Integrated computable general equilibrium (cge) micro-simulation approach

Conventionally, the analysis of macro-economic shocks and the analysis of income distribution and poverty require very different methodological techniques and sources of data. Owing to their economy-wide effects, macro-economic shocks are examined within the context of a computable general equilibrium (CGE) model based on consistent national accounts data but with the strong micro-economic assumption of one or a few representative households. On the other hand, income distribution and poverty issues are normally analysed within a partial equilibrium framework, based on household or individual level data to account for agents’ heterogeneity. Over the last decade however, the natural divide between both approaches has diminished, as evaluating the impact of macro-economic shocks on poverty and income distribution within a CGE framework complemented by household survey data has flourished.

Unlike the sequential approaches presented in the preceding papers, this paper focuses on explicitly integrating into a CGE model each household from a nationally representative household survey. The aim of this paper is threefold. First, we show that explicitly modelling each household in the CGE model addresses Kirman’s critique (1992) and overcomes the strong micro-economic assumption of representative agent. Second, it responds, albeit in a simple way, to the recommendation of Bourguignon and Perreira (2003) to integrate “real” households within a CGE framework rather than using representative households. Third, by providing applications to Nepal and the Philippines, we demonstrate that this technique is straightforward to implement and requires only a standard CGE model and a nationally representative household survey with information on household income and consumption.

The rest of the paper is organized as follows. Section 2 presents the conceptual problems raised by the representative household approach in general equilibrium models, and offers a fully integrated micro-simulation approach as an alternative. Section 3 describes the basic steps to follow to integrate and implement the integrated CGE microsimulation approach and offers some advice on data treatment through illustrations for Nepal and the Philippines. Section 4 discusses simulation results for both the Philippines and Nepal, while section five concludes on the usefulness of this approach.

Although poverty and income distribution analysis within a CGE framework began in the late 1970s (Adelman & Robinson, 1978; Lysy & Taylor, 1980; Dervis et al., 1982), it is not until the late 1990s that this line of research became widespread. One reason is the proliferation of nationally representative household surveys with detailed income information over the course of the decade. Another is that, until recently, CGE modellers have generally shown little interest in the distributive impacts of policies—preferring to focus on welfare effects in terms of equivalent variation (EV) or compensating variation (CV). Nonetheless, the use of CGE models complemented with household survey data is now recognized as well-suited to identifying the mechanisms by which macro-economic shocks affect poverty and income distribution (Winters et al. 2004; Hertel & Reimer 2005).

The traditional approach, as summarized by Lofgren et al. (2004), makes use of “representative” households rather than actual/real households. Distributive impacts are simply captured through extending the disaggregation of the representative households in order to identify as many household categories, generally corresponding to different socio-economic groups, as possible. This approach makes it possible to analyze the impacts of policies on incomes and welfare between groups (inter-group distribution) but not within groups as intra-household distribution is assumed to be fixed. It provides information neither on poverty impacts (as the poor may be found in many different socio-economic groups and in varying proportions) nor on intra-group distribution. In order to address the first limitation, some authors have applied a fixed income distribution function among households within each household group in order to compute poverty indices (such as the FGT indices). One way is to assume a log-normal distribution, where the variance is estimated from the base year data (De Janvry et al., 1991). Meanwhile, Decaluwé et al. (2000) argue that a beta distribution is preferable to other distributions because it can be skewed to the left or right and thus may better represent the types of income distributions commonly observed within household groups. Yet, Boccanfuso et al. (2003) underscore the difficulty of using restrictive functional forms as distribution could change before and after simulations, and large variations in poverty indices may arise depending on the functional form employed.

The representative household approach is based on a very strong theoretical assumption that could be formulated as follows: the choices of households belonging to a given category may be represented by the choices of a unique household that maximizes its utility in such a way that these choices coincide with the aggregated choices of a large number of heterogeneous individuals. Kirman (1992) argues that this hypothesis is not very realistic given that, outside the most restrictive behavioural hypotheses: (1) there is no theoretical justification to affirm that the aggregation of individual choices necessarily leads to the same solution as the choice of a representative individual, (2) there is no guaranty that the reaction of the representative household entails that any change in the model will be the same as the aggregated reaction of the individuals it represents, (3) lastly, the representative household approach may interfere with the individual preferences’ weak principle.

