simple demand for domestic currency. Then, households would decide between bank deposits, shares (firms’ equity) depending on the relative yield of these instruments. Banks would set deposit’s interest rates so as to capture households’ deposits and would then choose assets to cover their operating costs (payment of interest on deposits). Similarly, firms decide between the composition of their debt, e.g., determining the share of their borrowing requirement going to foreign or domestic financing based on relative cost of financing. If prevailing relative domestic borrowing rates for firms are high, despite foreign exchange risk, firms will most likely opt for an external financing strategy. The Government finances its borrowing requirements through issuing bonds in both foreign and local markets. Foreign issuing is limited and exogenous. Domestic government bonds, in turn, need at least to provide the yields that allow domestic commercial banks to remunerate household deposits. One condition thus for the financial sector is to have a structure of domestic interest rate spreads that equilibrates the portfolio of each of these agents.

c) The structure of domestic interest rates

Three domestic interest rates in local currency are endogenous in the model. The base –central bank rate—rate or the Selic when simulations are run after 1999; the domestic deposit rate for household deposits; the domestic borrowing rate for firms –working capital interest rate--. There is, in addition, one exogenous foreign –a short-term Libor—interest rate in USD dollar. But recall that Brazil went through an abrupt change in 1999 of its exchange rate regime, from an exchange rate anchor to a floating exchange rate with inflation targeting, with a corresponding change in emphasis for the Selic. We integrated this regime change for post-1999 simulations where the inflation-targeting objective prevails. For the 1999 simulation we decided not to model the base –central bank rate—rate. Once the base rate is defined, the structure of interest rates (domestic borrowing and deposit rates) is determined by the modeling of the spreads (see Favero and Giavazzi (2002), Cardoso (2002)). Spread over the base real policy rate will determine the real deposit rate given the supply of bank deposits by households and their first layer choice between local and foreign currency. Spread over the deposit rate will determine the real working capital rate given the supply of new credit by commercial banks after their first layer choice for Government domestic bonds and the demand for new credit by firms.

Page 12

rd

3 DRAFT

1.4. Estimation of the Macro-econometric Model

a) General Characteristics

The present version of the macro-econometric model, which generated the LAVs used in our simulations, comprises a total of 267 equations. About one third of these are identities, whereas the rest are estimated as behavioral equations. Most of the behavioral equations for the real economy were estimated in OLS with an error-correction mechanism.

The real economy part of the macro model comprises 72 equations, of which the BoP represents 42. The factor demand model comprises 68 equations and the financial sector model comprises 85 equations.

The availability of data imposes some constraints on the choices of estimators. Data on wages and employment for the disaggregated labor market could only be obtained through the household surveys, which start in the late seventies, and are available on a yearly frequency. Likewise, data for the financial sector is not available with consistent methodologies for a time span allowing a large sample. In spite of these inevitable constraints, we estimated several equations using two-stage least-squares when endogeneity of the regressors were considered to be an issue.

Since the model is estimated with time-series data, some equations were specified with a dynamic structure allowing for dynamics in the solution of the model. In the present version, the simulations describe essentially movements between long-term solutions in levels. However, one of the main planned extensions involves enhancement of the dynamics of the solutions by estimating movements of the variables in differences, through an error-correction mechanism. The idea, in line with the basic motivation for the proposed model, is to gain further insight into the paths of adjustment of the endogenous variables in response to shocks, allowing the analysis of trade-offs not only in terms of macro stability vs. social indicators, but also in terms of the time periods involved. A very relevant analysis in this context is the comparison of two alternate policies at the macro level, which involve some tradeoffs in terms of macro indicators, but are substantially different when it comes to the time involved in the transition from one situation in terms of social indicators to another.

b) Major Results of an Historical Simulation The real economy model fares reasonably well in an historical simulation mode where the 1999 financial crisis is simulated with historical exogenous data.

Table 1 below presents partial results.

The run

captures the slowdown in real private consumption and the fall in real private investment, which explain the modest real GDP growth rate in 1999. The major components of the external sector balances are

Page 13

rd

3 DRAFT

also reasonably well captured by the simulation. We should note that the Brazilian financial crisis is different from the massive output contraction that characterized the other financial crisis of the late 1990s.

Table 1: Some Results of the Real Economy Model for an Historical Simulation for 1999 Real Growth Rate

1998

1999

Actual

179,982

184,087

-7.2%

Simulated

174,535

174,050

-8.5%

Summary of Some Macro Aggregates

Error on real growth

Private Investment (nominal BRL million) -1.2%

Private Consumption (nominal BRL million) Actual

566,192

597,418

0.9%

Simulated

582,523

633,615

1.5%

0.6%

Gross Domestic Product (nominal BRL million) Actual

914,188

963,869

0.8%

Simulated

908,778

973,848

1.0%

0.1%

Exports of Basic Commodities (volume) Actual

120.20

130.60

8.7%

Simulated

119.37

126.80

6.2%

Actual

121.50

96.40

-20.7%

Simulated

130.16

92.26

-29.1%

-2.4%

Imports of Capital Goods (volume) -8.5%

Current Account Balance (USD Million) Actual

(33,435)

(25,335)

-24.2%

Simulated

(34,296)

(27,584)

-19.6%

Actual

29,381

16,982

-42.2%

Simulated

28,690

11,543

-59.8%

4.7%

Capital Account (USD Million) -17.6%

Source: Authors’ estimates based on Picchetti, Pereira da Silva and Castro (2003)

The financial sector model also depicts reasonably well in an historical simulation mode, the 1999 financial crisis (see some results in Table 2 below). Stocks, issuance and holdings of the key financial asset – Government domestic bonds—is relatively well captured by the financial sector model, although bank deposits by households is somehow underestimated. The model also captures adequately the path of changes in domestic prices (general price index, the CPI and the WPI) that are brought by the 1999 crisis. The change in regime and the “devaluation” of the Real resulting in a loss of 32.4% of its end period value against the USD is exogenous in this historical simulation. The model is also slightly overestimating the domestic interest rates.

Page 14

rd

3 DRAFT

Table 2: Major Results of the Financial Sector Model for an Historical Simulation for 1999

Summary of Some Financial Aggregates

1999 actual

1999 simul.

Error / Actual

Stock of Gov. Domestic Bonds outside BACEN (nominal BRL million) 464,507

464,507

0.0%

96,204

-1.3%

Stock of Government External Debt (USD million) 97,448

Stock of Bank Deposits by Households (nominal BRL million) 248,820

276,333

11.1%

Inflation measured by General Price Index (IGP-M)

11.300%

9.900%

Consumer Price Index (CPI)

4.865%

4.900%

0.7%

Wholesale Price Index (WPI)

16.600%

17.100%

3.0%

99.23

105.76

6.6%

Deposit

22.77%

25.30%

11.1%

Lending (working capital)

30.11%

32.02%

6.3%

Real Exchange Rate (index)

-12.4%

Domestic Interest Rates

Source: Authors’ estimates based on Picchetti, Pereira da Silva and Castro (2003)

2. Generating the LAVs to link the macro and micro models

2.1. The 1998-99 crisis in the macro model

During most of the first 4 year of the Real Plan, Brazil maintained a crawling-peg exchange rate regime (ERR). The event that the macro-financial model simulates is a Baseline (historical) scenario corresponding to the abandonment of the crawling-peg exchange rate regime. The crisis started around the third and fourth quarter of 1998 with pressure on the exchange rate through capital outflows and continued during the first quarter of 1999 after the de-pegging of the Real The counter-factual simulation in the exercise captures the crisis through a change the following variables:

The “float” of the currency on January 15, 1999 whose average annual parity with the USD goes up from R$1.161 to one USD to R$1.816 (e.g., a fall of 56.4% or a loss of 32.4% of its end-period value against the USD).

