How trade liberalization affected productivity in Morocco

Policy Research WQRKING PAPERS Trade Policy c Policy Research Department The World Bank February 1993 WPS 1096 How Trade Liberalization Affected Productivity in Morocco Mona Haddad Trade liberalization in Morocco improved productivity in manu- facturing firns, so they could exploit their comparative advan- tage and compete better with foreign firms. Psliy RoacraWorkingPap mdissaninteflndings of work in pnag nd oenowgethechangeofideus amongBank staff and alolts mmdsin vdovl iop n. etaThceepapan.dibued by theReshAdvisor; Staff,cnrty thenames of X autho,rfdlet only thiviewandsahodbeued and ci edacaordingly.Thefind.inaipretions.andconcluioDs arotheawhoseown.Theyahodd not be aunibued to the Wodd Bank. its Boad of Diatos, its managaen, or any of its manber counriea.  Policy Research | ~~~Trade Policy WPS 1096 This paper-a product of the Trade Policy Division, Policy Research Department-was prepared for the World Bank research project, Industrial Competition, Productive Efficiency, and Their Relations to Trade Regimes (RPO 674-46). Copies of the paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Dawn Ballantyne, room N 10-023, extension 37947 (February 1993, 39 pages). The economic literature now accepts theoretical the endogeneity of factor inputs or because arguments that liberal, outward-oriented trade managers have some knowledge about the noise policy is better than restrictive, inward-oriented in the production function. policies. Traditionally such arguments for the gains from trade have rested on the concept of Haddad then estimated the effect of various allocative efficiency. But a new argument for trade and market-structure variables on the level liberal trade has emerged: increased technical of TFP, as well as on the deviation of firm TFP efficiency or productivity. The best-known from the efficiency frontier. The results are not attempts to link trade policy and productivity are very sensitive to the different measures of TFP b r ed on "X-efficiency," economies of scale, and show that trade openness has a significant capacity use, increased competition, and techno- positive effect on firm productivity through: logical catch-up. i Outward orientation from export promotion. Haddad estimates total factor productivity (TFP) at the firm level using panel data from the * Import liberalization. Moroccan industrial census in a production- function framework during Morocco's period of * More direct foreign investment. trade liberalization (1984-89). Haddad corr.cted for several problems that usually bias the e.,- By splitting the sample into protected and mate of productivity. The use of panel data unprotected sectors, Haddad showed lower allowed Haddad to take into account the hetero- productivity in protected sectors. geneity across firms. These firm-specific effects were tested for randomness. Differences between The results are clear. Trade liberalization in large firms and small firms were checked. She M3rocco improved productivity in manufactur- also corrected for errors in measuring capital ing firns, so they could exploit their comparative stock, so common in data from developing advantage and compete better with foreign firms. countries, and for simultaneity bias because of ThePolicy Research Working PaperSeries disseminates the fndings of work under way in the Bank. Anobjectiveof the series is to get these fndings out quickly, even if presentations are less than fully polished. The findings, interpretations. and conclusions in these papers do not necessarily represent official Bank policy. Produced by the Policy Research Dissemination Center  How Trade Liberalization Affected Productivity in Morocco by Mona Haddad This paper was prepared for the World Bank research project "Industrial Competition, Productive Efficiency, and Thteir Relations to Trade Regimes (RPO 674-46)." The author c.nks Ann Harison, Oleh Havrylyshyn, Jaime de Melo, and James Tybout for their comments and support.  TABLE OF CONTENTS Page I. Introduction I 11. Specification and Estimation of a Production Model 3 1. Specification of the Production Model 3 2. Estimation Techniques with Panel Data 6 M. Trade Policy in Morocco 9 IV. Estimation of Firm-Level Productivity 11 V. Estimating the Link Between Productivity and Trade Policy 14 1. Estimation Model 14 2. The Results 17 3. High-Protection Versus Low-Protection Sectors 21 VI. Conclusion 23 Appendix 24 1. Data 24 2. Descriptive Statistics of the Moroccan Industrial Sector 25 3. Empirical Esmation of the Production Function 26  1. INTRODUCrION Theoretical arguments for the preeminence of liberal, outward-oriented trade policies over restrictive, inward-oriented ones are now widely accepted in the economic literature. Traditionally, these arguments for the gains trom trade rested on the concept of allocative efficiency, whereby an open economy is more likely to allocate its resources in areas where it has a comparative advantage. Yet another case in favor of more liberal trade has recently emerged In terms of increased technical efficiency or productivity. The best known attempts to link trade policy 3nd productivity are based on 'X- efficiency", economies of scale, capacity utilization, increased competition, and technological catch-up. First, trade liberalization can change the opportunity cost of leisure in such a way that managers work harder. That is, the return to entrepreneurial effort is increased by exposure to foreign competition, inducing managers to make an extra effort at eliminating inefficiency. Second, the existence of economies of scale implies that a widening of the market through trade should lead to reductions in real production