The impact of trade policies on income distribution in a planning model for Colombia

World Bank Reprint Series: Number 170 Jaime de Melo and Sherman Robinson The Impact of Trade Policies on Income Distribution in a Planning Model for Colombia Reprinted with permission from Journal of Policy Modeling, vol. 2, no. 1 (1980), pp. 81-100.  The Impact of Trade Policies on Income Distribution in a Planning Model for Colombia* Jaime de Melo, Georgetown University Sherman Robinson, World Bank This paper develops a multisector Computable General Equilibrium (CGE) model to simulate the effects of trade on the distribution of income among socioeconomic groups defined both hy the factors of production they own and the sector in which they work. The categorization of recipie.nts includes landless rural labor, land owners, workers in the urban traditional sector, and workers in the organized sector and capitalists. Experiments are conducted with an application to Colombia, a primary-exporting economy. The results indicate that, for such an economy, outward-looking policies with increased primary exports are likely to be more detrimental for the distribution of income in the medium term than inward-looking ones. INTRODUCTION Policy makers now have available a large collection of comparative studies analyzing the experience developing countries have had with their foreign !.-ade regimes. The focus of these national studies has been a detailed evaluation of the efficiency implicationis of industrialization policies based on a blend of inward- and outward-looking trade strategies. However, it has been difficult to distill evidence from these studies concerning the effects of trade policies on the functional or size distribution of income. In reviewing the impact of trade policies on income distribution from the country studies in the NBER project on Foreign Trade regimes, Bhagwati (1978, p. 201) concluded that "the Project evidence on the income distributional and egalitarian effects of the foreign trade regimes is somewhat sketchy and mixed." Having noted that although the functional distribution of income can be strongly related to foreign trade regimes in theoretical analysis and that the evidence from the country studies did not Address correspondence to Jaime de Melo, Economics Department, Georgetown University, Washington, D.C. 20057. *An earlier version of this paper was presented it the Seventh International Conference on Input-Output Techniques held at Innsbruck, Austria, April 1979. We wish to acknowledge financial support from the Agency for International DevelopmenL Our work has greatly bene- fitted from other work undertaken jointly with Kemal Dervis, and we should also like to thank Martha de Melo for comments on an earlier draft Neither they, AID, nor the World Bank are responsible for any views and interpretations expressed in this paper, Journal of Policy Modeling 2(i), 81-100 (1980) 81 OSociety for Policy Modeling, 1980 0161-8938/80/010081-20$01.75  82 Jaime de Melo and Sherman Robinson reveal a strong and predictable relationship, he concluded that the income distributional outcome may reflect more basic underlying distributions of wealth and power. Lacking an adequate theory of the determinants of the distribution of wealth and power, and in the face of inconclusive evidence from compara- tive country studies, one can attempt to quantify the likely effects of alternative trade strategies on the distribution of income by imposing on the analysis knowledge based on the barter theory of trade. This is the approach adopted here. We present a model for tracing out the effects of trade policy on the functional distribution of income, taking into account some of the characteristics of a developing economy where trade policy is likely to be the main vehicle of industrial development. Thz model is in the tradition of planning and trade models which focus on real flows. Prices enter only as relative prices, and macroeconomic variables such as the money supply and inflation are not considered. The analysis also abstracts from Keynesian issues on the composition of aggregate demand and focuses instead on issues of the sectoral structure of production, demand, and factor use. Two often-cited characteristics of a developing economy not emphasized in trade theory but that impinge on the effects of policy on income distribution are captured in the model. The first is the "structuralist" or "bottleneck" phenomenon, stressed in two-gap models, which emphasizes the difficulty of transforming foreign resources into domestic ones. This difficulty is captured on the one hand by assuming imperfect substitution between domestically produced goods and imported ones and, on the other, by specifying export functions that incorporate supp-ly limitations and the difficulties associated with penetrating new markets. The result is that protection raises the cost of essential intermediate inputs, which are often imported and cannot easily be produced locally, so that domestic produc- tion is not much stimulated by a protectionist policy. Likewise, export- oriented sectors, though they are somewhat more responsive to policy inducements than are import-competing sectors, will not expand much. The second characteristic of a developing country that calls for a multisector economy-wide model and has been the focus of much recent development literature is based on the pervasive dualism between