A temporary rise the central bank policy rate (BACEN’s Selic) during the period corresponding to October 1998 till May 1999. The monthly rate was raised from 1.47% in August 1998 up to 3.33% in March 1999 (corresponding to annualized rates of almost 50%). However, thanks to the rapid resolution of the crisis,

Page 15

rd

3 DRAFT

the annual average base nominal rate in 1999 ended up actually lower than in 1998. In nominal terms, the Selic was set in 1997, 1998 and 1999 respectively at 24.8%, 28.8% and 25.6%, corresponding to real average rates respectively of 16.1%, 26.6% and 4.7%.

A renegotiation of the terms of the SBA arrangement with the IMF in order to strengthen the credibility of the policy framework; and hence a tightening in the fiscal stance corresponding to a reduction of the Consolidated Public Sector Borrowing requirements (PSBR) from BRL68 billion to BRL56 billion (7.5% of GDP down to 5.8% of GDP, i.e. a cut of BRL12 billion or 1.7% of GDP) as depicted in Table 3 below:

Table 3: The changes in the fiscal policy stance to tackle the 1999 crisis 1998

In Millions of Current Reais

1999

Difference

PSBR, Nominal

68,229

56,284

Nominal Interest Payments

68,335

87,372

19,036

(31,087)

(30,981)

Primary Consolidated Fiscal Position Nominal Gross Domestic product In Percent of GDP

(106) 914,188

973,846

1998

1999

(11,945)

PSBR, Nominal

7.5

5.8

Nominal Interest Payments

7.5

9.0

1.5

Primary Consolidated Fiscal Surplus

0.0

-3.2

-3.2

-1.7

**From Table: Necessidades de financiamento do setor público - Fluxos em 12 meses - valores correntes BRL M

Among other aspects of the policy changes were the implementation of an inflation target anchor in 1999 as a replacement to the ER anchor; the provision of hedge to market participants (through the issuance of Government “foreign-exchange-indexed” domestic bonds) and, undisclosed occasional interventions on the spot market, drawing on international reserves, - within the limits agreed upon under the new arrangement with the IMF - .

2.2. The LAVs

The macroeconomic module generates twenty-four LAVs for occupational status (2 locations –rural and urban-- x 3 types of educational attainment –low, intermediate and high-- x 4 types of activity –formal tradable, formal non-tradable, informal and unemployment-, corresponding eighteen LAVs for incomes, and six LAVS for changes in prices. There are thus forty-eight LAVs in total.

The first 24 LAVs (in Table 4(a) are proportional changes in numbers of people in occupations, one for each of the four occupational categories (formal tradable, formal non-tradable, informal and unemployment), in each household group (urban or rural) and for each level (low, intermediate or high) of education).The next 18 LAVs (in Table 4(b) are proportional changes in wage rates, for each of the six

Page 16

rd

3 DRAFT

household groups (urban low-skilled, urban intermediate-skilled, urban high-skilled, rural low-skilled, rural intermediate-skilled, rural high-skilled).

The LAVs – i.e. the disaggregated macro-financial model – are telling the following story for occupational changes of the 1998-1999 crisis In urban areas, the crisis produces an overall decline in employment, a fall in formal sector employment (both tradable and non-tradable) particularly for low and high educational levels and an increase in informality in the labor market. In parallel, there is a relative increase of employment of workers with an intermediate level of education, as if there was a “down-grading” of workers with higher level of education. In contrast and in the rural areas, the crisis produces a positive response leading to an increase in employment in the tradable sectors at the low and intermediate levels of education. Unemployment falls at all levels of education.

In terms of nominal earnings (wage income), see column LAVs only in Table 4(b) the story is the following. In urban areas, there is an overall fall – across all occupational categories and levels of education – in nominal wage incomes in 1999 compared to 1998. In rural areas, conversely, there is an increase in nominal wage earnings except for workers with low education in the formal tradable sector and those with intermediate level of education in the formal non-tradable sector.

Page 17

Source: Authors’ estimates based on PNAD/IBGE 1998/1999

Page 18

Table 4(a) : The 24 Occupational Linkage Aggregate Variables (Nominal terms, Occupational status)

rd

3 DRAFT

Source: Authors’ estimates based on PNAD/IBGE 1998/1999

Page 19

Table 4(b) : The 18 Income Linkage Aggregate Variables (Nominal terms, Income)

rd

3 DRAFT

rd

3 DRAFT

3. The micro simulations

Standard applied general equilibrium analysis would stop at the LAVs. These represent the changes in nominal wages, prices and occupations for the six representative household types in our macro model. From the wages and prices, one would construct changes in real earnings. From the changes in employment and unemployment for each household type, one would reweigh the population shares to whom those earnings should accrue. Coupled to some functional form assumption (such as lognormal density functions within each RHG, these changes would generate a complete counterfactual income distribution.

That approach ignores heterogeneity in agent response to the devaluation within representative household groups. For instance, the occupational responses to a devaluation may differ across men and women within the same groups; or across women with different numbers of children. Or yet across workers with the same levels of education, but difference age and experience profiles. Earnings changes might be different depending on whether the informal non-tradable job is in manufacturing in a unionized sector or in own-account service provision in an urban slum. In order to capture some of the sources of heterogeneity across the diverse population of individuals and households lumped into these six broad groups of agents, we estimate a simple reduced form model of household income determination, which is based on Bourguignon, Ferreira and Lustig (1998) and Ferreira and Barros (1999). Once the model has been estimated, it can be used to simulate individual and household responses to the changes in the sectoral mean changes (in employment probabilities and in earnings), while respecting the conditional distribution of wages and employment on observed individual characteristics.

The model consists of three equations: one for the determinants of the occupational choice of the worker; another for the determinants of his or her earnings in that occupation, and a final rule for the aggregation of individual incomes into household incomes. The first equation represents the constrained choice of occupation by the worker as a function of his household and individual characteristics:

I j = s = I (zihγ s + ηih > zihγ j + ηih ∀j ≠ s )

(6)

I is an indicator function, which takes the value one if the inequality within the bracket holds, and zero otherwise. Z is a vector of observed individual and household characteristics, and η captures unobserved individual-level determinants of occupational choice. Elements of z include: education; labor market experience; gender; race; occupation of head; education of head; experience of head; race of head; region of Brazil; dummy if metropolitan area; housing status and a categorical variable for other incomes.

Page 20

rd

3 DRAFT

This occupational choice model may be estimated empirically by means of a discrete choice model such as a multinomial logit, where the frequency of persons in occupation s is modeled as:

Pr( j = s) =

e zihγ s +ηih

∑e

(7)

zihγ j +ηih

j

Six such models, with identical specifications, were estimated: one for household heads, another one for spouses, and a third for other household members, in each of rural and urban areas. The exact specification of the model, as well as the estimation results for all six categories of individuals (with inactivity as the base category) are presented in Annex 2, table 10(a) and 10(b). Marginal effects calculated from the estimated multinomial logit coefficients are presented in Annex 2, table 11.

Individuals choose among five possible occupations: inactivity; unemployment; work in the informal sector; work in the formal tradable sector and work in the formal non-tradable sector. They make that choice according to whether the criterion within the bracket is higher for that sector than for any of the other four. The parameter vector γ is specific to each occupation and can be interpreted in two different ways: either as a vector of the marginal utilities of each characteristic in z, in the occupation s; or as a descriptive parameter of the distribution of observed occupations, conditional on the elements of z.

The second equation of the model is a standard Mincer-type earnings equation:

log wih = α gs + xih β g + ε ih

(8)

This equation relates the earnings (w) of individual i in household h to his or her observed (x) and unobserved (ε) characteristics in the standard manner. The model is allowed to vary across occupations (denoted by s) and across household groups (g). The population was partitioned into the six sectors used in the macro model (Urban Formal Tradable; Urban Formal Non-Tradable; Urban Informal; Rural Formal Tradable; Rural Formal Non-Tradable; and Rural Informal), . Along the educational dimension, there are three groups: low (0-4 years of schooling); intermediate (5-11 years of schooling) and high (12 or more years of schooling). The interaction of these two dimensions yields a partition into eighteen groups. In each of these groups, the vector x includes the following characteristics: intercept, education, experience, occupation, race, region of Brazil; a gender dummy and a dummy for metropolitan area.

