Chapter

II. A Review of Macro-Micro Approaches for Evaluating the Distributional Impacts of Macroeconomic Reforms

Author(s):
Robert Gillingham
Published Date:
March 2008
Share
  • ShareShare
Show Summary Details

A. Introduction

The poverty and social implications of macroeconomic and structural reform policies are increasingly being recognized in IMF-supported programs and IMF policy advice. In 1999, the IMF replaced the Enhanced Structural Adjustment Facility, its assistance program for supporting low-income countries, with the Poverty Reduction and Growth Facility (PRGF), which explicitly gives poverty alleviation more prominence in its operations. In addition to its focus on promoting macroeconomic stability and growth, the PRGF program focuses on the relationship between macroeconomic policies and their poverty implications.

A new set of analytical tools and techniques is required, however, to evaluate the poverty and distributional effects of macroeconomic policies. The PRGF programs continue to rely on the IMF financial programming model (FPM) as a key analytical framework for a consistent set of macroeconomic policies. However, the FPM does not account for the diverse channels through which macroeconomic policies affect poverty. Therefore, it is necessary to complement the existing analytical framework with tools and techniques to address macro-poverty links.

Modeling the linkages among macroeconomic policies, growth, and poverty is a challenging task. The macro-poverty links are diverse and country specific. A number of economic mechanisms and economic agents are involved, and their roles in markets vary from one country to another. In addition, poverty is multidimensional. It is critical to recognize the characteristics of the poor—their consumption patterns, their sources of income, and the overall environment in which they operate. Furthermore, macroeconomic analysis uses different data sets from those used for traditional micro analysis of poverty.

Nevertheless, a variety of analytical approaches that address the macro-poverty links exist. The approaches vary across many domains. Some of the approaches focus on the consumption effects associated with a change in policy through their effects on prices, whereas other approaches focus on factor markets and how policies affect wages and employment. The methods also range from static to dynamic, with a partial equilibrium or a general equilibrium focus, and with or without behavioral responses.

The focus of this review paper is on a subset of these analytical approaches, those that are referred to as macro-micro techniques for evaluating poverty and distributional impacts of macroeconomic policies. These are techniques that follow an economy-wide framework of analysis and have been used in practice to evaluate the poverty and distributional impacts of economic policies. From a modeling perspective these techniques attempt to bridge the gap between two different sets of analytical approaches: those used to analyze macroeconomic policies and those used to analyze poverty and distribution issues. The objective of the review is to provide a brief and accessible guide to these techniques, highlighting how the macro-poverty links are modeled and the underlying assumptions; the trade-offs involved in terms of data, time, and resource requirements; and the typical policy questions addressed by these techniques.

A selected number of these techniques are reviewed here, including

  • Social accounting matrix (SAM) multiplier models;
  • SAM-based computable general equilibrium (CGE) models;
  • The 123PRSP (Poverty Reduction Strategy Paper) model;
  • The IMMPA (integrated macroeconomic model for poverty analysis); and
  • CGE-microsimulation-based models.

The first and second rely on the SAM database as an economy-wide framework of analysis, whereas the remaining three are extensions of the second approach. Typically the extensions involve additional data and modeling features to expand the analysis of poverty and distribution. The choice of which one of these techniques to use depends on the policy questions each technique can address and its requirements in terms of data, time, and skill. The chapter also considers the relative resource requirements associated with each technique and discusses some of the practical trade-offs involved.

B. Macro-Micro Analytical Approaches

WHAT IS A SAM?

A SAM is a comprehensive, complete, flexible, and consistent system for organizing the social and national accounts of a nation over a period of time, usually a year (Decaluwé and others, 1999). It is comprehensive in the sense that it covers all transactions in an economy, both within the domestic economy and between the domestic economy and the rest of the world. It is complete in the sense that it accounts for all incomes and outlays in the economy—that is, all payments, receipts, and transfers. It is flexible because it can be applied at the national level or can focus on a particular region, commodity, or institution. It can be aggregated or disaggregated to any level according to the requirements of the issue and availability of data. It is consistent in the sense that for each account, total receipts (income) and expenditure of each actor must balance (i.e., equal row and column totals).

A SAM is one approach to presenting the national accounts data of a country. It captures the circular flow of incomes from commodity markets through factor payments to households and back to product markets through spending on final goods (Figure 2.1). A typical SAM includes accounts for production (activities), commodities, factors of production, institutions (households, enterprises, and government), and the rest of the world, all of which receive income and demand goods (see Table 2.1).

Figure 2.1.Circular Flow of Income as Portrayed in a Social Accounting Matrix

Table 2.1.Structure of a SAM
Expenditure
ReceiptsActivity

-1-
Commodity

-2-
Factors

-3-
Institutions

-4-
Capital account

-5-
World

-6-
Totals

-7-
DomesticActivity income
1 - ActivitysalesYi
t12
IntermediateTransactionsFinalCommodity
2 - CommodityinputscostsdemandInvestmentExportsdemand

y2
t21t22t23t24t25t26
3 - FactorsValue added (wages/rentals)Factor income
t31y3
FactorInstitution
4 - InstitutionsTaxesTariffsincomeTransfersTransfersincome
t41t42t43t44y4
Capital
5 - CapitalSavingsinflowTotal savings
accountt56t54y5
Foreign exchange
6 - WorldImportsoutflow
t62y6
TotalCrossForeign
Total costsTotalfactordomesticTotalexchange
7 - Totalsabsorptionincomeincomeinvestmentinflow
y1y2y3y4y5y6
Source: Lofgren and others (2002)Note: GDP at factor cost - t31 and GDP at market prices t31- t41t24 (t24t62)
Source: Lofgren and others (2002)Note: GDP at factor cost - t31 and GDP at market prices t31- t41t24 (t24t62)