An alternative approach is to integrate separately all individual households from a household survey directly into a CGE model, making it possible to conduct an explicit analysis of the poverty impact of macro-economic shocks on each household. The representative household problem is avoided since individual household behaviour and income distribution is directly captured without the need to impose any functional form. The first applications of this approach date to the very end of the 1990s and are reviewed in Cockburn (2006).

To illustrate the approach, this paper focuses on two specific applications – Cockburn (2006) and Cockburn et al. (2008) – which fully integrate of 3,388 and 24,797 households for Nepal and the Philippines, respectively, without sacrificing the disaggregation of factors, sectors and products required to capture the links between macro-economic shocks and poverty and income distribution.

Constructing an integrated CGE micro-simulation model is technically straightforward since one merely shifts from a model with “representative” households to “real” households by integrating every household from a nationally representative household survey. As in the representative household approach, each household has an income and expenditure vector, but here these are all actual households. All the regular assumptions of a basic CGE model can be retained although, obviously, more sophisticated approaches can be envisaged. The only notable change in the CGE model code is to increase the number of households in the set defining household elements¹, while the manner by which the simulations are carried out remains the same. We used variants of the EXTER model (Decaluwé et al., 2001) as the basic CGE modelling platform.

For the Philippines, labour is disaggregated into skilled and unskilled workers, both segmented into agricultural and non-agricultural labour markets. Capital and land is sector specific. For Nepal, all sectors and factors of production are separated into three regions as households: urban, Terai, and Hills/mountains. Labour is classified into skilled and unskilled. Factors are mobile between sectors within each region but not between regions. Agricultural capital is only mobile among agricultural sectors, just as non-agricultural capital is mobile between all other sectors. National production in each sector is a constant-elasticity-of-transformation (CET) combination of regional productions. As they are expected to be close substitutes, high elasticities of substitution (equal to 10) are utilized. Poverty is analysed using Foster-Greer-Thorbecke (FGT) indices (Foster et al., 1984) computed using DAD software (Duclos et al., 2001).

To implement the approach requires simply a standard Social Accounting Matrix (SAM) and data from a nationally representative household survey with complete information on household incomes and expenditures. Survey data must be adjusted to establish a link to, and ensure consistency with, the SAM underlying the CGE model. Given differences in data sources and year, it is inevitable that there will be inconsistencies in data between the SAM and the household survey. Data reconciliation is necessary to the integration process but, as it brings to light limitations or errors in one or the other datasets, it should not be seen as a “drawback” to the CGE micro-simulation approach.²

Indeed, prior data adjustment in household survey data is normally required for standard poverty/inequality analysis even outside of an integrated CGE microsimulation framework. Deaton (1997) confirms that, in household surveys, it is more difficult to collect reliable information on income than on consumption, although consumption data are not without their faults. Income underreporting and measurement errors are likely to result from the desire to hide revenues from other family members, neighbours or eventual tax authorities, as well as from the long (usually one-year) recall periods involved for intricate sources of income such as returns to assets, agricultural output and seasonal activities. This is further aggravated by the fact that most households in developing countries receive their income from informal production activities in which family income and business revenues are often combined. Thus, the absence of formal accounting in these activities and within the household make it impossible to constitute an accurate picture of the income of producer households and often lead to underestimated income (or overestimated expenditures) generated by these activities.

There is no magic recipe for reconciliation. A thorough understanding of both datasets is required and reasonable assumptions have to be made with the ultimate aim of creating a better and coherent dataset with the least possible adjustments.