As indicated in Equation (8), a separate model is estimated for each of the eighteen sub-samples, corresponding to the aforementioned partition The full specification of the models, as well as the

Page 21

rd

3 DRAFT

estimation results for all eighteen groups in the base year of 1998 are presented in Annex 2, table 10(a) and 10(b).

yh =

6

1  nh ∑ n h  i =1

3

∑I s =1

s

 w ihs + y 0 h  

(9)

The third equation aggregates incomes across workers in their different groups and occupations, as well as non-labor incomes from the data set, into household incomes. Since we are interested in a measure of welfare, we choose to work with per capita household incomes, given as follows:

It is with respect to this variable (yh) that poverty and inequality measures are computed. Tables 10-12 in Annex 2 presented the results of estimating the above reduced-form model of household income determination for Brazil in 1998. We now describe the procedure employed to simulate a counterfactual distribution of earnings, and one of household incomes, based on the 1998 data and on the linkage aggregate variables generated by the macroeconomic module (described in Section 2).

Formally, the micro-simulations consist of finding the solution of the following system of twenty-four equations:

(

)

I sg zihγˆs + ηihs > zihγˆ j + ηihj ∀j ≠ s = f sg ......∀s, g

(10)

∑∑ Exp(αˆ

(11)

i∈g s∈g

gs

+ xih β g + ε ih ) = π sgω sg ........∀g

Equation (10) actually corresponds to the first six equations of the system: one for household heads, one for spouses, and one for other household members, separately in rural and urban areas. Equation (11) corresponds to the remaining eighteen equations: one for each of the sector-education cell for which an earnings regression was separately estimated. This system is over-identified if we allow for more than one element in each vector γ to vary. We therefore solve it for 24 unknowns, exactly: six γ0 and eighteen α terms.

6

It should be noted that, since the dependent variable in these regressions is total earnings, the omission of an hours variable on the right hand side implies that the coefficients are biased estimates of rates of return and should be interpreted as reflecting a labour supply component as well. This does not matter to the simulation results, provided both labour supply behaviour and the structure of returns are stable across counterfactuals.

Page 22

rd

3 DRAFT

The interpretation is the following. The first six equations - represented by (10) - require that the intercept term of the multinomial logit for occupation s (relative to inactivity) be such that the fraction of the overall population who belong to household group g and choose to work in occupation s be equal to the share of the population (who belong to that household group) which is predicted by the macroeconomic module to 7

be employed in occupation s. The remaining eighteen equations - represented by (11) - require that the intercept term of the earnings equation estimated for household group g be such that the mean real wage in the counterfactual distribution be equal to the sector/group wage predicted by the relevant LAV from the macro module.

The system is fully simultaneous, and is solved numerically by the application of a Newton-Raphson algorithm, which essentially alters values of the twelve unknowns progressively, so as to minimize the sum of squared differences between the left- and the right-hand sides of Equations (10) and (11). This procedure is analogous to the one used by Bourguignon, Robilliard and Robinson (2002). Like them, we offer no formal existence or uniqueness proofs for the equilibrium of the system. Like theirs too, our algorithm does converge to a seemingly plausible equilibrium. In the next section, we turn to a description of these results.

3.2. Results from the complete macro-micro simulation Once the system represented by (10)-(11) has been solved, yielding a vector of six γ0 and eighteen α terms, these can be substituted into the income-generating equations (6), (8) and (9). The predicted values of equation (6) will determine the new distribution of occupations in the population, in response to the shock. This new distribution reallocates workers across five occupations (inactivity, unemployment, employment in the formal tradable sector, employment in the formal non-tradable sector and employment in the informal sector),

Taking the new individual occupations into account, equation (8) determines the new predicted earnings for each employed worker. Equation (9) aggregates across the resulting counterfactual earnings distribution, thus generating a counterfactual distribution of household incomes. The result is a counterfactual occupational structure, a counterfactual distribution of earnings for main occupations, and 8

a counterfactual household per capita income distribution. Each is consistent both with the actual

7

One difficulty is that, whereas the macro module allows for changes in the relative skill composition of labour demand when constructing the employment LAVs, this micro-simulation does not allow for changes in the educational level of the worker. This may be economically realistic for the short term, but it implies that there are six actual g unknowns for potentially eighteen exogenous variables (fs targets). In the simulations, the adjustment occurs through the number of people left over for unemployment from each skill category, which corresponds to a quantity closure to the labour market. 8 It is important to bear in mind that these counterfactual distributions assume that a number of features of the population and economy remained constant at their 1998 levels. The spatial, racial, gender and educational

Page 23

rd

3 DRAFT

conditional earnings and occupational distributions observed in 1998, and with the predictions of the macroeconomic model for the effects of the devaluation on the Brazilian economy. In what follows, we call these distributions the counterfactual 1999 distributions, and compare them with the actual distributions observed in 1999.

a) The main results for the counterfactual structure of occupations

The main results for the counterfactual structure of occupations is presented in Table 5. The column under the heading "1998 starting point, household survey (PNAD) data" give the absolute frequency of workers in each category,. The next two columns, under the heading "1999 Simulated by Macro --> Micro models" present the corresponding counterfactual employment numbers predicted by both the macro and micro simulation models, and the proportional change implied by this number, with respect to the 1998 actual. The next two columns, under the heading “1999 comparison point, household survey (PNAD) data", present the real employment numbers from the 1999 PNAD, and the actual changes. The last two columns give the predictions differences between the counterfactual prediction and the actual changes.

A comparison of the actual/simulated columns in Table 4(a) illustrates the performance of the combined macro-micro model of the devaluation, as regards the distribution of employment. The model performed well for urban household groups of low or high skills. It performed less well for rural households with low and high levels of education. In essence, the model seems to be missing the occupational status of workers with intermediate levels of education.

In the urban areas for those workers with an intermediate level of education in the formal sectors (both tradable and non-tradable), predictions consistently over-estimated their employment level. And these errors were significant in size (about 13 to 17%) and with the wrong sign (employment actually contracted). In the rural areas, the prediction did not miss the direction of change but over-estimated in particular the employment of these workers in the formal tradable sector (and the informal sector –with a wrong sign). The performance of the model is reasonable for occupations in urban areas with the exception of the formal sector for intermediate skill level. It misses the sign of 2 out of 18 LAVs and the average simulation error – with these two exceptions – is around 2 to 4 percentage points.

composition of the population, the distribution of non-labor incomes, and the internal composition of the households are some of these features.

Page 24

rd

3 DRAFT

Table 5: Aggregate results from the macro-micro models – Occupations

1998 starting point, household survey (PNAD) data

1999 Simulated by 3 layer Macro --> Micro models

Number of Number of observations (in units) observations (in units)

Occupational categories*

Total Labor Force Low skill Intermediate skill High skill Unemployed Formal sector (tradable & non-tradable) Informal sector Urban employment Low skill Intermediate skill High skill Unemployed Formal sector (tradable & non-tradable) Informal sector Rural employment Low skill Intermediate skill Unemployment Formal sector (tradable & non-tradable) Informal sector

23,420,100 27,181,832 5,916,696 14,015,110 26,841,421 29,677,207 47,000,928 16,217,373 24,866,859 5,916,696 8,656,065 24,478,761 22,522,167 9,517,700 7,202,727 2,314,973 5,359,045 2,362,660 7,155,040

23,334,390 29,636,875 5,891,453 14,103,953 28,387,922 30,474,796 49,176,779 16,128,256 27,157,070 5,891,453 8,984,787 25,957,640 23,219,139 9,685,939 7,206,134 2,479,805 5,119,166 2,430,282 7,255,657

1999 comparison point, household survey (PNAD) data

Errors (simulated actual) in:

Change in Number of occupational observations (in category units) 1999 / 1998

Change in occupational category provided by runs of macro LAVs AND convergence of microsimulation model

Number of observations (in units)

Actual change in occupational category 1999 / 1998 (%)