According to its structure, each account is represented by a row and a column account, where as a matter of convention, incomes or receipts are shown along a row whereas expenditures or outlays are shown down a column. Activities pay for intermediate inputs (T21) factors of production (T31), and taxes (T41) whereas they receive payments from sales of commodities (T12) which at the same time are the commodity account payments for obtaining goods from activities (producers). Other commodity account payments include imports (T62) and tariffs (T42) to the government—an institution. At the same time, commodities receipts come from intermediate sales (T21), final demanders (households, government, and investment; T24 and T25), and the rest of the world (exports; T26). The capital account collects savings by institutions (T54) plus capital inflows (foreign savings; T56) from the rest of the world, which are used to finance domestic investment (T25) 1 Gross domestic product (GDP) at factor cost (payments by activities to factors of production) equals GDP at market prices (consumption plus investment plus government demand plus exports minus imports) less indirect taxes (T41)

Compared with input-output (IO) tables, a SAM includes additional information that tracks the circular flow of income from activities to factors and to institutions. A SAM keeps track of returns to factors by sector (T31)—the functional distribution of income—and its distribution among households (T43)—the size distribution of income. Table 2.2 provides a numerical illustration of a SAM representation of the functional and size distributions of income.

Typically, factors of production include labor and capital and could be disaggregated according to certain characteristics such as skilled or unskilled labor. Similarly, households may be classified according to their income sources or other socioeconomic characteristics such as residence in urban or rural areas. In the first panel of Table 2.2, the unskilled labor row (receipts) indicates that total unskilled labor income equals US$115 million (the sum of activity payments to the factor). The second panel of the table shows the percentage shares of this income by source (in this case its income is almost equally distributed among the three sectors). At the same time the third panel shows how factor income is distributed among different household groups. In this case, 60 percent of unskilled labor income accrues to rural households.

Table 2.2.Flow of Factors and Household Incomes in a SAM(US$ millions)
Distribution of Factor and Household Income in a SAM
Production Activities
AgricultureManufacturingServiceUnskilledSkilledCapitalTotal
Unskilled403738115
Skilled254362130
Capital354855138
Rural705542167
Urban457596216
Total100128155115130138
Share of Factor Income by Source
Production Activities
AgricultureManufacturingServiceTotal
Unskilled34.7832 1733.04100.00
Skilled19.2333.0847.69100.00
Capital25.3634.7839.86100.00
Distribution of Factor Income by Household
UnskilledSkilledCapital
Rural60.8742.3130.43
Urban39.1357.6969.57
Total100.00100.00100.00
Source: Illustrative SAM produced by author.Note: Labor is disaggregated into unskilled and skilled; representative households are split into rural and urban.
Source: Illustrative SAM produced by author.Note: Labor is disaggregated into unskilled and skilled; representative households are split into rural and urban.

The information in Table 2.2 shows how a SAM can be used to trace the distributional effects of a macro policy from its impact on sectoral production and in turn on factor incomes, and eventually onto households’ incomes. For example, if trade liberalization leads to higher demand for unskilled labor, and hence to an increase in its (relative) wage, then it follows that rural households will benefit more relative to urban households. In addition, if poverty is relatively high among rural households, then trade liberalization should lead to a decrease in poverty. In brief, a sufficiently disaggregated SAM reveals a lot of structural detail and interlinkages for an economy and can serve as a bridge between a macroeconomic framework, markets, and institutions (Round, 2003).

SAM Multiplier Analysis

A SAM by itself is not a model; it is a comprehensive database. However, a SAM can serve as a basis for simulating a variety of multisector economy-wide models. The SAM multiplier analysis is a simple SAM-based model. A SAM structure of row and column accounts can be translated into a set of linear equations in which each actor represented in the SAM accounts behaves according to fixed column coefficients (Robinson and Lofgren, 2005). For example, production technology follows a fixed Leontief input-output coefficients structure, returns to factors (value added) are distributed among households according to fixed shares, and households’ savings and spending decisions are fixed shares of total income. The multiplier follows from the Keynesian demand multipliers, because SAM-based multiplier models are driven by exogenous demand and yield an effect larger than the initial exogenous change in demand. The model thus implicitly assumes the existence of unlimited unemployed factor supplies, which severely restricts the usefulness of the results for policy purposes. However, the analysis can usefully identify important channels through which policies may affect household welfare.

Because equilibrium is defined as equal row-column sums in all SAM accounts and with a square matrix, it follows that if n - 1 row-column sums are equal then the nth account must also be equal. The SAM multiplier model becomes overdetermined unless a subset of accounts is exogenous. Assuming a partitioned SAM, with a set of endogenous and exogenous accounts, and denoting the vector of totals of n endogenous accounts by yn and the vector of totals of the exogenous accounts by x, then the incomes of the endogenous accounts (row totals) can be written as:

where An is an n x n square matrix of average propensities to consume. Equation (2.1) can also be written as:

provided (I - An)-1 exists.

The inverted matrix Mn in Equation (2.2) is the SAM multiplier matrix, which relates endogenous income yn to exogenous injections x (Stone, 1985). The matrix measures the response of the economy to an exogenous change in final demand where the notion of a multiplier is measured by the difference between the initial change in final demand and the total effect of this change on the economy. For example, if the value of the SAM multiplier is 1.5, then a $10 million increase in final demand x (say because of an increase in exports—an exogenous account) will generate a $15 million increase in output. Of this increase in output, $10 million is a direct effect of the change in final demand for the export sector, and the remaining $5 million is split between indirect effects—the change in other industries’ output to satisfy the inputs needed for the expanding export sector—and induced effects resulting from increased/decreased expenditures by households from the income gained/lost from the direct and indirect effects of the change in final demand for exports.