The first step is to multiply the household income and expenditure vectors by their respective vectors of sample weights in order to be able to extrapolate to national values as they appear in the SAM. Furthermore, if the survey and SAM data are collected for different years, the survey data must be adjusted to account for inflation in the intervening years. A first comparison of the aggregate household income/expenditure vectors from the survey and the SAM make it possible to evaluate the magnitude and direction of divergences in each of the income and expenditure accounts. There are a variety of reasons why the two sets of vectors will not match. If the initial SAM is based on a previous household survey, it is normal that household endowments and behavior would have changed over time. If instead it is drawn solely from national accounts, the margin for divergences is even greater. Some errors may stem from survey measurement and/or sampling errors. Coverage of informal and household production activities may also vary.

The choice of which dataset to use as an anchor boils down to the modeller’s judgment. For instance, one may keep the structure of the SAM and adjust the household survey data in consequence or vice versa. If errors are suspected on both sides, a reconciliation strategy requiring some adjustments to both datasets may be required. Substantial effort should be devoted to understanding and addressing all sources of divergence. However, ultimately, some arbitrary changes will need to be made in order to ensure coherency.

In this section, we present two country cases – Nepal and the Philippines – briefly explaining the procedure employed to prepare and integrate every household from a nationally representative household survey into a SAM.

To illustrate the integrated CGE microsimulation approach, we study the impact, in both Nepal and the Philippines, of the elimination of all import tariffs with a compensatory uniform consumption tax designed to maintain government revenue constant.⁵ As the consumption tax is applied uniformly to all goods, it does not create any distortions in the relative consumer prices, allowing us to focus on the impacts of the elimination of all tariffs. We present the results in terms of sectoral supply and demand, factor returns, household income and consumption, and finally poverty, bearing in mind that in a CGE model all variables interact and are determined simultaneously. We first discuss the simulation results for the Philippines then proceed with the discussion for Nepal.

The elimination of tariffs affects the agricultural and industrial sectors differently (Table 1). Since import-weighted average tariff rates are higher in agriculture, import prices (PM) fall more, and import volumes (M) increase proportionately more in agriculture than in industry. However, as import competition (intensity) is greater in the industrial sector, domestic prices for locally-produced goods (PD and, net of taxes, PL) and consumer prices (PQ) fall more in the industrial sector, whereas they rest almost constant for services, which are not imported. At the same time, faced with a fixed current account balance, rising imports lead to a real exchange rate devaluation and an expansion in exports. Industrial exports (EX) expand more than agricultural and service exports since industry is more export-oriented. Industry also benefits more from falling imported input costs which, together with the depreciating real exchange rate, make domestic industrial products more competitive in the international market. The net impact is a slight reduction in output (XS) in the inward oriented agricultural sector in favour of the service sector and the outward oriented industrial sector. Output prices (PX) also fall less in industry than in agriculture, despite the greater reduction in prices for domestic sales (PL), given industry’s greater export intensity, where export prices are assumed to be constant.

However, it is value added prices that drive factor returns and, eventually, income and poverty effects. Falling output prices bring a general reduction in both nominal returns to capital (R) and the average wage rate (Table 2). Greater input cost savings imply that value added priced (PVA) fall even less in industry relative to agriculture. The service sector, which is unaffected by import tariff liberalization, emerges with the smallest reduction in value added prices. As a result, returns to capital fall more in agriculture than in industry and services, and agricultural wages fall more than wages for non-agricultural workers.

Obviously, total nominal household income falls across all household categories (Table 3) due to the fall in both nominal wages and returns to capital. The decomposition in Table 3 shows that this income fall is driven primarily by the reduction in wage income, particularly among non-agricultural skilled workers, who contribute a high share in total household income, and agricultural workers, who are hardest hit by the wage reductions outlined in Table 2. Given their reliance on agricultural income, income reduction among male headed households is stronger, when compared to their female counterparts. Similarly, less educated households experience greater nominal income losses than their higher-educated counterparts.

On average, consumer prices fall more than nominal household incomes (Table 4). This is not the case in rural areas, where nominal incomes fall more given the above-noted strong reductions in the returns to agricultural factors. This impact can be traced primarily to households headed by low educated males, for whom nominal incomes fall more than consumer prices, as they are the most dependent on agricultural income.