costs, mainly in terms of increased demand through export expansion. The same argument holds for increased capacity utilization. Third, in a protected market dominated by several firms, trade reform will lead to increased competition, and hence a reduction of monopolistic inefficiency. Finally, trade reforms are likely to accelerate the transition to state-of-the-art technologies since domestic producers are more exposed to foreign competition. The handfil of studies which attempted to quantify the allocative gains from liberal trade policies found, in general, weak results. However, much greater benefits are likely to emerge from improvements in productivity. Unfortunately, the latter are more difficult to measure and the empirical literature does not offer definitive evidence on the effect of trade reform on productivity. Several recent overviews of the links between trade regimes and productivity gains (Tybout 1991, Havrylyshyn 1990, Bhagwati 1988, Nishimizu and Page 1987) suggest that the evidence is mixed. I  One possible explanation for the lack of conclusive results may depend on how productivity is measured. The empirical research on industrial productivity has suffered from two major shortcomings. First, a large number of studies' were based on the traditional measure of total factor productivity, pioneered by Solow (1957). The consistency of this measure depends on the validity of the assumptions it makes, namely perfect competition, constant returns to scale, and perfect mobility of all inputs. Yet, although the potential biases of the productivity estimates which take place when these assumpdons are violated have long been recognized2, litde was actually done to correct for these errors. Second, even when the problems of scale economies, quasi-fixed factors, and non-competitive pricing are successfilly dealt with, the problem of aggregation remains. Most studies which attempted to estimate productivity have used macro or sectoral data, implicitly assuming that a well defined prciuction technology describes all plants within the industry, sector or country of analysis. Tybout (1991) points out that "if technological innovation takes place through a gradual process of efficient plants displacing inefficient ones, and/or through the diffusion of new knowledge, the approaches to productivity measurement based on 'representative plant' behavior are at best misleading. At worst, they fail to capture what is important about productivity growth altogether, as Nelson (e.g. 1981) has long argued". In this study, we will first attempt to get a consistent estimate of productivity by using industrial census data and taking into account the heterogeneity across firms. Second, we will ask the question: Does trade liberalization actually increase firm-level productivity? In section 11, the production model and estimation techniques will be discussed. In section m, recent changes in the Moroccan trade policy will be reviewed and evaluated. Section IV describes the estimated TFP. In section V, the estimation results of the link between productivity and trade are presented. The conclusion is given in section VI. 'See for example Nishimizu and Robinson (1984) or Krueger and Tuncer (1982). 2See for example Nishimizu (1979) or Kim and Kwon (1977). 2  II. SPECIFICATION AND ESTIMATION OF A PRODUCIION MODEL 1. Specification of the production models 3:he pron techngy: We begin with a stochastic Cobb-Douglas production function: (1) Yb, = A La, Kf, e where the subscripts i and t represent the firm and the time period respectively. The industry subscript has been suppressed. Y is value added, L is labor measured in efficiency units, and K is true capital stock. A is the average level of Hicks-neutral technical efficiency within an industry. a and 0 are scalars for which the sum represents returns to scale for each industry. The error term u,, is assumed to have three components: (2) us + T + t where y, is a firm-specific effect that reflects firm efficiency and management skills; Tt is a time effect common ^o all firms that reflects industry-level changes such as general fluctuations in capacity utilization, technological innovation, and returns to scale; ,, is a random disturbance reflecting the remaining noise across firms and time which represents factors such as luck, weather conditions, and unpredicted variation in machine or labor performance. All error component are unobservable to the econometrician; however, both 1A and ,, may be observable to the managers. In this case, they will be correlated with the exogenous variables as will be shown later. On the other hand, the errors represented by {,k are uncorrelated with the exogenous variables and are assumed to be independently and identically distributed across firms and time. In this 3This model is an extension of Tybout (1990). 3  production function, pi wUI depict the firm-level technical efficiency which we would like to esdmate and wUll be represented as a fixed or a random variable. The producer behavior: Produeers are assumed to maximize short-run profits. However, because of the stocL>stic nature of the production process, any given level of inputs will result in an uncertain level of output, and therefore, in an uncertain profit. The concept of profit maximization becomes ambiguous due to the presence of the random elements. It is therefore necessary to gear the problem towards the maximization of eected profits. However, this will involve the inclusion of the variance of the production function disturbance (see Zellner, Kmenta, and Dreze 1966). In order to avoid carrying along this extra term, we assume median profit maximization (see Kumbhaker 1987). Furthermore, we assume that