SAM multipliers are larger in value than are the standard IO model multipliers. This fact follows from the structure of a SAM, which embodies the IO table accounts in addition to accounts describing the distribution of factor incomes and the distribution of disposable income. The difference is the induced effects generated from specifying the factor accounts and household groups as endogenous in the system. Thus a SAM multiplier allows for induced feedback effects as respending of incomes occurs.

In the SAM multiplier analysis, it is important to decide which accounts to make exogenous. Sadoulet and de Janvry (1995) note that the multipliers are affected by the choice of exogenous accounts and that a choice of an account should be justified on the basis of theory and empirical realism. One criterion is to choose an account for which expenditures are set independently of income. For example, household expenditures depend on household disposable income; thus, choosing a household account as one of the exogenous accounts might not be useful from a policy perspective. Depending on the purpose of the study and the policy questions, the choice for exogenous accounts typically includes one or a combination of the following accounts: (1) the government, (2) the rest of the world, and (3) the capital account.

With a functional and a size distribution of income as specified in the SAM, it is possible to use a SAM-based multiplier analysis to assess the distributional and poverty impacts of a number of policy issues, such as changes in government expenditures or transfers, investment, exports, and remittances. For example, Bautista, Robinson, and El-Said (2001) analyze the growth and income distribution effects of increased investment allocation by economic sector in Indonesia. The results indicate that increased investment in agriculture, compared with food processing and light manufacturing industries, resulted in a more egalitarian growth of the economy. Other examples of SAM multiplier analysis include applications to Mexico (Adelman, Taylor, and Vogel, 1988), India (Hazell, Ramasamy, and Rajagopalan, 1991), and Ghana (Powell and Round, 2000).

It is important to note that SAM multiplier models capture the effects of exogenous injections into the economy on the levels of incomes of the endogenous accounts following a fixed-coefficient (Leontief) technology. In addition, as indicated above, the SAM multiplier analysis assumes fixed prices, implying that production activities’ response to the exogenous shock falls entirely on changes in factor supply. In other words, the results are generated under no supply-side constraints—the model is demand driven with an infinite supply of unemployed factors of productions. Income effects arise from increased employment of previously unemployed factors and not from changing factor and product prices. A key contribution of computable general equilibrium models, which are considered in the following section, is the introduction of endogenous flexible prices (for commodities and factors) and a model of behavioral responses by optimizing economic agents reallocating resources between competing uses.

C. Computable General Equilibrium Models

Following the structure of a SAM, a CGE model is a system of simultaneous equations that provides a complete and consistent picture of the circular flow of income in an economy, and at the same time accounts for all market-based interactions among economic agents (Robinson, 1989). In the model, producer and household behavior is based on standard microeconomic theory whereby households maximize utility subject to budget constraints, and producers seek to maximize profits given existing technology and factor and product prices. The model satisfies equilibrium conditions, known as macro closure rules in the CGE literature. These are constraints that have to be satisfied by the general equilibrium model, but are not considered in the optimizing decision of any micro agent (Robinson, 1989).2

Using a CGE model for simulating the impact of economic policies involves a SAM calibration process and carrying out at least one policy simulation. CGE model calibration is a step that ensures that the initial model solution, a solution that is typically called the base solution, reproduces the input SAM. In other words, the SAM or the base solution is taken as a benchmark equilibrium—the state of the economy as portrayed in the SAM structure. The next step is to change a key policy parameter in the model to account for a policy shock, then re-solve the model and compare the new solution, which represents a new state of equilibrium, with the base solution to assess the impact of the introduced policy.

Early CGE models that focus on macroeconomic issues of growth, income distribution, and poverty were applied by Adelman and Robinson (1978) to Korea and by Lysy and Taylor (1980) to Brazil. These models investigated the impact of various macroeconomic policies and programs intended to improve the relative and absolute incomes of the poor. Much of this and later work emphasized issues of growth and income distribution, whereas less work was done on poverty.

CGE models address a number of weaknesses in SAM-based multiplier models. First, prices are endogenous. Both prices and quantities adjust in response to an exogenous shock to clear markets. Second, the model specifies factor supply constraints that address the problematic supply assumptions underlying multiplier models. Third, a CGE model is capable of using alternative functional forms to model economic agents’ behavioral responses, including fixed-share coefficients as in the SAM multiplier analysis.

However, like SAM multiplier models, CGE model applications classify households into a few broad representative categories—by income and/or by locality. That is, an average representative household (RH) is representative of all the households in its group. Within-group variation is suppressed. For example, when the income of a rural household group grows by x percent, that signifies that the mean of the distribution of income for rural households as a group has grown by x percent. Consequently, a typical CGE model determines (endogenously) only the variation in the between-group distribution of income. This is a limitation of the approach because within-group heterogeneity is a factor that cannot be ignored when conducting an analysis of inequality or poverty.3

Figure 2.2 is another way to illustrate the link between household and factor incomes in a SAM, as presented in Table 2.2 In the context of a typical CGE modeling framework, the figure illustrates the mechanism through which a macroeconomic policy change trickles down and affects the household at the micro level. At the top level are the production activities, usually disaggregated by sector, that demand factors of production. If profit-maximizing behavior is assumed, the changes in relative prices following a policy or an exogenous shock influence the demand for factors and, ultimately, factor incomes and the functional distribution of income (the second level in Figure 2.2). The third level of Figure 2.2 traces the macro shock to an intermediate, or meso, level where the structure of household ownership of the factors of production (derived from the SAM as fixed shares) is transferred to the micro or RH level. However, according to the RH group mean income (rural and urban groups in Figure 2.2), only changes among group distributions are endogenously determined. Addressing within-group variation requires supplementing the modeling framework with additional information from either survey data or analytic distributions (the lower level in Figure 2.2).

