This paper analyzes economic developments in El Salvador during 1990–97. The paper assesses the prospects of the Salvadoran economy in facing the new challenges coming from its reinsertion into the global economy. The paper describes the evolution of trade competitiveness and evaluates the divergence from equilibrium of the real effective exchange rate. It concludes that the exchange rate behavior in the last decade appears to have followed roughly the path predicted by a long-term equilibrium exchange rate model consistent with a current account deficit of about 2 percent of GDP.

Abstract

This paper analyzes economic developments in El Salvador during 1990–97. The paper assesses the prospects of the Salvadoran economy in facing the new challenges coming from its reinsertion into the global economy. The paper describes the evolution of trade competitiveness and evaluates the divergence from equilibrium of the real effective exchange rate. It concludes that the exchange rate behavior in the last decade appears to have followed roughly the path predicted by a long-term equilibrium exchange rate model consistent with a current account deficit of about 2 percent of GDP.

V. Sources of Growth 14

A. Introduction

51. The section evaluates the contribution to growth from structural factors and macroeconomic policies on the basis of an error-correction model for a standard Cobb-Douglas production function for 1970-1995. An area of particular interest is to determine if the acceleration of growth observed in the 1990s was related to the structural reforms undertaken after the advent of peace. The civil war that lasted from 1978 to 1990 represents half of the sample period that constitutes the scope of the econometric analysis. The period preceding the war was one of structural disarray, with segments of the civil society organized in quasi-military movements since the beginning of the 1970s. After the advent of peace, a process of economic reform took place at the same time that the government pursued prudent financial policies based on fiscal discipline while broadening democratic participation as agreed under the peace agreements.

52. For analytical purposes, the analysis differentiates the chaos period (1970-1991) from the reform period (1992-1995). The main findings are: (a) an annual long-run growth rate of 2.1 percent for the chaos period and 5.0 percent for the reform period; (b) for the chaos period, an upward deviation from the long-run trend equivalent to around 9 percent of long-run GDP up to 1981, followed by a downward deviation equivalent to about 16 percent of long-run GDP during the intensification of the civil war; (c) for the reform period, an upward deviation from the long-run trend of 2 percent of long-run GDP as the economy moved quickly towards A05app01tability, and an acceleration of growth based on a massive return of resources to productive use and rebound effects until 1992 as the economy returned to its long-run trend after the war; (d) evidence of a decline in total factor productivity as defined for the Cobb-Douglas production function at an average rate of annual 0.6 percent during the chaos period, that was not reversed in the first half of the 1990s; (e) a significant positive impact of education improvements and a negative one of losses of competitiveness on total factor productivity; and (f) a significant positive impact of positive expectations reflected in changes in the fraction of people willing to undertake superior education, and a negative impact on growth of adverse macroeconomic factors, reflected in the average rate of inflation.

53. In terms of growth patterns, a peculiar characteristic of El Salvador is the sustained decline in the ratio of capital per worker and consequent decline in the ratio of GDP per worker even after per capita GDP started to recover after the war. It appears that the divergence between the evolution of average productivity of labor and per capita GDP during the 1990s is mainly explained by the massive reincorporation of labor to productive activities, although there is evidence of an important role of formalization of formerly underground employment creating distortions in the measurement of labor in the reform period.

54. For the medium term, there is ample room for a positive impact on growth of sustained structural reforms and the consolidation of macroeconomic A05app01tability, mainly through increases in the ratio of capital per worker and improvements in total factor productivity equivalent to 0.6 percentage points of growth per year. This would allow the economy to gradually converge to its long-run growth path, with long-run growth rates increasing at about 6 percent per year. Reversal of policies to try to accelerate this process would be regrettable at a time when productivity gains could start a virtuous trend for the first time in 25 years.

55. The underlying econometric model is presented in part B, while growth patterns in the evolution of labor, capital and output over the sample period are discussed in part C. Part D outlines the econometric results, and part E analyzes a decomposition of the sources of growth and projections for the medium term.