industry and agri- culture. Although this dualism is not well understood and certainly not easily quantifiable, it is probably best captured by a categorization of income recipients along lines that would include landless rural labor, farmers, workers in the urban traditional sector, workers in the organized modern sector, and capitalists (perhaps further broken down by sector). This categorization of income earners into broad socioeconomic groups with different consumption expenditure patterns allows one to trace out the  TRADE POLICIES AND INCOME DISTRIBUTION 83 effects of trade policy on the cost of living and hence on the real income of each of these different groups. Following the presentation of the main characteristics of the model and the important mechanisms by which changes in trade policy affect the distribution of income, we describe the application of the model to Colombia. Results are then presented for a number of simulation experiments designed to explore the impact of different policy regimes on the distribution of income followed by our conclusions. Model specifi- cations are suemmarized in the Appendix. OUTLINE OF THE MODEL The recent emphasis on distributional issues in developing countries has led to the resurgence of models designed to investigate the determination of the functional distribution of income. Broadly speaking, these models fall into two categories. On the one hand there are short-term aggregate macroeconomic models that focus on the determination of the wage share. These models typically investigate the effects of alternative closure rules and focus on the distributional shifts caused by changes in real incomes induced by changes in prices and wages in aworld where some income flows are fixed in nominal terms (Taylor 1979). On the other hand, there are medium-term multisector microeconomic models that have been used to investigate the effects of policy on the distribution of income among socioeconomic groups (Adelman and Robinson 1978). Neither group of models has properly focused on the role of the external sector in income distribution. The model outlined here firmly belongs to the second category. It is a microeconomic general equilibrium model in the spirit of the discussions found in the literature on planning and on the barter theory of trade, where money is a veil and the government transfers income between groups in a lump-sum, nondistortionary manner. The model allows no interaction between trade and economic growth, which significantly narrows the scope ot discussion. Also, the important issues of the distributionai aspects of trade-induced growth are skirted. This limits the present analysis to what is an essentially -static framework and implies a closure rule whereby the aggregate level of investment is fixed in real terms. Given the specified technology and behavioral rules, the model endo- genously determines wages, profits, product prices, and the exchange rate; sectoral production, employment, consumption, investment, exports, and imports; the nominal flow-of-funds, including both the government and private sector accounts; and, finally, the distribution of income to socio- economic groups and the overall size distribution by individuals.  84 Jaime de Melo and Sherman Robinson Production Technology and Factor Markets Each sector produces output with a two-level Cobb-Douglas production f-nction. Different categories of labor are combined via a Cobb-Douglas .Si_-;tion into a sectoral labor aggregate. Capital goods are combined in fixed proportions (that vary by sector) into aggregate capital. Sectoral output is a Cobb-Douglas function of the sectoral labor and capital aggregates. Intermediate inputs are required according to fixed input- output coefficients.' Each sector is assumed to maximize profits and thus hires labor until the wage equals its marginal revenue product. Thus nonlinear labor demand functions by sector Eaid skill category are given by the first-order conditions for cost minimization. Coupled with fixed aggregate labor supplies by skill category, the model solves for a set of average wages that equate the aggregate supply of and demand for labor. We distinguish three categories of labor in the model. The first is agricultural landless labor, which is tied to agriculture, i.e., is mobile only across the two agricultural sectors. The second is unskilled labor, which is perfectly mobile across all sectors. The third is skilled (or "modem" sector) labor, which is mobile across the urban sectors only. The rather limited mobility of labor between the agricultural and nonagricultural sectors is a reasonable specification for a medium-term model of a developing country. The capital stock in each sector is assumed to be fixed during a given period and hence is not mobile across sectors. The rental rates of capital (or the profit rates) will not generally be equal across sectors. They are thus sector specific and are computed residually for each sector. Income Distribution and Product Markets The model deterrnines all the flow-of-funds accounts within the economy. The model distributes factor income-payments to labor and capital-to the three types of labor, land owners of three different farm sizes, owners of capital in the manufacturing sectors and in the service sectors, and government. For all of these socioeconomic groups, the model determines both income and number of members. The model thus solves for the functional distribution and also for what might be called the socio- economic distribution of income. The latter is interesting in its own right because the groups, if appropriately defined, reflect social and political divisions in the society that are especially relevant for policy analysis. One might argue that given the political significance of the groups, the 1The Cobb-Douglas specification abstracts from distributional shifts caused by changes in the wage rate.  