Figure 2.2.Functional Distribution and Size Distribution of Income (CGE Model)

Approaches for Modeling Within-Group Variation in CGE Models

Two common approaches are used to address within-group household income distributions within a CGE framework. One approach assumes that the within-group distribution of household incomes follows a parametric probability distribution. For example, Adelman and Robinson (1978) use a lognormal distribution, and de Janvry, Sadoulet, and Fargeix (1991) use both lognormal and Pareto distributions. More recently Decaluwé and others (1999) used a beta distribution and went further to endogenously determine the monetary value of the poverty line.

Another approach suggested by Coady and Harris (2004) in their analysis of a targeted cash transfer program in Mexico maps each household income (or consumption) observation in a household survey to a corresponding RH group in the CGE model. Doing so attaches a within-group distribution of income to each RH group. Then the within-group distributions are updated by scaling each mapped observation in a particular representative group by the absolute change in the group’s mean income (or consumption) plus a household-specific real income change owing to price changes. Under this approach, computed poverty and income distribution indices account for between-group and within-group variations in the distribution of income.

The two approaches are top-down, tracking a macro policy shock at the macro level as it reverberates through the economy and eventually affects the households at the micro level. Following a top-down approach simplifies the analysis because it distinguishes between the economy-wide analysis and the household analysis. This approach is useful especially when adopting a within-group distribution following the second approach because there is no need to integrate the household survey data within the SAM (Lofgren, Robinson, and El-Said, 2003). In this case, the two levels of analysis interact through information—typically prices and wages—passed between the macro level onto the micro level.

Although the analysis in a top-down approach typically ignores household demand and supply responses to price changes, which would be valid only for marginal price changes, these behavioral changes can be easily accounted for by estimating compensating variation or equivalent variation using the functional forms specified in the CGE model. In addition, in a typical CGE model, wages are equalized across sectors. So to the extent this is not a true characterization of the economy, as when labor markets are segmented, important income and distributional effects at the micro household level are ignored. Microsimulation models, discussed below, provide an extension to the analysis as it explicitly addresses some of these issues.

Macroeconomic Closure Rules

Closure rules maintain the equilibrium of the models’ macroeconomic balances, which are globally satisfied by the model independent of individual agents’ optimizing decisions. The macroeconomic balances in a CGE model are for the same exogenous accounts typically chosen in multiplier analysis. These accounts are for (1) the (current) government balance, (2) the external balance (the current account of the balance of payments), and (3) the savings-investment balance. Technically, a closure rule involves a decision on a choice variable to clear each of the macroeconomic balances.

Because they influence the model results, it is always useful to experiment with different closure rules to gain additional insights into trade-offs involved with alternative macroeconomic closure rules.4 This is an area of intense debate in the CGE literature because a number of alternative closure rules are possible for each macroeconomic balance.5 The issue is further complicated by the fact that a choice of one closure rule for one of the balances does not constrain the choice for the other two balances (the appendix discusses commonly used closure rules for the three macroeconomic balances).

In practice, the choice of closure rules should depend on the policy issues being analyzed and the time frame for the analysis. For example, for policy questions involving welfare analysis, a set of closure rules that involves fixed foreign savings, real investment, and real government consumption variables is preferred in order to avoid the un-accounted-for welfare changes resulting from changes in foreign savings and investment. In a static CGE model, these results will not include the welfare changes in later periods that involve a higher foreign debt and lower capital stock accumulation (Robinson, 2005). Similarly, it is analytically plausible to assume fixed government consumption because the model does not keep track of its direct and indirect welfare contributions.6

D. The 123PRSP Model

The 123PRSP model is a quantitative framework developed at the World Bank to address the impact of macroeconomic policies on poverty. It is a multilayered model combining different modeling approaches with a simplified core CGE model. Whereas the models are independently used to evaluate one or more of the macro-micro transmission channels—growth and income distribution—the models are also layered to allow model results to feed into one another (Figure 2.3). For example, an estimated 5 percent short-run growth rate obtained from a trivariate VAR (vector autoregression) model is used as an input to the core 123CGE model, which in turn solves for a new set of relative prices consistent with a 5 percent short-run growth rate (Devarajan and Go, 2003).

Figure 2.3.The 123PRSP Macro-Poverty Links

Accordingly, evaluating the impact of macroeconomic policy on poverty is split among a number of modeling layers as opposed to a full-blown multisector and multihousehold CGE model as described above. At the top is a macro layer consisting of three macro models (a financial programming model, a “Get Real” long-run growth model, and a trivariate VAR short-run growth model) to evaluate the effects of macro policies on growth under the assumption of fixed relative prices, wages, and the composition of output. A meso layer, consisting of a simplified 123CGE model (further described below), takes as input the growth rate of the economy generated from the macro layer to generate changes in relative prices, wages, and sectoral output—under the assumption of full employment. Then a micro layer uses the new set of prices, in combination with income and consumption data, to compute the welfare effects across households (Essama-Nssah 2005).