prices (of output, labor, and capital) are either known with certainty or statistically independent of the production function disturbance term. More specifically, with a short- run production function, capital is fixed and labor is variable. It becomes natural to assume that the price of capital is known with certainty since, typically, capital is purchased before it is used in production. On the other hand, let the expected real wage for labor be related to its actual ex=ost value for each firm according to the following equation (3) WI,, = Whe*. The expected short-run profit funcidon can now be written as (4) E(rk) =E(Y) -W,,L,, A Lf. Kf, E(e) - Wit E(eb) Lk 4  By taking the first order condition of the median profit, after decomposing the error term ul, into its three components and assuming that ;I and r, are observed by managers, we get: (5) dx1/dfl, = (a/L,"Ykev - Wihe = 0 In logarithmic terms we have (6) lnL, = Ina + InYj, - lnWk, - 4 - {it From equation 6 it becomes clear that the demand for labor by the firm not only depends on output and wages in the same period, but a1so on the unforseen random elements in both production and real wages. By combining the first order condition with the production function, we get the reduced form for employment (7) hd, = (1-a) tUna + I1A + PnIC - InW,, - ;4 - -rt - f From equation 7, we can see that labor is only affected by the components of the production function's error term that are observed by managers (i4 and ,j and not by the unobserved component (,). Ihereibre, whenever managers have knowledge about a portion of the production function's disturbance, the employment decisions will be affected by it. In this case, simultaneity problems arise and labor cannot be taken as exogenous in the production function. However, if managers do not have knowledge of any portion of the production function's random element, equation 7 will be completely independent of u, and the simultaneity problem is eliminated. 5  Whether 4 is observable or not to managers, It will represent our technical efficiency esdmate, whilo the sum of the estimated a and 0 will represent an index of returns to scale. 2. Estimation techniques with panel data Given the nature of our data (cross-section, time-series), the empirical estimations for this model are based on panel data techniques. The use of panel data improves the efficiency of the econometric estimates and allows the introduction of firm-specific effects (representing technical efficiency in our production model) which can be treated as fixed constants or as random variables. Each case is briefly discussed below', assuming for the moment that all inputs are exogenous. The fixed-effect model: The firm-level productivity p4 is assumed to be fixed and can therefore be estimated as an intercept which varies across firms by introducing dummy variables. Assuming for simplicity that there are no time-specific effects, we have the following model (8) Y,, = 14 + 'X; + tk where i = 1, ..., N and t = 1, ..., T. Y, is the dependent variable (output) for the P flrm at time t, X& is a Kxl vector of K exogenous variables (inputs), y' is a lxK vector of constant parameters, and p, is a lxl scalar constant representing the effects of the variables specific to the i' firm and invwart over time'. The ^ for each i is obtained by including i dummy variables which take the value 1 for the corresponding i and 0 otherwise. The error term t, represents the effects of the omitted variables that are 'More details can be found in the econometric literature on panel data (see for example Hsiao, 1986). 'Note that we are using vector notation. 6  both time and cross-sectional variing. Assuming that ki, Is independently and identically distributed, the OLS estimator for A is: (9) = - i;V11 where Y, = (IM)2:Y, and X! = (lITEXk, and j- is the OLS estimator of y. The estimator of y obtained from the fixed-effect model is sometimes called the covariance estimator or the within-group estimator, because only the variation within each group is utilized in forming this estimator. It is known from the literature that the covariance estimator 4: is unbiased. It is also consistent when either N or T or both tend to infinity. However, the estimator for the intercept A, although unbiased, is consistent only when T tends to infinity. the random-effect model: In the previous section, we treated the firm-specific technology effects pz as fixed constants over time. Alternatively, these firm-specific effects can be treated as random variables, like El. It is standard in regression analysis to assume that factors which affect the dependent variable, but are not explicitly included as independent variables, can be appropriately summarized by a random disturbance. In the case of panel data where some omitted effects vary across time but are firm- invariant, and others vary across firm but are time-invariant, it is natural to assume that the residual ui consists of three random components (see equation 2). Because the error term has several components, this model is often referred to as the error- component model. Again, we assume th"p , = 0 for all t. It is clear that the presence of pi produces a correlation among residuals of the same cross-sectional unit, though the residuals from different cross- sectional units are independent. Therefore, the least-squares estimate of y (irv) is not efficient, although 7  It is stlll unbiased and consistent. In the case of correlated errors, the generalized-least-squares (GLS) estimator is the BLUE estimator. Given the GLS estimate of y (ia,|), we can recover estimates of the individual cross-sectional unit's intercept yj from the residuals. Following Schmidt and Sick!ls (1984), if we define the residuals as 4 = Yb - X;, j0As, we can estimate Az by the mean, over time, of the residuals for the individuai cross-sectional unit i (10) (IM E 4 In our production model, this estimate will represent technical efficiency at the firm l_vel in a random-effect model. Fixed versus random effects models: How can we decide whether to assume fixed or random fim-specific effects? The GLS estimation, although being more efficient than the within estimation when N is large and T is small, requires the assumption of uncorrelatedness between the error term pi and the regressors. If the firm-specific TFP is correlated with input choices, the estimated regression coefficients will be biased and inconsistent. On the other hand, the advantage of the covariance model is that it protects against a specification error caused by such a correlation, but its disadvantage is a loss of efficiency because of the increased number of parameters to be estimated. Following Hausman (1978), we can test the null hypothesis that no such correlation exists [H.: EAjX'1) = 01, in order to assess thIi appropriateness of using a random-effect model. 8  m. TRADE POLICY IN MOROCCO Since 1983, the Moroccan government has been pursuing trade liberalization measures, within the framework of the structural reform, aimed at gradually reducing the and-export bias and rationalizing the incentives to import substitution. There are basically three major import regimes in Morocco: imnport taxes, quantitative restrictions, and reference prices. The import tariff is the most important taxation instrument for protection from fbreign competition and a significant source of tax revensie There are five individual taxes on imports: the customs duty, the special import tax, the stamp duty, the value added tax, and the excise tax. The customs duty is considered the major fiscal instrument of protection and is levied on the c.i.f. value of the imported goods for domestic use. Prior to the liberalization in 1983, the customs duty was subject to a wide variation both across and within sectors. In 1988, the nraximum rate declined to 45%, with 26 levels. The customs stamp tax is levied at 10% of the sum of all other import taxes administered by customs. Although it is applied uniformly, it magnifies the protective effect of both customs duty and special import tax. The special import tax (SIT) is a uniform tariff levied on the c.i.f. value of imports. In 1988, the SIT and the customs stamp tax were replaced by a Fiscal Levy on Imports (Pr6lbvement Fiscal sur les Importations or FF1), applicable in principle to all Moroccan's imports at the rate of 12.5% of the c.i.f. value. Contrary to the declining maximum tariff trend observed since 1983, this entailed an increase over the sum of the two abolished taxes. Although the intentioE wvas to generate additional fiscal revenue rather than to provide protection, in effect it also confered protection. The authorities proposed uniformity of rates in order to avoid discriminatory incentives. However, there are in fact numerous exemptions from the PFI (in 1988, over one-fourth of all imports were exempt from the PFI). The value-added tax is levied on the c.i.f. value of imports inclusive of customs duty and the PFI tax and is neutral in terms of resource allocation. The excise tax is levied by customs at the port of entry for a limited number of products (primarily petroleum, petroleum products, sugar and beer). These two 9  taxes cannot be regarded as trade policy instmments, as they apply regardless of the origin -domestic or foreign- of the goods and do not create a wedge between domestic production and imports. Next, consider the role of quantitative restrictions (QRs). They were regarded in the past as the principal instrument of domestic protection but were significantly reduced following the establishment of a generalized control of imports in March 1983. An annual General Import Program classifies goods by tariff line into three lists: goods in list A which can be freely imported without prior authorization, goods in list B which necessitate a prior authorization to be imported, and goods in list C for which imports are prohibited except in special circumstances. In 1986, list C has been formally abolished. Moreover, since 1983, products have steadily transferred from list B to list A which represented, in 1988, 81.8% of the imported products (six-digit CCCN tariff codes) as opposed to 67.6% in 1984 (Table la). Nowadays, import licenses for list B goods are almost automatically granted and the authorities consider that by 1992 list B would also disappear. Finally, there is the system of reference price which is, in principle, intended as a safeguard against dumping and unfair trading practices by foreign producers. Reference prices are limited to 367 tariff headings (mainly ceramic tiles, end-of-series and second-hand clothing, used auto-parts). They are used to alleviate the concerns of domestic producers about the liberalization of QRs. However, there are questions arising about the reference prices being actually binding. Despite the liberalization effort, the Moroccan economy is still far from being an open economy. Simply looking at the share of restricted imports and the average tariff rates is misleading and actually exaggerates the extent of the liberalization. First, the share of domestic protduction whose competing imports are subject to licensing is a more meaningfiz measure of protection. Indeed, although the share of imports which require an import license (List B) dropped to 12.7% in 1988, 40% of the value of industrial production is stil protected by import licenses. With import substitutes (which are calculated