-0.37% 9.03% -0.43% 0.63% 5.76% 2.69% 4.63% -0.55% 9.21% -0.43% 3.80% 6.04% 3.09% 1.77% 0.05% 7.12% -4.48% 2.86% 1.41%

23,346,262 28,382,451 6,050,970 15,274,013 27,037,613 30,742,070 47,960,119 16,062,275 25,846,874 6,050,970 9,596,041 24,551,054 23,409,065 9,819,564 7,283,987 2,535,577 5,677,972 2,486,559 7,333,005

-0.32% 4.42% 2.27% 8.98% 0.73% 3.59% 2.04% -0.96% 3.94% 2.27% 10.86% 0.30% 3.94% 3.17% 1.13% 9.53% 5.95% 5.24% 2.49%

(11,872) 1,254,424 (159,517) (1,170,060) 1,350,309 (267,274) 1,216,660 65,981 1,310,196 (159,517) (611,254) 1,406,586 (189,926) (133,625) (77,853) (55,772) (558,806) (56,277) (77,348)

-0.05% 4.61% -2.70% -8.35% 5.03% -0.90% 2.59% 0.41% 5.27% -2.70% -7.06% 5.75% -0.84% -1.40% -1.08% -2.41% -10.43% -2.38% -1.08%

Source: Authors’ estimates based on PNAD/IBGE 1998/1999

However, when we look at the results of the macro-micro model re-aggregated, table 5, we have the following picture. The model is under-stating the rise of unemployment (particularly in rural areas). This result mirrors the over-prediction for the growth in formal employment levels both tradable and nontradable. In terms of skills, the model captures well the decline in the employment of low skilled workers, the rise in the employment of intermediate skill workers but slightly under-estimates that of workers with high educational level.

Overall, the model captures a good part of the occupational effect of the 1999 devaluation on the occupational structure in Brazil. The shock led to a significant increase in both rural and urban unemployment and also a decline in the employment of urban workers with low skills. There is also a sizeable increase in the level of informality in both locations. These general characteristics are picked up by the model, although the rise in unemployment is substantially underestimated.

However, the predictive performance of the model is much worse for earnings. Four out of the eighteen counterfactual changes in earnings for household groups reported in Table 4(b) went in the opposite direction to the changes actually observed for those groups between 1998 and 1999. For these four errors, three were wrong by more than ten percentage points. In addition, the model is making three additional significant mistakes not in direction but in magnitude.

Page 25

rd

3 DRAFT

Table 6: Aggregate results from the macro-micro models - Earnings 1998 starting point, household survey (PNAD) data

Occupational categories

Number of observations (in units)

1999 Simulated by 3 layer Macro --> Micro models

1999 comparison point, household survey (PNAD) data

Change in average Change in average income of category Income of Income of the income of category Income of the Number of Number of the category provided by runs of provided by runs of category category observations (in units) observations (in units) macro models (average) macro models (average) (average) (LAVs) only (LAVs) only

Errors (simulated - actual) in:

Change in average income Number of Income of of category observations (in the category provided by runs units) (average) of macro models (LAVs) only

Brazil Low Intermediate High

23,420,100 27,181,832 5,916,696

278.17 484.85 1,593.53

23,334,390 29,636,875 5,891,453

267.88 488.38 1,506.04

-3.7% 0.7% -5.5%

23,346,262 28,382,451 6,050,970

280.14 475.08 1,546.16

0.7% -2.0% -3.0%

(11,872) 1,254,424 (159,517)

(12.26) 13.30 (40.12)

-4.4% 2.7% -2.5%

Low Intermediate High

16,217,373 24,866,859 5,916,696

316.17 499.07 1,593.53

16,128,256 27,157,070 5,891,453

302.01 503.61 1,506.04

-4.5% 0.9% -5.5%

16,062,275 25,846,874 6,050,970

315.61 489.88 1,546.16

-0.2% -1.8% -3.0%

65,981 1,310,196 (159,517)

(13.60) 13.73 (40.12)

-4.3% 2.8% -2.5%

Low Intermediate High

7,202,727 2,314,973 158,332

192.63 332.09 1,362.29

7,206,134 2,479,805 158,017

191.49 321.60 1,374.30

-0.6% -3.2% 0.9%

7,283,987 2,535,577 185,208

201.94 324.20 1,520.18

4.8% -2.4% 11.6%

(77,853) (55,772) (27,191)

(10.45) (2.60) (145.87)

-5.4% -0.8% -10.7%

Urban

Rural

Source: Authors’ estimates based on PNAD/IBGE 1998/1999

For Brazil as a whole, the model underestimated the changes in earnings for both low-skilled and highskilled workers, while overestimating earnings performance for the intermediate group. These errors were driven by the urban areas, which account for 80% of the population. In the rural areas, the underprediction for the low-skilled was also observed, but the pattern was reversed for intermediate and high-skilled workers, with the model overestimating losses to the former, and gains to the latter. For the six basic groupings underpinning Table 6 (the three skill levels in urban and rural areas), the model prediction was only near the target in one single case: that of high-skilled urban workers.

Overall, the actually observed changes in (nominal) earnings in Brazil from 1998 to 1999 may be summarized as follows. Mean earnings fell for all urban categories of workers,. The picture is much more mixed in rural areas. There, the only winners among low-skilled workers were those employed in the formal non-tradable and the informal sectors (and this is well predicted by the model). The main losers among intermediate skilled workers were those in the formal tradable sector (but this is not predicted by the model). And the main winners among high-skilled workers were those in the formal tradable sector.

Figures 2 and 3 present the disaggregated results for the distributions of household incomes. Figure 2 compares three different distributions with the actual 1998 distribution. In all cases, the curves plot the difference in logarithms of the mean incomes in each hundredth of the distribution. For instance the difference in logs between the actual 1999 and the actual 1998 incomes for each percentile of the distribution is given by the dotted line. The thick line plots the differences between the macro-micro counterfactual and the actual 1998. The thin solid line plots the differences, with respect to 1998, of another counterfactual distribution: that obtained from feeding the actual mean changes (rather than the LAVs) to the micro-simulation module, and running it.

Page 26

rd

3 DRAFT

When we compare the actual changes in 1999 with the simulated ones, we find that the model correctly predicted the fall in real incomes for the entire distribution, seen with respect to the line representing inflation during the period. That granted, the model substantially underestimates the changes in earnings in the lower segments of the distribution, ; overestimates the changes in earnings for the upper-middle segments, and once again underestimates changes at the very top of the distribution. The micro model on its own (i.e., operating with observed - rather than simulated - LAVS) performs better, as one would expect. Still, some of the error, particularly in the bottom third of the distribution, originates from the micro, rather than from the macro –simulation.

Figure 2 could compare the two counterfactual distributions with the actual change between 1998 and 1999, because all distributions were ranked by their own current income ranks. If one is interested in identifying winners and losers, however, one is fundamentally interested in the churning that takes place within a distribution, as it evolves over time. Figure 3 presents proportional changes (differences in logs) between the two counterfactual 1999 distributions and the actual 1998 one. This can not be done for the actual 1999 distribution, since the PNAD sample is not a panel. Regression towards the mean is clearly apparent in these simulated models. Since we know that they tend to be a regular feature in actual distributions (see Quah, 1993), this is an encouraging feature of the model predictions.