B. Analytical Framework

56. The basic framework for the model is a standard Cobb-Douglas production function with constant returns to scale (standardized by labor units), converted to a logarithmic expression for tractability:

Y=AKαL(1-α)y=YLk=KL(1)
Logy=LogA+αLogkWhere0<α<1(2)

Where Y is output, K the capital stock, L the labor stock, A technology (total factor productivity), and a the long-run contribution of capital to output.

57. An error-correction model on the basis of this equation permits to incorporate information about long-run equilibrium forces and at the same time to allow the data to play a strong role in the specification of the dynamic structure. An additional assumption of weak exogeneity of the explanatory stationary variables, allows to summarize the model in a single equation instead of a system, which permits enough degrees of freedom in spite of sample size limitations. This error-correction model identifies long-run equilibrium relationships among economic variables, which if not exactly satisfied will set in motion forces affecting the variable being explained while stationary variables may affect the magnitude of the deviations.

58. In the case of the production function, it was deemed convenient to test the highest possible number of variables explaining total factor productivity (which was facilitated by the lack of significance for lags of more than one year), in such a way as to determine simultaneously the pure technological parameters of a Cobb-Douglas function together with the structural variables affecting the way factors of production are combined. The magnitude of the deviations from this long-run trend would depend on short-term factors, and it would be expected to decline as the economy reaches more stability.

59. Consistent with this approach, A is allowed to change over time as a function of nonstationary variables z (determinants of total factor productivity), while GDP converges to its long-run path (given by equation 2) at a speed of adjustment reflected in the coefficient δ (0 < δ < 1) (equation 4, below).

LogA=f(z)(3)
dLogy=ζdLogk-δ(Logy[-1]-αLogk[-1]-f(z[-1]))+f(z)+g(w)(4)

Short-run growth will also be affected by fluctuations of z and k (with a short-run contribution given by ζ and other exogenous stationary variables w (macroeconomic conditioning factors). Rival hypothesis about the values for α and ζ can be tested, to evaluate the weight of market frictions and decisions at the firm level (i.e., if in the short-run the contribution of capital is higher than in the long run as a result of more rigidities for shutting down a plant compared to firing workers, for example).

60. To the extent that not all variables affecting total factor productivity can be identified, the inclusion of a trend variable will be tested to account for absent variables. Also, although all long-run variables may share the same structural break with GDP during the war, a level dummy for the period of major deviations from the long-run trend will be tested after trying to incorporate most of the impact of the war in the measurement of capital, if it appears evident that the deviations are a result of a remaining impact of the war.

C. Growth Patterns, Evolution of GDP, and the Use of Inputs

61. To analyze the evolution of output in the long run, it is useful to take as a reference the stylized facts typifying growth as listed by Kaldor, 15 among them: (a) per capita output grows over time, and its growth rate does not tend to diminish; (b) physical capital per worker grows over time; (c) the ratio of physical capital to output is nearly constant; and (d) the shares of labor and physical capital in national income are nearly constant.

62. In the case of El Salvador, the identification of growth patterns implies to overcome difficulties in: (a) identifying capital destruction and distraction to nonproductive uses during the war period; (b) interpreting labor statistics in the face of migration of about 40 percent of the working population between 1975 and 1990; (c) inferring long-run trends based on information available only since 1970; and (d) analyzing overlapping structural patterns brought about by a rapid process of economic reform after the advent of peace.

63. Labor is measured by the number of private contributors to the Social Security Health System, assuming a constant share of about 20 percent in the total labor force. 16 The capital stock was calculated from data on gross capital formation in the national accounts (excluding changes in stocks). 17 Some statistical issues result from the heterogeneity of the indexes that were used: First, labor measures actual employment and the capital stock measures availability. 18 Second, coverage is different as the capital stock incorporates public and private investment while labor estimates correspond only to private sector employment. The former issue was not addressed, as it was considered that the costs in terms of the risk of information loss exceeded the potential benefits of correcting the data, specially with information for half of the sample period affected by a war. The latter was not considered a major problem, as public investment is expected to play a major role conditioning private economic decisions.