These income and consumer price changes can be traced to each individual household from the FIES in order to analyze the poverty impacts. Overall, the national poverty headcount index falls marginally, whereas both the national poverty gap and severity index increase marginally, implying that the poor become even poorer (Table 5). Indeed, the poor are overwhelmingly concentrated in rural areas and thus hit hardest by the fall in agricultural income. A clear pro-urban bias in the poverty results emerges as all poverty indices fall significantly in the National Capital region and compare favourably among other urban households relatively to rural households, for whom all poverty indices increase. Households with highly educated heads also benefit more from tariff reductions than their counterparts with low educated heads, given their greater relative reliance on non-factor and non-agricultural income and skilled wages. Moreover, poverty falls dramatically among households headed by highly educated females given their high share of income from non-factor and non-agricultural sources. As the underlying data is available at the individual household level, analysis by other socio-economic categories would, of course, straightforward.

Sectoral supply and demand effects of trade liberalization in Nepal are presented in Table 6. As in the Philippines, the sectors with high initial tariff rates (TM) – paddy, other food crops, mining, and gas/electricity/water – experience the greatest increase in import volumes. The impact of this increase in import volume is tempered however, as imports represent a small share of local consumption in all sectors, except manufacturing and, to a lesser degree, the transport/communication, mining and trade sectors.

In general, the export-oriented sectors – hotel/restaurant, transport/communication and trade – post the greatest output increases. Manufacturing exports are insufficient to offset the sector’s vulnerability to import competition (47 percent of local consumption) such that it posts the greatest output contraction next to mining, which does not export at all. The agricultural sector has very low export intensities, 2 percent or less, and thus posts output contractions and the greatest reduction in producer prices.

The fall in agricultural producer prices translates into a reduction in the remuneration of factors used intensively in agriculture, namely land, unskilled labour and agricultural capital (Table 7). Note that unskilled wages in the agricultural sector falls by about 4 percent, which is roughly as much as the fall in the prices of agricultural output. In contrast, the decline is smaller for urban unskilled labour as it is not so tightly linked to the agricultural sector.

For a better understanding of the household income impacts, the average income changes for households in each region are decomposed into changes in income from each factor (Table 8). Since Terai and Hill/mountain households derive their income primarily from unskilled labour and land, households in these two regions have a more substantial loss (3.3 percent) in nominal income than urban households (1.8 percent). In contrast, urban households earn nearly one-third of their income from non-agricultural capital and nearly a quarter more from skilled wages, which are the two factors that experience the smallest reductions in terms of remuneration rates. These changes imply that, on the income side, trade liberalization favours urban households over rural households.

At the same time, trade liberalization affects consumer prices (Table 9) through the reduction in import prices and, in the face of this import competition, in the prices of local sales by domestic producers. At the same time, a uniform compensatory consumption tax is introduced to keep government revenue constant, which raises consumer prices proportionally in all sectors. Generally speaking, consumer prices fall most in the initially highly protected sectors – paddy, other food crops, mining and gas/electricity/water sectors – and the initially moderately protected but import-intensive manufacturing sector.

What conclusions can we draw in terms of poverty? Making full use of the household-level results, it is possible to draw curves showing the variation in headcount ratios and the poverty gap (Figures 1 and 2) over a wide range of poverty lines (from zero to twice the median income). This gives us a much more detailed view of how the impacts of trade liberalization vary over the whole income distribution. These figures show that results are very sensitive to the choice of poverty line. While there is some evidence of a slight reduction in the number of the very poorest (under 900 rupees, or $US 43, per capita annual income), the number of moderately poor increases as a result of trade liberalisation (Figure 1). Similar results are obtained in terms of the poverty gap, with the very richest clearly emerging as beneficiaries of trade liberalization (Figure 2).

Figure 1

Download asset Open asset

Figure 2

Download asset Open asset

This analysis can be pushed further to examine the impacts among particular socio-economic or regional groups. In results not shown here, the incidence of poverty in urban areas is found to fall, whereas it increases in the two rural areas. Indeed, the urban poor have a greater endowment of non-agricultural capital and depend less on income from land and unskilled labour. Here, too, it is possible to go further into the analysis to contrast, for example, impacts on the extreme versus moderate poor.