TRADE POLICIES AND INCOME DISTRIBUTION 85 socioeconomic distribution is more important than either the aggregate functional distribution (to capital and labor) or the overall size distribution of personal or household income. Such an emphasis, however, neglects the interest policy makers have in questions such as IHow many people are living in poverty? What is the socioeconomic composition of the poverty group? and What is the degree of inequality in the overall size distribution? We approach these questions by incorporating a technique for generating the overall size distribution from information about the socioeconomic dis- tribution and separate information about the within-group distributions. The model determines the overall size distribution of income in two steps. First, the distribution of income within each socioeconomic group is described by a two-parameter lognormal probability distribution function. The logvariance for each group is specified exogenously and the logmean is computed given the logvariance and group mean income (which is solved endogenously). Second, the overall size distribution of income is then computed by aggregating the set of within-group distributions.2 Statistics describing the overall distribution, such as the Gini coefficient, number of people in poverty, socioeconomic composition of the poverty group, and so forth, are all computed numerically from the aggregated distribution. Thus, a particular functional form is not assumed for the overall distribution, but only for the separate within-group distributions. This technique for generatiing the overall size distribution involves a number of important assumptions. Most important, it is assumed that the policies whose impact is being analyzed do not affect the within-group distributions (since the logvariances are assumed fixed). Thus experiments only affect the intergroup distribution. For the range of trade policies being considered in this paper, such an assumption seems reasonable. The assumption that the within-group distributions are described well by lognormal distributions is a testable hypothesis, given data. We were forced to rely on secondary sources for our within-group logvariance estimates [ Berry and Urrutia ( 1976), and Cordoba et al. (197 1)1, and so were not able to test the hypothesis statistically. However, we are most concerned with comparing the socioeconomic composition of the poverty group across experiments, a result that is likely to be quite robust to the exact specification of the within-group distributions. The behavior of the government sector is simulated by the closure rules, which require that expenditure and investment add up to total net government revenue (includirig direct and indirect taxes). It is assumed that government consumption and investment remain fixed in real terms 2This technique for gencrating the size distribution of income is used in Adelman and Robinson (1978). The algorithm is described in Robinson (1976).  86 Jaime de Melo and Sherman Robinson (deflated by the overall price index) at their values in the reference solution to which all trade policies are z-mpared. This is achieved by a system of proportional (to income) trans.ers between the government and each socioeconomic group, Finally, to isolate the analysis from the interactions between trade policies and growth via changes in the level of investmient (determined by the propensities to save of the different groups and ihe government), a fixed level of aggregate real investment is maintaiaied throughout all experiments. Group savings rates are adjusted propor- tionately to validate the fixed investment level. Each socioeconomic group is assumed to have different consumption patterns and hence different expenditure equations, which are based on Stone's linear expenditure system (LES). Hence consumer demand is a fiunction of relative prices as well as of the distribution of income. Prices are then solved so that demand equals supply in each sector. As discussed below, the exchange rate is endogenously determined to maintain the balance of trade. Foreign Trade The mechanisms whereby a change in trade policy affects the distribution of income between factors of production is well understood under the assumption that domestically produced and foreign goods are perfect substitutes in use. If factors of production are perfectly mobile and world prices are given, the percentage change in the domestic price for each sector is equal to the percentage change in the tariff which, in turn, is equal to a weighted sum of changes in factor prices, with the weights given by the distributive shares of each factor in the total product of that sector. Consider, for instance, the case where there are two sectors, agriculture and manufacturing, and let the capital stock in each of these sectors be fixed. Then capitalists' rents are detennined residually. What happens to the wage of the mobile factor, labor, depends both on the