64. The main results of the analysis of the evolution of GDP and the use of inputs are:

  • The upward trend in per capita GDP observed in the seventies is interrupted during the war and resumed during the reform period. Nonetheless, GDP per worker has consistently declined, even after 1990. 19 The divergence of trends between these variables in the 1970s and the 1990s is explained by increases in the participation rate (as measured by the share of labor force in total population at working age), which shifted from 24 percent in 1970 to 30 percent in 1980 mainly reflecting the incorporation of a greater number of women into the labor force; subsequently it remained in the range of 20-25 percent for the remainder of the 1980s (war period); and climbed again to about 35 percent by 1995, as a substantial segment of the population was reincorporated to productive uses between 1990 and 1995—employment expanded at 11 percent a year in the period (Figure 1). The decline of the average labor productivity in the reform period should not be surprising considering the massive incorporation of unskilled labor force. The conclusion with respect to labor is then similar to Young’s study for the East-Asian economies: 20 labor productivity growth underperforms per capita GDP growth because of massive transfers of labor into productive activities. 21

  • The reduction in average GDP per unit of labor appears related to the reduction in capital per worker, i.e., capital has not increased over-time as much as employed labor. Even in recent years, while the economy went through a process of recapitalization, capital per worker remains low.

  • The capital-output ratio fluctuates around 1.4 in the 1970s and 1.6 in the 1990s 22 (see Figure 1). Large fluctuations in the 1980s result from GDP falling at rates that reached 10 percent per year. The derived evolution of the capital stock mimics that of gross capital formation and reflects the impact of the war.

  • It appears evident that capital, labor and GDP show the same structural break due to the war.23 For this reason, an econometric analysis is preferred to determine if it is possible to find nearly constant contributions of labor and physical capital to the determination of output, in spite of the sizable impact of the war on per capita output and the ratio of capital per worker that precluded long-term growth to follow the patterns stated by Kaldor. Lack of market prices for most of the period makes the analysis of input shares on national income accounts less relevant.

Figure 1
Figure 1

El Salvador: Evolution of Output, Capital, and Labor

Citation: IMF Staff Country Reports 1998, 032; 10.5089/9781451834703.002.A005

D. Determinants of Growth

65. Following Hendry et al. (1984), an error-correction model is constructed starting from an unrestricted equation that is reformulated to incorporate an error-correction once the relevant variables are identified. The error-correction model allows to differentiate explicitly factors affecting short-term deviations from factors affecting total factor productivity (which is a limitation of growth-accounting procedures). As regards policy implications, this technique serves the purpose of identifying the most relevant determinants of growth in the long run, thereby helping in the design of relevant structural policies to raise sustainable growth. Furthermore, it helps to explain short-run fluctuations from the long-run path of output to the extent that they are related to financial policies. For a case like El Salvador, it serves also as a test to evaluate the quality of the data, to the extent that excessive divergence of results from expectations may indicate severe problems of data construction, which would be a major drawback for a case with a distortion such as a twelve-year war. The expected results based on available information for testing were the following:

  • Positive impacts on total factor productivity of education (with enrollment in high school and superior education as a measure of completion of basic education), integration of Central American markets and terms-of-trade improvements; and a negative impact of the deterioration of competitiveness as measured by the real effective exchange rate. An impact of remittances of an uncertain sign was to be evaluated, resulting from offsetting repercussions of availability of resources vs. Dutch disease effects.

  • A positive impact of improved expectations as measured by the increase in enrollment in superior education; and a negative impact of macroeconomic in stability on growth, with inflation as the main proxy for the quality of macroeconomics (as shown in cross-section studies), 24 and of variables closely associated with war-related patterns, such as the migration rate and life expectancy.

  • Short-term deviations from the long-run path more severe during the war, not the least because of measurement problems resulting from uneven impacts of the war. These deviations are expected to diminish as the economy reaches more A05app01tability.

  • As physical capital and human capital are expected to be correlated, 25 and given that human capital is not included in the specification of the production function, the output elasticity of capital is expected to be close to a joint share of between two thirds and three quarters. 26

66. The list of variables as well as a brief explanation about their properties is provided in Table 1. An illustration of the evolution of the variables is provided in Figure 2. The unrestricted equations in Table 2 (the t values appear in parentheses) show the results for the variables that were significant. From equation 1, the main findings are the following:

Table 1.