This paper has shown how to explicitly integrate into a CGE model each household from a nationally representative household survey to overcome the representative household critique. To the extent that the analysis of macro-economic shocks has differing impacts on individual households, an explicit analysis that integrates household-level income sources and consumption patterns is essential.

We have also shown that constructing a standard integrated CGE micro-simulation model is straightforward, since one merely shifts from a model with several “representative” households to a model with a large number of “real” households. Practically no modification is required to the standard CGE model code and the only data requirements are the SAM and other data underlying a standard model and a recent household survey containing complete income and expenditure data.

The main challenge of this approach is to reconcile the national accounts and household survey data. However, reconciliation is just as necessary in the representative household and sequential microsimulation approaches presented in the other papers of this volume, even if it is often ignored. There is no magic recipe for reconciliation; a thorough understanding of the datasets are required and reasonable assumptions must be made, with the ultimate aim of creating a more reliable and coherent database with minimal data alteration.

Through applications to Nepal and the Philippines, we have shown how this approach can generate rich and nuanced insights into the full distributive impacts of trade liberalization and other macro shocks. A rich agenda for future research remains, particularly in integrating some of the more sophisticated labour market modelling found in sequential CGE microsimulation models. Another extension could be in terms of modelling the motors of growth in order to explicitly identify how each household is affected by and contributes to growth.

1. 1
   Income Distribution Policy: A Computable General Equilibrium Model of South Korea

   1. I Adelman
   2. S Robinson

   (1978)

   Stanford: Stanford University Press.

   • Google Scholar
2. 2
   Poverty, income distribution and cge modeling: does the functional form of distribution matter?. Cahier 03-32

   1. D Boccanfuso
   2. B Decaluwe
   3. L Savard

   (2003)

   Montreal: CIRPEE, Université Laval.

   • Google Scholar
3. 3
   Conclusion: Where do we go from here?

   1. F Bourguignon
   2. LA Pereira da Silva

   (2003)

   In: F Bourguignon, LA Pereira da Silva, editors. The Impact of Economic Policies on Poverty and Income Distribution: Evaluation Techniques and Tools. New York: World Bank and Oxford University Press. pp. 339–349.

   • Google Scholar
4. 4
   Trade liberalisation and poverty in Nepal: a computable general equilibrium micro simulation analysis

   1. J Cockburn

   (2006)

   In: M Bussolo, J Round, editors. Globalization and Poverty: Channels and Policies. London: Routledge. pp. 171–194.

   • Google Scholar
5. 5
   Poverty effects of the Philippines’ Tariff Reduction Program: Insights from a CGE model

   1. J Cockburn
   2. E Corong
   3. C Cororaton

   (2008)

   Asian Economic Journal 22:287–317.

   • Google Scholar
6. 6
   How to build an integrated CGE microsimulation model : step-by-step instructions with an illustrative exercise

   1. J Cockburn
   2. M-H Cloutier

   (2002)

   mimeo, Poverty and Economic Policy (PEP) research network, www.pep-net.org/fileadmin/medias/pdf/microsim.pdf.

   • Google Scholar
7. 7
   Growth distribution and poverty in Madagascar: learning from a microsimulation model in a general equilibrium framework. TMD Discussion Paper 61

   1. D Cogneau
   2. A S Robillard

   (2001)

   Washington, D.C.: International Fool Policy Research Institute.

   • Google Scholar
8. 8
   Analyzing the impact of trade reforms on welfare and income distribution using cge framework: the case of the philippines

   1. C B Cororaton

   (2004)

   Philippine Journal of Development 57:25–64.

   • Google Scholar
9. 9
   Measuring poverty and inequality in a computable general equilibrium model. Working paper 99-20

   1. B Decaluwé
   2. J-C Dumont
   3. L Savard

   (2000)

   Quebec: Centre de recherche en économie et finance appliquées (CREFA), Université Laval.