Page 27

rd

3 DRAFT

Figure 2: Effect of the Financial Crisis on the overall distribution of income Percent Changes in Household Per Capita Income 1999 compared to 1998 Ranked by final distribution of income - Nominal Terms 10.0%

8.0%

Average annual inflation rate in 1999 measured by the core consumer price index (IPCA)

6.0%

Log Difference

4.0%

2.0%

0.0%

-2.0%

-4.0%

-6.0%

Actual change 1999 / 1998 according to HHS data Change simulated by the run of Micro model only for 1999 Change simulated by the run of both Macro & Micro models for 1999

-8.0% Percentiles of the distribution

1

-10.0%

Percentiles

Source: Authors’ estimates based on PNAD/IBGE 1998/1999

Figure 3: Winners and Losers from the Financial Crisis Percent Changes in Household Per Capita Income 1999 compared to 1998 Ranked by the original distribution of income - Nominal Terms 10.0% 8.0% 6.0%

Average annual inflation rate in 1999 measured by the core consumer price index (IPCA)

2.0% 0.0% -2.0% -4.0% Change simulated by the run of both Macro & Micro models for 1999 -6.0% Change simulated by the run of the Micro model only for 1999 -8.0% -10.0%

1

Log Difference

4.0%

Percentiles

Source: Authors’ estimates based on PNAD/IBGE 1998/1999

Page 28

rd

3 DRAFT

b) The main aggregate results for Poverty and Inequality

From the counterfactual distributions whose differences were plotted in Figures 2 and 3, it is straightforward to construct standard aggregate poverty and inequality measures. Table 7 lists four inequality measures and three poverty measures for each of these three distributions: actual 1998, counterfactual 1999 and actual 1999. P(0, 1 and 2) are the standard Foster, Greer and Thorbecke (1984) poverty measures, calculated with respect to an absolute poverty line for Brazil, which takes the value of R$74.48 per person in a household located in the metropolitan region of São Paulo in 1999 (R$69.98 in 1998), and which is regionally deflated for other areas. For details of the construction of the line, see Ferreira, Lanjouw and Neri (2002). The four inequality measures are the Gini coefficient and three members of the Generalized Entropy Class of inequality measures.

9

Table 7: Poverty and Inequality: Actual and Simulated (Nominal Terms) Results for 1999

Brazil - Nominal Terms Departure: Actual data from the HHS (PNAD)

Simulated by the Macro & Micro Models (B)

Comparison: Actual data from the HHS (PNAD)



(a)

(b)

(c)

(b)/(a)

P0 Poverty Headcount

28.1%

29.8%

29.4%

1.060

P1 Poverty Gap

11.6%

12.6%

12.3%

1.082

p2

6.5%

7.1%

6.9%

1.097

e0

0.662

0.655

0.649

0.990

e1

0.715

0.695

0.695

0.971

e2

1.731

1.630

1.563

0.942

Gini

0.593

0.588

0.588

0.992

257.30

260.46

257.75

1.012 (0.9531)

Mean of HPCI in R$ / month

Source: Authors’ estimates based on PNAD/IBGE 1998/1999 Note 1 – Change in Real Per Capita Income

The actual changes in the distribution of income observed in Brazil between 1998 and 1999 can be summarized as follows. Nominal mean income remained unchanged, and there was a small reduction in measured inequality (due to relatively large income declines at the top end of the distribution). Poverty 9

See Cowell and Jenkins (1995) for a discussion. E(0) is the Theil-L coefficient, also known as the mean log deviation. E(1) is the Theil-T index. E(2) is one half of the square of the coefficient of variation.

Page 29

rd

3 DRAFT

increased as a result of the crisis: incidence and severity rose a little. Table 7 indicates that our macromicro model captures these trends rather well. This may seem surprising, since predictions of household incomes were poor. The reason is that on average, these errors compensate each other and the model predicts well both the direction and the magnitude of the change in inequality and the poverty level for 1999. The Brazilian crisis – measured by output contraction and the fall in real per capita income (-4.7%) was less severe than in Indonesia (-18.1%). The Indonesia crisis was a major depression whereas Brazil’s was a milder downturn.

4. Concluding Remarks

In this paper we proposed a macro-micro model of the Brazilian economy, intended to investigate the link between macroeconomic shocks and the distribution of employment, earnings and household incomes. The approach was to estimate a macro-econometric model on time-series data, and a micro-econometric model on household-level cross-section data. The macroeconomic model generates three sets of linkage aggregate variables – employment and unemployment levels per household group and sector, wage levels per household group and consumer price levels per household group and sector - which are then used to recalibrate parameters in the earnings and occupational models at the microeconomic level, and thus to simulate changes in the household level distribution of earnings and incomes.

This approach, adapted from Bourguignon, Robilliard and Robinson (2002) was applied to an investigation of the employment, earnings and income distribution effects of the 1998/99 devaluation of the Real. We took advantage of the benefit of hindsight, and compared the counterfactual distributions generated by our model for 1999, with the distributions actually observed in 1999.

The shock observed –together with its standard policy response of tightening both the monetary and fiscal stances to ensure price stability after a de-pegging of the currency—was expected to be rather negative. However, the devaluation in Brazil (a loss of 32% of the Real’s value) did not result - like in East Asia - in a collapse of the financial sector with a devastating effect on the credit market and eventually on 10

the real economy . Increases in poverty were correspondingly smaller in Brazil than in Indonesia, Thailand or, for that matter, Argentina. Nevertheless, real incomes fell across the entire distributions, and aggregated poverty measures rose accordingly. Since income falls were largest for the richest households, inequality fell for most commonly used measures.

The main effect on the distribution of occupations was a substantial increase in unemployment levels across sectors and areas, but predominantly in urban areas, and for more skilled workers. In urban areas, 10

The GDP real growth rate in Brazil in 1997, 1998 and 1999 was respectively 3.27% 1.32% and 0.81% still in positive territory as opposed to the dramatic changes from positive 6-8% real growth down to negative –5% to –15% in the same period in East Asian crisis-hit countries such as Thailand and Indonesia.

Page 30

rd

3 DRAFT

the informal sector registered small increases in employment, whereas the formal sector retrenched by 24%, regardless of skill-level or the tradable nature of the goods. In rural areas, the picture is mixed. There was a pronounced move from employment in the informal and formal non-tradable sector, towards the formal tradable sector (the sector that reacted to the real devaluation). In rural areas, the predictive performance of the model was good for low and intermediate skill groups, but poor for highly-skilled workers, due to a failure to predict the considerable changes in occupational status that affected that group in the formal sector. More importantly, the model failed to predict the large increase in unemployment during the crises, probably because the labor market closure and parameters, estimated from a 20-year time-series, are inadequate to predict the behavior of the labor market under stressful conditions.

The actual effects on the distribution of earnings were also muted, at least in urban areas. The changes were generally equalizing, in the sense that skilled workers had greater declines than those with fewer years of schooling. The predictive performance of the model for earnings was less satisfactory than that for occupations. In particular, the model over-predicted the earnings of workers with intermediate skills, and under-predicted those of low and high skilled workers.

This may be due to an insufficient

desegregation of the wage LAVs across occupations, or to the construction of factor remuneration – deriving strictly from our production function-- in the macroeconomic modeling. The end result in terms of the counterfactual income distribution for Brazil in 1999 was a picture where the model both under and over estimated incomes for households. But with compensating errors, this led to a relatively good prediction of the poverty and inequality levels, when compared with those actually observed in 1999.

All in all, we can not claim that this approach has delivered the ability to predict the distributional outcomes of macroeconomic shocks or policy packages with any accuracy. Nevertheless, it did get the sign of real income changes right, across the entire distribution. Its performance was also acceptable in predicting cross-sectoral changes in employment, and in predicting overall changes in poverty and inequality measures. Improvements are clearly required in disaggregated earnings predictions, and in predicting changes in unemployment levels.

It is clear that this procedure needs further testing and development, possibly in the context of a medium and longer term scenario, rather than that of a “financial crisis” which, by definition, tends to generate complex adjustments in markets (factors, etc.). Our results, however, compare well with those of other similar attempts such as Indonesia. Brazil and Indonesia have some similarities in the macro responses to the crisis as real devaluation, export boost and import contraction. Besides, the two exercises seem to show that –using both the original distribution of income —:(a) financial crises do not cause huge increases in inequality, because the original poorest percentiles loose less or be “relative winners”; and the middle and upper percentiles seem to be the hardest hit in relative terms; (b) however, these crises

Page 31

rd

3 DRAFT

do seem to send a new type of previously less poor people into extreme poverty. In summary, our results provide a reasonable decomposition of how each of the two levels of the model contribute to the explanation of the changes in poverty and income distribution. Given how important the question is, the search for better answers should continue.