Augmented Dickey-Fuller Test of Unit Roots of Variables in the Model

Critical values: 5%=-3.004 1%=3.767; constant included

t-adf

article image

All Log variables were found to be nonstationary, as well as super, and all differentials (ineluding average inflation) were found to be stationary with the exception of DLsuper, but its beta coefficients indicate that it is in all likelihood stationary as well.

As the Fund does not make the calculation for years prior to 1980, the series was completed based on the main four partners for 1970-1979: USA, Germany, Japan and Guatemala, whose trade volume with El Salvador is equivalent to 60 percent of Salvadoran trade.

Table 2.

El Salvador: Econometric Results: Variables Explaining Percent Change (GDP/Labor)

article image
Figure 2
Figure 2

El Salvador: Variables Explanatory of Growth

Citation: IMF Staff Country Reports 1998, 032; 10.5089/9781451834703.002.A005

  • The speed of adjustment to the long-run growth path is about annual 26 percent of the deviations (i.e., it takes some 11 years to reach long-run growth after a given shock).

  • The long-run output elasticity of capital is 0.76 and the short-run 0.86. The null hypothesis that both coefficients have the same value cannot be rejected.

  • The relative impact of education and the real effective exchange rate show the expected sign (positive and negative respectively) on the long-run trend of output.

  • Central American trade shows an apparent negative impact on growth contrary to expectations.

  • Periods of higher enrollment in superior education and periods of lower inflation, are periods more favorable to higher rates of growth.

67. The negative impact of regional integration, may result from either stronger trade-deviation than trade-creation effects or a combined impact on competitiveness in conjunction with the real effective exchange rate. This latter hypothesis is tested by creating an index that weighs the real effective exchange rate using the volume of Central American trade (the main market for nontraditional exports), substituting the real effective exchange rate by this index, and running a new regression. The new unrestricted equation (number 2 in Table 2), shows a positive impact of integration on growth (90 percent confidence) while the competitiveness index keeps the same coefficient of the real effective exchange rate in equation 1 and all other variables in the equation keep the same coefficients and significance.

68. An appreciation of the exchange rate does not necessarily imply a deviation from long-run equilibrium (see Section III), although in the case of El Salvador the prolonged recurrence to a fixed exchange rate may have caused a bias toward appreciating the exchange rate, which may make current account adjustments more costly in terms of output. 27 In addition, structural factors may have disfavored the relative price of tradeable goods against nontradeable goods, with an impact on growth. Moreover, the equilibrium exchange rate may have moved toward a more appreciated value by exogenous factors affecting growth (for example Dutch disease caused by remittances in the 1990s).

69. In general, the coefficients for the determinants of growth may also reflect collinearity with variables excluded from the final equation, and even if these variables were identified it would be difficult to incorporate them without seriously affecting the degrees of freedom of the estimation, given the small sample size. 28 To some extent this has been done deliberately by accepting a coefficient for physical capital that in all likelihood embodies the elasticities for human capital, and by using changes in enrollment in secondary education as a measurement of expectations about the future.

70. In an attempt to minimize the impact of additional unforeseen collinearities, in spite of the overall good performance of the test-statistics, a trend variable as well as a dummy variable were included for the period in which the deviations from the long-run trend based on equation 2 were more severe (not surprisingly, for 1984-1987, the core period of the civil war). As a result, equation 3 is obtained, for which the independent impact of Central American trade becomes much less significant, 29 the coefficient for competitiveness declines substantially and the impact of education appears to be stronger. The coefficient for the quality of macroeconomics reflected in average inflation declines slightly, while a linear restriction imposing the same coefficient for short-run and long-run elasticities of capital is now rejected, with a higher value for the short-run elasticity of capital as expected (0.49 for the long run vs. 0.76 for the short run. De Gregorio finds values within this range for a sample of Latin American countries).