factor intensities (as measured by distributional shares) and on the elasticity of substitution between capital and labor. If the elasticity of substitution is high in agriculture and low in industry, the wage will closely follow the price behavior of industry.3 The wage will thus be related to the effective rate of protection in industry. The key component in the mechanism just described is the domestic price system, which in a small open economy is entirely determined by world prices. Furthermore, as Samuelson (1953) has shoNvn, the country will specialize and produce at most as many goods as it has Flactors of production. At the other extreme, some sectors, such as services where quantities traded are usually a small fraction of total supply, are classified 3An analysis of this mechanism is provided by Mussa (1974).  TRADE POLICIES AND INCOME DISTRIBUTION 87 as nontraded goods and their price is assumed to be entirely determined on domestic markets. From an empirical point of view, this traded/nontraded good dichotomy is too coarse, and the assumption that domestic and foreign goods are perfectly interchangeable inevitably overestimates the distri- butional effects of changes in trade taxes. Following Armington (1969), we assume that domestic and foreign goods are imperfect substitutes in use and can be aggregated into a composite commodity. Under this assumption, which helps capture some of the structural characteristics of a developing economy, the demand for imports and domestically produced goods are derived demands, analogous to those forfactor inputs in a tradition.', model. In contrast with the standard trade model, in which the domestic price system is entirely determined through foreign trade, dom.stic prices are endogenously determined by the model and acquire a substantial degree of autonomy. On the expoit side, the ratio of exports to domestic production is assumed to be a decreasing function of the ratio of the domestic price to the export price. The functional form adopted for the export function is an asymmetric logistic function with the inflection point at the base-year export ratio, an upper asymptote of 1, and a lower asymptote of 0. As with the import demand functions, this specification precludes complete specialization since sectoral expdrts can neith\ r disappear nor include all production (in those sectors for which there were some exports iri the base-year data).4 Although this specification is somewhat restrictive, it is plausible for an empirical study where the degree of sectoral aggregation is high and one could reasonably expect intra- (rather than inter-) industry specialization. It is also a specification that could be justified on the basis of placing reasonable constraints on potential increases in export supplies due to rising selling costs for export sales (relative to domestic sales). The implications of our speciFication of trade for the autonomy of the domestic price system are important Sectors are no longer either traded or nontraded; rather, they are characterized by their degree of tradability. according to whether changes in the domestic price are closely linked to, or relatively independent of, changes in the export and import price. For example, a 1 0% tariff on machinery imports will afffect the price of domestic machinery through its effect on the price P of aggregate machines. In another paper (de Melo and Robinson 1978) we show that the respon- siveness of the price of domestic machinery to a change in the import price is an increasing function both of the elasticity' of substitution betwvecn imported and domestic machines and of the share of imported machines in 4Notethat ifilic logistic is verv steep. thenii tl model approachesan issumptiin of rnosupply constraint. In this extreme case, jfpd > pe, there are no exports at all. Ifpd = p, then exports are residually dctcmiincd as the difference between domestic demand and domestic supply.  88 Jaime de Melo and Sherman Robinson the aggregation function. In this miodel, the only pure nontraded sectors are those for which the import shares are zero and there are no exports (e.g., construction and housing in the Colombian, application). What is the exchange rate in this model, and what does an adjustment in its value mean? First, as in the barter theory of trade, only relative prices are determined in the model. Some numeraire or normalization rule is therefore required. We choose to normalize around the aggregate value of all composite goods, using base-year quantity weights in the index [see Appendix, equation (12)]. Therefore the exchange rate in this model is the relative price of imports and domestic goods. For instance, a tariff leads to an improvement in the trade balance matched by an increased demand for domestic goods. Therefore, the price of domestic goods would tend to rise relative to that of imports. With a fixed price level, this may be translated into an appreciation of the exchange rate. APPLICATION OF THE MODEL TO COLOMBIA We now turn to a simulation of alternative trade strategies on the distribution of income in Colombia.5 Three alternative policy regimes have $For an interesting study simulating the likel) effects of growth on income distribution C.ad employment using a multisector framework, see Thorbecke and Sengupta ( 1972). Table 1: Structure of the Colombian Economy in the Free Trade Base Run (1) (2) (3) (4) (5) Gross Value Output Ratios (