5. References

Adelman and Robinson, S. (1988), Macroeconomic Adjustment and Income Distribution: Alternative Models Applied to Two Economies, Journal of Development Economics, 29(1), 23-44. Agenor P.R., Miller M., Vines D. and Weber A. (eds.) (1999), “The Asian Financial Crises: Causes, Contagion and Consequences”, Cambridge University Press, 1999. Artus P., Deleau M. and Malgrange P.(1986), Modélisation Macroéconomique, Economica, 1986. Baldacci, Mello and Inchauste (2002) “Financial Crises, Poverty, and Income Distribution” , IMF Working Paper Series 0204 Bourguignon, F., Ferreira, F.H.G.and Leite, P.G. (2002): 'Beyond Oaxaca-Blinder: Accounting for Differences in Household Income Distributions across Countries', World Bank Policy Research Working Paper 2828. Bourguignon, F., Ferreira, F.H.G.and Lustig, N. (1998) ‘The Microeconomics of Income Distribution Dynamics in East Asia and Latin America’ (mimeo), World Bank, DECRA: Washington DC. Bourguignon, F. and Pereira da Silva, L. (eds.) (2003): " Evaluating the Poverty and Distributional Impact of Economic Policies: a Toolkit of Techniques", http://www.worldbank.org/poverty/psia/tools.htm, The World Bank, Washington, DC, March 2003 Bourguignon, F., Robillard, A.-S. and Robinson, S. (2002): "Crisis and Income Distribution: A Micro-Macro Model for Indonesia", mimeo, IFPRI, Washington, DC, April 2002 Bourguignon, F.; Branson, W.H. and de Melo, J. (1992): “Adjustment and income distribution. A micromacro model for counterfactual analysis”, Journal of Development Economics 38 , pp.17-39. Bourguignon, F., Branson, W.H. and de Melo, J.(1989), Macro - economic Adjustment and Income Distribution : A Macro - Micro Simulation Model, Washington, The World Bank. Cardoso E. (2002), “Implicit and Explicit Taxation of Financial Intermediaries in Brazil: The Effect of Reserve Requirements on Bank Spreads”, mimeo Cowell, F.A. and Jenkins, S.P. (1995): "How much inequality can we explain? A methodology and an application to the USA", Economic Journal, 105, pp.421-430. Elbers, C., Lanjouw, J.O., Lanjouw, P. and Leite, P.G. (2001): "Poverty and Inequality in Brazil: New Estimates from Combined PPV-PNAD Data", World Bank, DECRG Mimeo. Fallon P. and Lucas B. (2002) “The Impact of Financial Crises on Labor Markets, Household Incomes, and Poverty: A Review of Evidence”, World Bank Research Observer ( web site) v 17 (1)21-45

Page 32

rd

3 DRAFT

Ferreira, F. H.G., Lanjouw, P. and Neri, M. (2002): "A Robust Poverty Profile for Brazil using Multiple Data Sources", Revista Brasileira de Economia, Forth Coming Ferreira, F.H.G. and Barros, R. (1999) ‘The Slippery Slope: Explaining the Increase in Extreme Poverty in Urban Brazil; 1976-96’, Revista de Econometria, 19 (2): 211-96. Fernandes, R. and Menzes-Filho, N. (2001) “Escolaridade e Demanda Relativa por Trabalho : Uma Avaliação para o Brasil nas Décadas de 80 e 90” , mimeo. Fischer, F., Kein, L. and Shinkai, Y. (1965) “Price and Output Aggregation in the Brookings Econometric Model” The Brookings Quarterly Econometric Model of the US, Duesenberry, L. (ed.). Brookings Institution, 653-679. Foster, J.E.; Greer J and Thorbecke, E. (1984): “A Class of Decomposable Poverty Indices”, Econometrica, 52, pp.761-766. Favero C.and Giavazzi F. (2002) “ Why are Brazil’s Interest Rates So High” Bocconi University - IGIER and IGIER (Bocconi University) SSRN Electronic Paper Collection Glick R., Moreno R. and Spiegel, M. (2001), “Financial Crisis in Emerging Markets”, Cambridge University Press, 2001. Lokshin M. and Ravallion M. (2000), “Welfare Impacts of the 1998 Financial Crisis in Russia and the Response of the Public Safety Net”, The Economics of Transition, V 8 (2) 2000, pp.269-295. Kakwani, N. (1998) “Impact of Economic Crisis on Employment, Unemployment and Real Income”, National Economic and Social Development Board. Krugman, P. (1979), ‘A Model of Balance of Payments Crises’, Journal of Money, Credit and Banking 11: 311-25. Masson, P (1999). Contagion: Macroeconomic Models with Multiple Equilibria," Journal of International Money and Finance 18:587-602. Obstfeld M. (1994), “Risk-Taking, Global Diversification and Growth,” American Economic Review, Vol. 84, No.4. Picchetti, P. Pereira da Silva L. and Castro A. Samy de (2003), A macroeconometric financial model for Brazil estimated with the PNAD, (forthcoming), mimeo, The World Bank, 2003. Quah D. (1993) Empirical Cross-Section Dynamics in Economic Growth, European Economic Review, 37, n. 2-3, April. Stern N., (1989), The Economics of Development: A Survey, Economic Journal v99, n397 (September 1989): 597-685

Page 33

rd

3 DRAFT

Annex 1 Macroeconomic Model Regression Results

Table 8 - Factor Demands Rural Non-Tradable Formal Sector RNF_NL Coefficient

RNF_NI

Std. Error

Coefficient

Std. Error

RNF_NH Coefficient

Std. Error

RNF_K Coefficient

Std. Error

C

-1.06256

3.434591

6.832893

4.236494

12.66641

2.238485

-8.25891

0.240471

RNF_Y

0.055036

0.021078

1.332698

0.507967

0.108299

0.040333

1.009547

0.005871

RNF_NL(-1)

0.886186

0.269272

-0.03185

0.042257

AGG_PLQ-AGG_PLH

-0.06885

0.215309

RNF_PY-AGG_PLA

-0.00488

0.094192

-0.05684

0.030686

-0.23011

0.190907 0.060967

0.034285

R-squared

0.999609

RNF_PY-AGG_PLL

-0.04525

0.087043

AGG_PLU-AGG_PLL

-0.07052

0.090867

UCK-AGG_PLA

-0.02549

0.03018

RNF_Y(-1)

-0.162368

0.051529

AGG_PLI

-1.169801

0.488612

UCK-AGG_PLI

0.056814

0.056092

RNF_PY-AGG_PLI

-1.416644

0.538317

UCK(-1)-AGG_PLA RNF_NIQ(-1)

0.171037

0.29797

RNF_NH(-1) RNF_PY-UCK(-1) R-squared

0.747075

R-squared

0.85895

R-squared

0.970424

Table 8 (b) Rural Tradable Formal Sector RTF_NL Variable

RTF_NI

Coefficient

Std. Error

Coefficient

RTF_Y

0.045992

0.053488

RTF_Y(-1)

-0.0315

0.034206

RTF_NL(-1)

1.006323

0.013598

UCK

-0.01764

0.03224

C

UCK-AGG_PLA

RTF_NH

Std. Error

Coefficient

13.08778

3.70928

0.988568

0.531133

-0.14834

0.066269

-0.02398

0.060461

AGG_PLQ

-0.80136

0.534637

RTF_PY-AGG_PLA

-0.80249

0.654628

RTF_NIQ(-1)

-0.14794

0.267973

Coefficient

11.26326

2.114545

-6.140105

0.327099

-0.23182

0.275791

1.009916

0.008133

-0.08243

0.03746

0.005776

0.035023

0.339177

0.333986

0.049467

0.046341

R-squared

0.999251

RTF_NH(-1)