71. The impacts captured by the trend and the dummy variables are negative, significant and relatively modest, which implies that even though there may be variables absent from the equation, their contribution to the explanation of growth is not as high as the variables that were already included. Nevertheless, the inclusion of the trend and dummy variables have a non trivial role in improving the specification of the final equation.

72. Based on the unrestricted equation number 3, the final error-correction model is formulated. The result in terms of the Cobb-Douglas production function are the following:

Y=AK0.4907L0.5093Logy=LogA+0.4907Logk(5)
LogA=5.3072-0.3407Logcompe+0.0426super-0.1681dummy(1983-87)-0.0267trend(6)
dLogy=0.758dLogk-0.25209(Logy[-1]-0.4907Logk[-1]-LogA[-1])+0.13295dLogsuper-0.002084Avginflaction(7)

73. The first two equations express the production function as such. The third equation shows the different identifiable variables that help explain the evolution of total factor productivity (long run equation), and the fourth equation shows the short-run dynamics.

74. The variables that did not show any evident impact on growth in this exercise were terms of trade and remittances. Consistent with results by Sala-i-Martin (1994), 30 neither the terms of trade nor coffee prices relative to different subsets of import prices proved to be significant. Likewise, remittances were not found to have a direct impact on growth. Other variables related to macroeconomics, namely the public sector deficit and the interest rate, proved to have some significance, but much weaker than inflation. Moreover, their inclusion resulted in simultaneity problems.

75. Attempts to introduce the war explicitly into the model in addition to the discount factor to determine the capital stock, using as proxies the migration rate and life expectancy, were fruitless. Explanations for this are: (a) the main explanatory variables share a structural break conditioned by the war; (b) relations may be nonlinear (as for some forms of capital distraction, explained in part C); and (c) the proxies for the war are not only affected by the war itself: surprisingly life expectancy declined long before the civil war started. The migration rate appears to be a better proxy for the war in general.

76. Figures 3 to 5 show the fitness of the final equation, a comparison between actual evolution of output and its estimated long-run path, and the evolution of total factor productivity (with an alternative view smoothed using the Hodrick-Prescott filter). The patterns are in general as expected. A05app01tability of parameters is acceptable (Figures 6 and 7).

Figure 3
Figure 3

El Salvador: Actual vs. Fitted Growth of GDP per Worker

Citation: IMF Staff Country Reports 1998, 032; 10.5089/9781451834703.002.A005

Figure 4
Figure 4

El Salvador: Actual vs. Long-run GDP

Citation: IMF Staff Country Reports 1998, 032; 10.5089/9781451834703.002.A005

Figure 5
Figure 5

EI Salvador: Computed Total Factor Productivity Growth

Citation: IMF Staff Country Reports 1998, 032; 10.5089/9781451834703.002.A005

Figure 6
Figure 6

Recursive estimates (error-correction model)

Citation: IMF Staff Country Reports 1998, 032; 10.5089/9781451834703.002.A005

Figure 7
Figure 7

Recursive residuals (error-correction model)

Citation: IMF Staff Country Reports 1998, 032; 10.5089/9781451834703.002.A005

EI. Sources of Growth in the Period 1990-1995 and Medium-Term Prospects

Growth dynamics

77. The model predicts quite closely the rates of growth observed in the chaos period (average 1.0 percent annually forecasted vs. 1.0 percent observed) and the reform period (average 6.4 percent vs. 6.8 percent observed) (Table 3). As Figure 3 shows, the economy was below their potential during the last part of the chaos period, which allowed an extra impulse to growth coming from the correction term in the nineties (rebound effects from the war), that was basically exhausted in 1993. 31 As a result, the model predicts that for 1996 and 1997 the pace of growth may have decelerated, which actually occurred. 32

Table 3.

El Salvador: Sources of growth

Percentage of Annual Growth Explained

article image

78. Macroeconomic conditions play a significant role affecting growth adversely in the chaos period. Consequently, growth is lower than what the contribution of inputs would explain. A period of negative correction of the rate of growth (1971-1982) is followed by positive corrections but huge downward deviations from the long-run trend (1983-1993). In the reform period, a declining adverse impact of macroeconomics reflected on a reduction of the average inflation rate, contributed to a convergence towards the long-run growth path.