Core 123CGE Model

The 123CGE model is designed to analyze one country, two sectors, and three commodities. The two sectors—exports (£) and domestic goods (D)—produce two commodities designated for the export market and domestic sales. Imports (M) are the third commodity. The model assumes a constant elasticity of transformation, Ω, between exports and domestic goods, and demand for imports and for domestically produced goods, treated as differentiated goods, is characterized by a specified constant elasticity of substitution, σ The country is assumed to be a price taker in world markets, implying exogenous world prices for imports pwm and for exports pwe. The domestic prices of exported and imported products are given by:

and

where R is the exchange rate, and tm and te are the implicit tariff and export tax rates.7

The desired ratio between m and d is a function of their relative prices. Similarly, producers’ desired allocation between E, and d is set to maximize profits subject to their relative prices. That is,

and

where CES* and CET* refer to the first-order conditions for utility maximization and profit maximization, and ° is the price of domestic goods.

In spite of its simplicity, the 123 model incorporates features capable of addressing a number of policy issues, such as the impact of trade liberalization and devaluation on poverty. Given that D is a domestic nontradable good, the relative price of E or M to D is a real exchange rate (the relative price of tradables to nontradables). Therefore, the effect of macroeconomic policies on an important relative price—the real exchange rate—can be endogenously determined by the 123 model. In addition, underlying the transformation frontier between E and D is a market for labor and capital. That is, with every equilibrium price PD there is also an equilibrium wage rate, which enables the model to generate wages, sector-specific profits (computed as a residual of output after the wage bill), and relative prices (for £), M, and E) in a general equilibrium setting.

Long- and Short-Run Growth Models

The financial programming model is a macroeconomic framework designed to ensure accounting consistency when setting macro and fiscal targets aimed at achieving broad macroeconomic goals (IMF, 1987). In the context of the 123PRSP modeling framework, the FPM provides medium-term projections for the macro variables, which are considered a consensus forecast or a reference growth path for the economy. Alternative macro policies result in a new growth path that is evaluated in terms of deviations from the reference growth path.8

The “Get Real” model is a reduced-form model employing cross-country regressions to estimate long-run growth coefficients for a set of policies. The growth coefficients include policy-determined coefficients, such as inflation, the ratio of M2 to GDP, the real exchange rate, secondary school enrollment, infrastructure (telephone lines/1,000), and a black-market premium. Other estimated coefficients include the effects of economic shocks, such as terms of trade, interest payments on external debt, and growth in an OECD trading partner.

The trivariate VAR model estimates the short-run growth impact of a policy shock (in terms-of-trade shock, a change in government expenditure, and a change in the real exchange rate) on growth rates. The estimated parameters are interpreted as short-run growth elasticities.

The Household Module

The household module evaluates the impact of macroeconomic policies on household welfare. The module uses survey data on households’ labor income and consumption and follows an envelope approach (i.e., first-order welfare analysis) to calculate the net welfare impact of macroeconomic policies through linkage variables determined in other layers of the model. According to the envelope approach, changes in the arguments of the indirect utility function—wages, profits, and prices (linkage variables)—are used to approximate the welfare impact at the household level. This impact is equal to the sum of (1) the initial level of labor income multiplied by the relative change in the wage rate, (2) the change in profit income, and (3) the negative of initial consumption times the relative change in commodity prices Essama-Nssah 2005).

An application of the 123 model to the case of Zambia evaluated the welfare effects of increased public expenditures and deterioration in copper terms of trade (Zambia’s main export commodity). Findings were reported for short-run as well as long-run growth estimates for Zambia’s GDP. In the short run, reported results indicate a positive welfare gain associated with increased public expenditures. However, the gains are relatively low compared with the amount of increase in expenditures and the overall impact on the fiscal deficit. For the terms of trade shock, household welfare generally declines and the distributional impact across household deciles shows the poorest households experiencing higher welfare losses. In the long run, the direction of welfare changes is reinforced and the distributional effects, though harsher on poorest household groups, tend to even out across all groups.

Summary

The 123PRSP modeling framework integrates different modeling approaches and allows for a set of welfare and poverty measures consistent with a set of macroeconomic policies. Compared with a full-blown CGE model, the 123 model provides a flexible framework that allows results from different submodels to be linked in a layered structure where results of one model feed into one another. However, the aggregate nature of the sectoral classifications would suggest that the 123 modeling framework is not useful for looking at specific sectoral reforms, such as rice price liberalization. In addition, the layered structure of the 123PRSP model follows a one-way direction of causality from macroeconomic policies to poverty—with no feedback effect of changes at the micro level on macroeconomic balances.

E. The Integrated Macroeconomic Model for Poverty Analysis

The IMMPA is a dynamic quantitative macroeconomic framework developed at the World Bank for analyzing the impact of policy and exogenous shocks on income distribution and poverty in low-income, highly indebted countries as well as middle-income developing economies (Agénor, Izquierdo, and Fofack, 2003). At its core is a financial CGE model with a number of modeling extensions that focus on labor market segmentation, informal activities, credit market imperfections, and the composition of public expenditures.

The IMMPA modeling framework has a relatively rich modeling structure compared with the approaches described above. Its specification of the labor market distinguishes between rural and urban sectors of the economy, and segments the urban labor market between formal employment (in public and private sector) and informal employment. In addition, labor is heterogeneous in the formal sector where skilled workers earn a different wage than unskilled workers. Furthermore, the model allows for labor migration from the rural sector to the urban unskilled labor market according to an expected wage differential between the two markets.

The financial sector treatment in the model links the financial sector with the real side of the economy, which makes it possible to simultaneously analyze the impact of structural reforms on relative prices and output and the impact of short-term stabilization policies and other financial shocks on the economy. At the same time, the model framework allows for analyzing the effects of alternative government expenditures (infrastructure, education, and health) on productivity of the economy and the accumulation of physical and human capital in the private sector.