-0.05141

0.187434

AGG_PLA

0.338375

0.278635

AGG_PLH-AGG_PLQ

-0.27396

0.278283

RTF_PY-UCK(-1) R-squared

0.464169

R-squared

0.838774

RTF_K

Std. Error

R-squared

0.81857

Std. Error

Page 34

rd

3 DRAFT

Rural Non-Tradable Informal Sector RNI_NL

RNI_NI

RNI_NH

Coefficient

Std. Error

Coefficient

Std. Error

Coefficient

Std. Error

C

15.86946

4.513833

8.542033

3.020678

5.46586

2.171236

RNI_Y

0.014543

0.00756

0.305519

0.303825

0.006325

0.010475

RNI_PY-AGG_PLA

-0.09079

0.105993

-0.10448

0.11051

AGG_PLU-AGG_PLL

-0.06187

0.054378

-0.07189

0.059728

RNI_NL(-1)

-0.00419

0.282491

0.50384

0.217196

R-squared

0.838781

RNI_NIQ(-1)

0.114874

0.286017

RNI_PY-AGG_PLQ

0.187363

0.126773

AGG_PLQ

-0.24797

0.301685

RNI_NH(-1) R-squared

0.27958

R-squared

0.907394

Urban Non-Tradable Formal Sector UNF_NL

UNF_NI

UNF_NH

UNF_K

Coefficient

Std. Error

Coefficient

Std. Error

Coefficient

Std. Error

Coefficient

C

17.05956

4.701364

14.31808

3.163831

14.52338

0.206298

-8.17699

0.233099

UNF_Y

0.041757

0.012365

0.050514

0.037048

0.026193

0.002318

1.008911

0.005141

UCK(-1)-AGG_PLA

-0.03985

0.013702

AGG_PLU(-1)-AGG_PLL(-1)

0.011642

0.02506 0.072182

0.031666

R-squared

0.999657

UNF_NL(-1)

-0.12619

0.316331

0.150061

0.194909

UNF_Y(-1)

-0.04568

0.012845

-0.02742

0.034728

UNF_PY-AGG_PLA

-0.04203

0.028069

AGG_PLU-AGG_PLL

-0.0014

0.015684

UCK-AGG_PLA

0.004169

0.010741

-0.08582

0.031842

UNF_PY-UCK(-1) R-squared

0.783053

R-squared

0.841848

R-squared

0.933526

Urban Tradable Formal Sector UTF_NL Coefficient

Std. Error

UTF_Y

0.017923

0.016689

UTF_NL(-1)

0.998179

0.002179

UTF_Y(-1)

-0.02145

0.016437

UTF_PY-AGG_PLA

UTF_NI

UTF_NH

Coefficient

Std. Error

Coefficient

Std. Error

19.19142

5.103581

2.412677

0.842959

0.005733

0.344419

0.0406

0.022511

0.02003

0.059576

0.012982

0.439321 0.345946

AGG_PLQ

0.00808

UCK-AGG_PLA

-0.07466

0.069322

UTF_NIQ(-1)

-0.19215

0.310624

UTF_NH(-1)

0.819584

0.060315

AGG_PLL-AGG_PLI

-0.01921

0.010109

UTF_PY-UCK(-1) R-squared

0.805478

R-squared

0.606582

R-squared

0.962233

UTF_K Coefficient

Std. Error

1.050455

0.035276

0.878386

0.05332

R-squared

0.982553

Page 35

Std. Error

rd

3 DRAFT

Urban Non-Tradable Informal Sector UNI_NL

C

UNI_NI

Coefficient

Std. Error

15.50333

4.034316

UNI_NH

Coefficient

Std. Error

Coefficient

Std. Error

UNI_Y

-0.00124

0.019959

0.055097

0.122716

0.0668

0.018069

UNI_Y(-1)

0.006752

0.019642

-0.05062

0.120666

0.0564

0.017654

UNI_NL(-1)

0.007264

0.258672

UNI_NIQ(-1)

0.502054

0.223026

UNI_NIQ(-2)

0.505659

0.222739 1.004875

0.002695

R-squared

0.948452

UNI_NH(-1) R-squared

0.257916

R-squared

0.686649

Table 9 - Earnings Rural Non-Tradable Formal Sector RNF_WH

RNF_WI

RNF_WL

Coefficient

Std. Error

Coefficient

Std. Error

Coefficient

C

0.068528

7.60313

4.279924

5.762198

-14.2862

4.585718

RNF_Y(-1)

1.313402

0.693592

0.748655

0.553947

1.239955

0.621363

-1.1912

0.468283

0.126195

RNF_WI(-1)

0.814944

0.477382

RNF_WI(-2)

0.591887

0.525366

AGG_URQ_R

-1.83452

1.327938

RNF_Y(-2)

-1.28904

0.57603

RNF_WH(-1)

0.41543

0.574788

RNF_WH(-2)

0.51841

0.606784

AGG_URH_R

-0.45281

0.449789

Std. Error

RNF_WL(-1)

0.280542

0.293677

RNF_WL(-2)

-0.44556

0.301249

AGG_URL_R

-2.46207

2.967608

R-squared

0.969356

R-squared

0.990948

R-squared

0.99163

Rural Tradable Formal Sector RTF_WH

RTF_WI

RTF_WL

Coefficient

Std. Error

Coefficient

Std. Error

Coefficient

Std. Error

-0.830759

5.957318

3.891722

6.677299

-28.8765

9.550492

RTF_Y(-1)

1.321735

0.580787

1.124469

0.524451

3.545119

1.303221

RTF_Y(-2)

-1.123766

0.561407

-1.598306

0.567492

-0.07827

RTF_WH(-1)

0.40252

0.532592

0.661469

0.465303

RTF_WH(-1)-RTF_WI(-1)

-0.154209

0.993299

RTF_WH(-2)

0.366384

0.54335 -0.062863

0.112787

0.757109

0.556819

-1.532952

1.121033

C

RTF_WI(-1)-RTF_WL(-1) RTF_WI(-2) AGG_URH_R

-0.146147

0.376121

AGG_URQ_R RTF_WL(-1)

-1.88601

RTF_WL(-1)-RTF_WI(-1)

2.201093

1.130696

RTF_WL(-2)

-0.36427

0.247371

AGG_URL_R

-0.00371

0.00573

R-squared

0.977009

R-squared

0.993878

R-squared

0.994595

1.163861

Page 36

rd

3 DRAFT

Rural Non-Tradable Informal Sector RNI_WH Coefficient

RNI_WI

Std. Error

RNI_WL

Coefficient

Std. Error

Coefficient

Std. Error 3.064699

C

-8.49821

3.32669

-5.90826

3.583113

-5.25426

RNI_Y

0.99613

0.066731

0.923553

0.085694

0.89482

0.078441

RNI_Y(-1)

0.105686

0.283655

-0.42713

0.297777

-0.59319

0.280074

0.139612

0.300796

0.2258

0.284835

RNI_WL(-1)

0.751492

0.332356

RNI_WL(-2)

-0.31973

0.331146

-0.41338

0.435849

R-squared

0.999104

RNI_Y(-2)

-0.05697

0.202923

RNI_WH(-1)

-0.25626

0.29082

RNI_WH(-2)

0.156947

0.212332

RNI_WI(-1)

0.498711

0.35309

RNI_WI(-2)

-0.18448

0.355356

AGG_URH_R

-0.2453

0.125259

AGG_URQ_R

-0.13417

0.538503

AGG_URL_R R-squared

0.999162

R-squared

0.998954

Urban Non-Tradable Formal Sector UNF_WH

UNF_WI

UNF_WL

Coefficient

Std. Error

Coefficient

Std. Error

Coefficient

C

-9.520144

0.460872

-11.1356

0.279559

-25.1968

2.026513

UNF_Y

2.018166

0.117349

2.092114

0.100913

1.064131

0.052107

-1.13434

0.099104 -2.79428

1.855842

UNF_Y(-2)

-1.041915

0.116498

AGG_URH_U_T-AGG_URH_U_T(-1)