   • Google Scholar
10. 10
    La politique économique du développement et les modèles d’équilibre général calculable

    1. B Decaluwé
    2. A Martens
    3. L Savard

    (2001)

    Montréal: University of Montreal press.

    • Google Scholar
11. 11
    Politically feasible and equitable adjustment: some alternatives for Ecuador

    1. A De Janvry
    2. E Sadoulet
    3. A Fargeix

    (1991)

    World Development 19:1577–1594.

    • Google Scholar
12. 12
    The analysis of household surveys: a microeconometric approach to development policy

    1. A Deaton

    (1997)

    Baltimore: Johns Hopkins University Press for the World Bank.

    • Google Scholar
13. 13
    General Equilibrium Models for Development Policy

    1. K Dervis
    2. J de Melo
    3. S Robinson

    (1982)

    Cambridge: Cambridge University Press.

    • Google Scholar
14. 14
    MIMAP Programme

    1. JY Duclos
    2. A Araar
    3. C Fortin

    (2001)

    MIMAP Programme, International Development Research centre, Government of Canada and CREFA, Université Laval, Montreal.

    • Google Scholar
15. 15
    Microsimulations in Computable General Equilibrium: Procedures for analysing and reconciling data

    1. I Fofana
    2. J Cockburn

    (2003)

    Microsimulations in Computable General Equilibrium: Procedures for analysing and reconciling data, mimeo, Poverty and Economic Policy (PEP) research network, www.pep-net.org/fileadmin/medias/pdf/Reconciliation.pdf.

    • Google Scholar
16. 16
    A class of decomposable poverty measures

    1. J Foster
    2. J Greer
    3. E Thorbecke

    (1984)

    Econometrica 52:761–766.

    • Google Scholar
17. 17
    Predicting the poverty impacts of trade reform

    1. T Hertel
    2. J Reimer

    (2005)

    The Journal of International Trade and International Development 14:377–405.

    • Google Scholar
18. 18
    Whom or what does the representative individual represent?

    1. A Kirman

    (1992)

    Journal of Economic Perspectives 6:117–136.

    • Google Scholar
19. 19
    Poverty and inequality analysis in a general equilibrium framework: the representative household approach

    1. H Lofgren
    2. S Robinson
    3. M El Said

    (2004)

    In: F Bourguignon, AL Pereira da Silva, editors. The Impact of Economic Policies on Poverty and Income Distribution: Evaluation Techniques and Tools. Washington D.C.: The World Bank. pp. 325–339.

    • Google Scholar
20. 20
    The general equilibrium income distribution model

    1. F Lysy
    2. L Taylor

    (1979)

    In: L Taylor, EL Bacha, EA Cardoso, F Lysy, editors. Models of Growth and Distribution for Brazil. London: Oxford University Press. pp. 128–139.

    • Google Scholar
21. 21
    Using weather variability to estimate the response of savings to transitory income in Thailand

    1. C H Paxson

    (1992)

    American Economic Review 82:15–33.

    • Google Scholar
22. 22
    Regionally disaggregated social accounting matrices of Nepal, 1986/87, mimeo, Himalayan Institute of Development, Kathmandu, Nepal

    1. PR Sapkota

    (2001)

    Regionally disaggregated social accounting matrices of Nepal, 1986/87, mimeo, Himalayan Institute of Development, Kathmandu, Nepal.

    • Google Scholar
23. 23
    Trade liberalization and poverty: The evidence so far

    1. LA Winters
    2. N McCulloch
    3. A Mckay

    (2004)

    Journal of Economic Literature 42:72–115.

    • Google Scholar

Author details

1. John Cockburn

   Poverty and Economic Policy (PEP) research network and Université Laval, Canada

   For correspondence

   jcoc@ecn.ulaval.ca

2. Erwin Corong

   Centre of Policy Studies Monash University, Australia

   For correspondence

   erwin.corong@buseco.Monash.edu.au

3. Caesar Cororaton

   Global Issues Initiative, Virginia Polytechnic University, United States

   For correspondence

   ccaesar@vt.edu