The IMMPA model follows an RH approach for the distributional analysis; the IMMPA framework can be used to evaluate the distributional effects of a variety of macroeconomic policies that arise in the context of development strategies. For instance, the model can address the distributional effects of shifts in resource allocation associated with structural adjustment policies: What is the best pro-poor allocation of additional debt relief resources in terms of their equity and poverty alleviation effects? What is the impact of a permanent cut in domestic credit to the government? How do employment and poverty effects vary between stabilization and structural adjustment policies? What trade-offs are involved in terms of sequencing policy implementation? What are the short- and long-term policy implications?

Although the IMMPA framework is capable of addressing a wide range of policy questions, its implementation is constrained by the availability of data and the time required to construct its database. The detailed model specification requires a relatively disaggregated labor market and financial sector database, which could be lacking in many developing countries. In addition, a large number of the modeling extensions require additional parameters and elasticities, which increases the data requirements.

F. Microsimulation and CGE Models

A recent advance has been to merge microsimulation and CGE models in an integrated framework. This approach addresses the limitation of the top-down approach to modeling within-group variations by (1) making the within-group distribution variance endogenous and (2) integrating the household-level behavioral response. Though the CGE-microsimulation merger is fairly recent, microsimulation models have long been applied to tax incidence analysis in developed countries (see, for instance, Orcutt, 1957).9

Previous microsimulation models were partial equilibrium models designed to evaluate the distributional effects of changes in tax policies on household incomes and their impact on the budget constraints and resulting decisions faced by households (Atkinson and Bourguignon, 1991). Household surveys, income tax records, and other surveys are the main source of information for these models.

Merging CGE models and microsimulation models, in which each class of models uses different techniques and sources for data, is a challenging implementation task that raises a new set of issues. The integration between economy-wide analysis at one level and household analysis at another level requires the reconciliation of national accounts data with the data obtained from the household surveys. In addition, from a modeling perspective, an approach is needed to integrate the top-down and bottom-up effects in a consistent modeling framework.

According to the literature, two common approaches are followed in combining CGE models and microsimulation models. In one approach, the two models are layered, and in the second, the two models are completely integrated (see Figure 2.4) (Davies, 2004).

Figure 2.4.Layered and Integrated CGE-Microsimulation

Layered Approach

According to the layered approach, two models—a CGE model and a household microsimulation model—interact through a number of common channels to converge to a consistent solution. Robilliard, Bourguignon, and Robinson (2003) use this approach to address the distributional effects of the 1997 financial crisis in Indonesia. They use a CGE model for Indonesia to capture the impact of the crisis on prices, the exchange rate, and the sectoral structure of production. At the same time, they use a household income microsimulation (HIMS) model—a reduced-form income generation and occupational choice model—to capture the heterogeneity in income sources and other household characteristics (factor endowment, sociodemographic characteristics, and behavior with respect to resource allocation and subject to institutional constraints). The CGE model was solved first, and provided the HIMS model with new data on commodity prices, wages, and employment by sector. Then the HIMS model generated changes consistent (with the CGE model) in individual wages, self-employment incomes, and employment status to arrive at a new size distribution of income.

The Robilliard, Bourguignon, and Robinson (2003) model does not fully incorporate bottom-up effects, however. It is better seen as a constrained top-down approach, because the interaction between the two modeling layers imposes consistency constraints from the CGE model on the HIMS model. For example, the HIMS model employs an occupational choice equation to simulate household employment decisions (work or become self-employed) subject to sectoral employment constraints generated by the CGE model solution.

Savard (2003) extends the constrained top-down approach and follows a bidirectional relationship model, which he calls “top-down/bottom-up,” to capture the feedback effects coming up from the microsimulation model and passing it on to the CGE model. Consistency between the two models is achieved by recursively iterating solutions between the two models to generate a convergent solution.10

Both Robilliard, Bourguignon, and Robinson and Savard compare the microsimulation results with results from a representative household approach. Robilliard, Bourguignon, and Robinson conclude that the RH approach underestimates the inequality and poverty impacts of the crisis compared with their microsimulation results and argue convincingly that the microsimulation improves the distributional analysis. Savard concludes that, although the RH and microsimulation approaches produce similar qualitative and quantitative results at the macroeconomic and sectoral levels, the poverty and inequality results are markedly different at the micro level. Even the signs are reversed when the analysis is carried out under the assumption that capital is sector specific in the short run. Savard concludes that it is quite possible for an RH approach to produce misleading results.

Fully Integrated

An alternative approach is to have a CGE model with as many households as in the household survey. In this case, there is no need to specify transmission channels between two models because the micro-level household heterogeneity—factor endowments, labor supply, and consumption behavior—is subsumed within the CGE modeling framework. Two studies following this approach are Cogneau and Robilliard (2000) and Cockburn (2001). The former is a study of the distributional impact of growth on income distribution in Madagascar, and the latter looks at the trade liberalization effects on poverty in Nepal.

The two studies arrive at contrasting results. Cogneau and Robilliard find that overall poverty results using a parametric distribution (log normal) are not significantly different from those obtained from a fully integrated microsimulation model. However, within a household group, the poverty levels are quite different. Cockburn uses a relatively disaggregated factors of production structure and notes that integrating data from the Nepalese household survey within a CGE framework uncovers enormous variation in the distributional detail within the modeled geographic regions when compared with a RH household approach. However, his final conclusion under the two approaches is the same, eliminating tariffs on imported food in Nepal lowers food prices in the cities and hurts agricultural producers, and the final outcome is lower poverty in the cities and higher rural poverty. In conclusion, integrating the CGE and microsimulation models adds a lot of detail to the distributional analysis of macroeconomic policies, but it remains unclear whether the fully integrated or layered approach more accurately reflects the distributional effects of macro policies (Davies, 2004).