-25.27212

15.87965

UNF_Y_WBILL_HIGH

0.27885

0.360434

AGG_URH_U

-0.63007

0.82419

UNF_Y_WBILL_INT(-1)

-0.65493

0.485005

AGG_URQ_U

-0.20083

0.639335

Std. Error

UNF_Y_AGG_W_BILL_LOW

-2.53156

0.57875

AGG_URL_U_T-AGG_URL_U_T(-1)

5.746846

31.97453

R-squared

0.988295

R-squared

0.998398

R-squared

0.998528

Page 37

rd

3 DRAFT

Urban Tradable Formal Sector UTF_WH Coefficient

UTF_WI

Std. Error

Coefficient

UTF_WL

Std. Error

Coefficient

Std. Error

C

-2.144822

2.039826

-5.13273

2.248914

-24.1029

14.20426

UTF_Y

1.005595

0.018227

0.960437

0.020515

1.551016

0.730521

UTF_Y(-1)

-0.837418

0.159374

-0.62537

0.166985

-1.88551

1.135379

UTF_Y(-2)

-0.048694

0.033524

-0.08015

0.039561

1.542614

0.655927

UTF_WH(-1)

0.826591

0.148641

UTF_WI(-1)

0.693009

0.151761

UTF_WI(-2)

0.044305

0.029834

UTF_Y_AGG_W_BILL_INT

-0.96853

0.028828

UTF_Y_AGG_W_BILL_INT(-1)

0.596124

0.151385

UTF_Y_AGG_W_BILL_LOW

-2.35034

0.805308

UTF_Y_AGG_W_BILL_LOW(-1)

0.681465

0.991209

UTF_WL(-1)

-0.37017

0.17517

-3.38639

2.218798

R-squared

0.990164

UTF_WH(-2)

0.052138

0.026955

UTF_Y_AGG_W_BILL_HIGH

-0.975886

0.030044

UTF_Y_AGG_W_BILL_HIGH(-1)

0.813349

0.143145

UTF_WL(-2) AGG_URH_U

-0.039806

0.092364

AGG_URQ_U

-0.04831

0.099537

AGG_URL_U R-squared

0.999989

R-squared

0.999985

Urban Non-Tradable Informal Sector UNI_WH

UNI_WI

UNI_WL

Coefficient

Std. Error

Coefficient

Std. Error

Coefficient

C

-7.64119

2.541425

-10.69719

3.639443

-9.88829

3.29472

UNI_Y

0.952634

0.051748

0.952917

0.024673

1.170188

0.513993

UNI_Y(-1)

-0.07563

0.281221

0.358382

0.283942

-1.24522

0.903057

UNI_Y(-2)

0.289429

0.304266

0.05741

0.269796

1.157793

0.621382

UNI_WH(-1)

0.112432

0.283085

-0.274344

0.295688

0.587908

0.257598

UNI_WH(-2)

-0.26724

0.300263 -0.086134

0.257937 -0.46742

0.258442

-5.28433

3.254422

R-squared

0.974622

UNI_WI(-2) UNI_WL(-2) AGG_URH_U

-0.27822

0.296013

AGG_URQ_U

-0.6599

0.177751

AGG_URL_U R-squared

Std. Error

0.99966

R-squared

0.999904

Page 38

rd

3 DRAFT

Annex 2 Main Equations of the Micro-Simulation Model and Results of Econometric Estimates Table 10(a)

Page 39

rd

3 DRAFT

Table 10(b)

Page 40

rd

3 DRAFT

Table 11

Page 41

rd

3 DRAFT

Table 12(a)

Page 42

rd

3 DRAFT

Table 12(b)

Page 43

rd

3 DRAFT

Annex 3

The Household-Level Data and the Micro-econometric model

The micro-econometric module of the model is estimated on household-level data from the Pesquisa Nacional por Amostra de Domicílios (PNAD), which is fielded annually (except in Census years) by the Instituto Brasileiro de Geografia e Estatística (IBGE). We use the unit-record data for the 1998 survey which had a sample size of 88,356 households (and 333,074 individuals) - and the 1999 survey - which had a sample size of 91,523 households (and 340,986 individuals).

The main variables we used from these surveys were: (i) earnings, defined as total monthly labour earnings from the worker's main occupation (in the reference weeks of 20-26 September, 1998 and 19-25 September, 1999); (ii) occupation of the worker in that reference week and in that main occupation; and (iii) total household income per capita, aggregated over all earnings of all individuals in the household and all reported non-labor incomes, as indicated in Equation (4) below. Workers are all those who reported working at least one hour in the reference week.

11

The PNAD is the staple household survey for analysis of the Brazilian income distribution. It is representative of both urban and rural areas in all five regions of Brazil, except in the North where, for cost-related reasons, rural areas are only surveyed in the State of Tocantins. Income data from the PNAD does, however, suffer from considerable measurement error. The PNAD questionnaires, although much improved during the 1990s, still contain insufficient detail on capital incomes, production for own consumption and incomes in kind. As a result, there is some evidence that some incomes are underreported, particularly in rural areas. It is likely that this problem is most severe at both tails of the distribution.

12

In what follows, we chose to include rural incomes for the sake of completeness of

coverage. We urge the reader to beware, however, that the income levels reported are likely to reflect substantial measurement error.

11

Workers were only included if they reported to be active in the reference week, and were aged 15 or above. Households were purged of domestic servants and lodgers (pensionistas), and of their incomes. 12 See Ferreira, Lanjouw and Neri (2002) and Elbers, Lanjouw, Lanjouw and Leite (2001) for an assessment of these measurement problems, based on comparisons between the PNAD and an alternative Brazilian household survey, the Pesquisa de Padrões de Vida (PPV).

Page 44

rd

3 DRAFT

Annex 4 Derivation of Estimable Factor Demand Functions •

Demand for Low-Skill Labor

∂y ∂Ll

=

∂y ∂La ∂LU ∂La ∂LU ∂Ll

=

PLl PLl PLU PLa = P PLU PLa P

Log-linearization:

ln Ll = ln y + σ ka (ln P − ln PLa ) +

σ QU (ln PLa − ln PLU ) + σ li (ln PLU − ln PLi ) + σ

β − ρσ ka (1 − λ )σ ka (1 − δ ) QU ξ σ li 144444 42444444 3 cst l



Demand for High-Skill Labor

∂y ∂y ∂La ∂LQ = ∂Lh ∂La ∂LQ ∂Lh

=

PLh PLh PLU PLa = P PLU PLa P

Log-linearization:

ln Lh = ln y + σ ka (ln P − ln PLa ) +

σ QU (ln PLa − ln PLQ ) + σ ih (ln PLQ − ln PLh ) + σ

β − ρσ ka (1 − λ )σ ka δ QU (1 − φ )σ ih 144444 42444444 3 cst h

• Note:

Demand for Intermediate-Skill Labor

PLiU = PLiQ = PLi

Unqualified Jobs

∂y ∂LiU

=

∂y ∂La ∂LU ∂La ∂LU ∂Li

=

PLi PLi PLU PLa = P PLU PLa P

Log-linearization:

ln LiU = ln y + σ ka (ln P − ln PLa ) +

σ QU (ln PLa − ln PLU ) + σ ih (ln PLU − ln PLi ) + β − ρσ ka (1 − ξ )σ li δ

σ QU

σ

(1 − δ ) QU 14444442444444 3 cst iU

Qualified Jobs

Page 45

rd

3 DRAFT

∂y ∂LiQ

=

∂y ∂La ∂LQ ∂La ∂LQ ∂Li

=

PLi PLi PLQ PLa = P PLQ PLa P

Log-linearization:

ln LiQ = ln y + σ ka (ln P − ln PLa ) +

σ QU (ln PLa − ln PLQ ) + σ ih (ln PLQ − ln PLi ) + β − ρσ ka φ σ ih δ

σ QU

144244 3 cst iQ



∂y ∂K

Demand for Capital

=

PK P

Log-linearization:

ln K = ln y + σ ka (ln P − ln PK ) + ln( β − ρσ ka λ σ ka )

Page 46