G. Relative Resource Needs

When one chooses among techniques for integrating macro models and models of household behavior, practical trade-offs must be systematically evaluated. What are the advantages and disadvantages of each approach? What kind of policy questions is each technique capable of addressing? How much time does it take to apply these approaches? For example, how much time does it take to construct a SAM for a specific country? If a SAM is readily available, what additional data is needed to implement a CGE or an IMPPA model to address poverty and distribution issues? How much distributional detail is gained when one adopts a CGE-microsimulation modeling framework? What follows summarizes the approach in the context of these questions.11

SAM-Based Multiplier Analysis

Common policy questions addressed using SAM-based multiplier analyses include issues related to government expenditures and transfers, investment, and exports. A SAM would capture the interaction between changes in factor income and household incomes and needs to be supplemented with household survey data to capture the impact of a policy on the size distribution of income. Economic structure and linkages among production sectors, factors of production, and households are the main channels through which macro policies affect the distribution of income and poverty. Constructing a relatively detailed SAM for analysis takes at least one month, with a large variation in the time requirement depending on data availability.12 In addition, the underlying assumptions are often unrealistic: SAM-based multiplier analysis assumes prices are fixed and the results are demand driven with no supply constraints; the analysis is static and economic agents’ behavior follows fixed-share coefficients.

CGE Model

The structure of a CGE model follows the structure of a SAM. Compared with SAM multiplier analysis, a CGE models the supply side of the economy, makes prices endogenous, and considers different functional forms to specify household behavioral responses. CGE models have been applied to a wide range of issues. These include tax policy analysis, subsidies, changes in production technology, public spending, trade liberalization, and devaluation. In addition, the model sectoral detail allows analysis of specific sectoral reforms as well as macro reforms. Provided that a SAM and a CGE model are available, an experienced modeler with substantial exposure to CGE models can generate a base solution using a CGE model in a few days. However, if the policy issue requires adjustment to the model or additional data (e.g., an agriculture-focused CGE model, which would require modeling of crop production and hence a more disaggregated SAM), it can take from a few months to a year to construct a CGE model. CGE model results are sensitive to choice of closure rule and hence require a careful understanding of the macroeconomic paradigm underlying each closure rule.

The 123PRSP Model

Common policy issues to which the 123PRSP model has been applied include changes in government spending, trade liberalization, and changes in terms of trade and foreign capital inflows (or outflows). Growth and income distribution are the main transmission channels through which macro policies affect poverty. Following a layered modeling structure provides flexibility in terms of implementation. Typically, a financial programming model is available, and growth models can be implemented fairly quickly. Implementing a 123CGE model is relatively straightforward and can rely on national accounts data (at the expense of losing sectoral detail). With all the models available, as well as household survey data, an experienced researcher with knowledge of financial programming and time series analysis can set up and implement a 123PRSP model in a few days.

IMMPA

The IMMPA model is a dynamic financial CGE model with a rich modeling structure that can address a number of issues related to stabilization and structural adjustment programs. Compared with a standard CGE model, the IMMPA model has the detailed labor market structure that is often considered essential to the analysis of poverty and income distribution. However, the time and data requirements of the IMMPA model are more demanding. The process of implementing a fully specified IMMPA model requires at least a few months.

CGE-Microsimulation Models

CGE-microsimulation integration is a relatively recent approach in which both macro and micro behavioral models are linked. In a layered approach, the two models iterate recursively until the solution converges to an equilibrium. The integrated approach drops the representative household approach, and all survey household observations are included in a consistent modeling framework. This approach is increasingly applied, but the applications are still evolving. The approach is relatively demanding in terms of time needed to implement. At least three to six months are needed for an experienced modeler to implement a standard CGE-microsimulation model.

H. Conclusions

The chapter reviewed a number of modeling approaches for evaluating the poverty and distributional impact of macroeconomic policies with emphasis on macro-micro techniques that combine economy-wide general equilibrium models with micro models applied at the individual level. The techniques included SAM multiplier analysis and SAM-based CGE models. Three other approaches that build on the CGE modeling framework and also address the poverty and distribution analysis were also considered: the World Bank 123PRSP and IMMPA models and CGE-microsimulation-based models. The list of models selected is not intended to be exhaustive; rather, the idea is to provide a brief and accessible guide to the approaches and techniques common to the analysis of poverty and distribution using this class of models. What are their underlying assumptions and modeling features? What types of policy questions are typically addressed by these models? What trade-offs are involved in terms of data, time, and resource requirements?

These models can be applied to evaluate the poverty and distribution impacts of a number of policy issues relevant to the IMF programs. A choice of using one modeling framework or another depends on the issue at hand, data availability, and the time frame for carrying out the analysis. Typical macro issues would include trade liberalization, devaluation, tax policy, subsidies, composition of public expenditures, and foreign borrowing. Availability of a country SAM (or at least the information embodied in a SAM) is a starting requirement for implementing any of these approaches (SAM is a common database underlying all of the analytical approaches reviewed in this chapter). However, the data requirements vary from one modeling approach to another.

Appendix 2.1. Closure Rules Commonly Used in the CGE Modeling Framework13

The choice of closure rules imposes a particular macroeconomic interpretation onto computable general equilibrium (CGE) model results, and it is essential to understand the implications of the choice for the model behavior. Macroeconomic closure rules are constraints that have to be satisfied by the economic system, but are not considered in the optimizing decision of any micro agent (Robinson, 1989).

There are three macroeconomic balances: the (current) government balance, the external balance (the current account of the balance of payments, which includes the trade balance), and the savings-investment balance. The mechanisms by which the model satisfies these constraints are called the closure rules. A variety of combinations as to how each constraint is equilibrated is possible, and Appendix Table A2.1 provides a summary for a set of closure rules commonly used in the CGE modeling framework.

Appendix Table A2.1.Alternative Closure Rules for Macro System Constraints
Constraint
GovernmentRest of worldSavings-investment
GOV-1:ROW-1:SI-1:
Flexible government savings; fixed direct tax ratesFlexible real exchange rate (via a flexible nominal rate); fixed foreign savingsFixed capital formation; uniform MPS point change for selected institutions
GOV-2:ROW-2:SI-2:
Fixed government savings; uniform direct tax rate point change for selected institutionsFlexible real exchange rate (via a flexible domestic price level); fixed foreign savings (operationally equivalent to ROW-1)Fixed capital formation; scaled MPS for selected institutions
GOV-3:ROW-3;SI-3:
Fixed government savings; scaled direct tax rates for selected institutionsFlexible foreign savings; fixed real exchange rate (via fixed nominal rate and fixed domestic price index)Flexible capital formation; fixed MPS for all nongovernment institutions
SI-4:
Fixed investment and government consumption absorption shares (flexible quantities); uniform MPS point change for selected institutions
SI-5:
Fixed investment and government consumption absorption shares (flexible quantities); scaled MPS for selected institutions
Source: Lofgren and others (2002)Notes: Choice of one closure rule for one of the balances does not constrain the choice for the other two. For example, a possible combination of closure rules could include GOV-1, ROW-2, and SI -3. MPS = marginal propensity to save.
Source: Lofgren and others (2002)Notes: Choice of one closure rule for one of the balances does not constrain the choice for the other two. For example, a possible combination of closure rules could include GOV-1, ROW-2, and SI -3. MPS = marginal propensity to save.

The Government Balance. A typical specification (GOV-1) is to have government savings as a flexible residual with all tax rates for domestic institutions (households and enterprises) being fixed. The other two closures for the government assume fixed government savings with alternative treatments as to how the direct tax rates of domestic institutions are endogenously adjusted to meet changing government spending. For one of these alternatives (GOV-2), the base-year direct tax rates are adjusted endogenously by the same number of percentage points. For the second alternative (GOV-3), the tax rates of selected institutions are multiplied by a flexible scalar.14 A fourth alternative is to have government consumption act as an equilibrating variable while savings and the tax rates are fixed. However, it is typical to keep government consumption either fixed in real terms or as a share of total absorption (along with household consumption and investment demand).

The External Balance. The existence of the rest of the world in the model requires an explicit treatment of how the flows of foreign exchange are equilibrated. Typically, the real exchange rate, defined as the relative price of traded to nontraded goods, is the equilibrating variable.15 Alternatively, foreign savings (net foreign capital transfers) could serve as an equilibrating mechanism. The model provides a choice among three closures with either a flexible exchange rate and fixed foreign savings, or vice versa. Under the first closure (ROW-1) a flexible nominal exchange rate is assumed whereas foreign savings and a domestic price index (here the price index for nontradables was selected) are both fixed. The second closure (ROW-2) differs in that the nominal exchange rate is fixed along with foreign savings, whereas the price index for nontradables is flexible. The third closure (ROW-3) assumes a fixed nominal exchange rate (indexed to the model numéraire) whereas foreign savings is flexible.

ROW-1 and ROW-2 are operationally equivalent, and there is an assumed relationship between the real exchange rate and foreign saving. For instance, under a flexible exchange rate and fixed foreign savings, an increase in foreign savings leads to an appreciation of the exchange rate—a rise in the relative price of nontraded to traded goods—which leads to a fall in exports as output is channeled to domestic markets and imports rise as consumers switch to imported commodities, which leads to a new equilibrium in the external balance with a new increased level of foreign savings.

The Savings-Investment (S-I) Balance. Two common closures for the S-I balance are the savings-driven and the investment-driven closures. Under the savings-driven closure, savings rates are fixed and investment demand adjusts to match the level of savings (a neoclassical closure). For the investment-driven closure, investment demand is fixed and the value of savings adjusts (a Johansen closure). The CGE model provides five alternative closure rules. The first closure (SI-1) is investment-driven, where savings adjust by the same number of percentage points across all institutions. The second closure (SI-2) is also investment driven; however, here, savings are adjusted according to a scalar. The third closure (SI-3) is savings-driven, and all savings rates are fixed whereas investment in each sector is multiplied by a scalar to equate aggregate saving and investment.

The remaining closures (SI-4 and SI-5) are “balanced” closures that can be seen as extensions of the investment-driven closures combined with assumptions on the adjustment mechanism for government consumption. The notion is that both the classical and the Johansen closures can be seen as extreme cases. Robinson and Lofgren (2005) note that in situations in which the aim of the analysis is to evaluate the role of complementary policies, it is preferable to choose a balanced closure where adjustment to macro shocks is distributed across the components of consumption and investment (as opposed to either consumption (investment-driven) or investment (savings-driven)) to absorb the full effect of the shock. Accordingly, the nominal absorption shares of investment and government consumption are fixed at base levels, and similarly the residual share for household consumption is also fixed.

    Other Resources Citing This Publication