Long-run growth path

79. The estimated annual long-run growth rate of GDP goes from 2.1 percent in the chaos period to 5.0 percent in the reform period. The chaos period shows a negative evolution of total factor productivity, with education offsetting partially the adverse impact of competitiveness. In the reform period, the contribution of total factor productivity to growth is even more negative, reflecting measurement problems and frictions in the reaccommodation of resources to productive use with respect to the impact of structural reforms.33

80. The ratio of capital per worker does not increase until 1995, reflecting the massive reincorporation of labor to productive use analyzed in part C. In both the chaos and the reform periods, capital has a lower impact on growth than labor.

Projections

81. On the basis of: (a) a moderation of the rate of incorporation of new labor force in 1996-2000 convergent to an annual 3 percent a year through 2002; (b) an increase in the rate of capital accumulation of 7.5 percent per annum; (c) further improvements in education; (d) a reduction of the inflation rate to less than 3 percent for 1997-2002; and (e) no further loss of competitiveness; 34 the model predicts an average annual growth rate of 5 percent for 1996-2002, with most of the growth explained by investment and an increase in the availability of capital per unit of labor force. An improvement in total factor productivity equivalent to 0.6 percent of GDP per year would result from education more than offsetting the impact of the accumulated exchange rate appreciation. This positive impact could be higher considering factors that were absent in the explanation of long-run growth in the last 25 years. 35 More important, the long-run rate of growth would increase to around 6 percent toward the end of the decade, rate to which the economy would converge gradually based on the above-mentioned assumptions. Continuous improvement of macroeconomic conditions would minimize deviations from this upward path.

Table 1.

El Salvador: National Income Accounts

(In millions of colones)

article image
Sources: Central Reserve Bank; and Fund staff estimates.

Public expenditure and exports of nonfactor services are reclassified by the central bank from the fiscal accounts and the balance of payments respectively to be incorporated into the national accounts, except for public investment for 1995-1997 at current prices.

Includes changes in inventories.

Table 2.

El Salvador: Components of Aggregate Demand

(Annual percentage change)

article image
Source: Statistical Appendix Table 1.

Public expenditure and exports of nonfactor services are reclassified by the central bank from the fiscal accounts and the balance of payments respectively to be incorporated into the national accounts, except for public investment for 1995-1997 at current prices.

Includes changes in inventories.

Table 3.

El Salvador: Financing of Investment

article image
Sources: Central Bank; and Fund staff estimates.

Includes changes in inventories.

Current account balance excluding official transfers.

Table 4.

El Salvador: Gross Domestic Product by Sector

article image
Source: Central Reserve Bank.
Table 5.

El Salvador: Value Added in the Agricultural Sector

(In millions of colones at 1990 prices)

article image
Source: Central Reserve Bank.
Table 6.

El Salvador: Main Agricultural Products

(Area in thousands of manzanas; production in thousands of quintals; yield in quintals permanzana) 1/

article image
Source: Central Reserve Bank.

One manzana equals 0.699 hectares and one quintal equals 101.2 lbs.

Volume data in thousands of short tons; one short ton equals 2,000 lbs.; yield in short tons per manzana. Each ton of cane yields on average just over 190 lbs. of sugar.

Table 7.

El Salvador: Value Added in the Manufacturing Sector

(In millions of colones at 1990 prices)

article image
Source: Central Reserve Bank.
Table 8.

El Salvador: Electricity Indicators

article image
Sources: Lempa River Hydroelectric Commission (CEL); Ministry of Economy; and Central Reserve Bank estimates.

Includes public lighting.

Reflects the use of electricity by generating plants.

Table 9.

El Salvador: Consumer Price Indices

article image
Source: Central Reserve Bank
Table 10.

El Salvador: Nominal Wages by Sector

article image
Sources: Salvadoran Social Security Institute (ISSS); Ministry of Finance; and Fund staff estimates.

January-October for private sector wages.

Includes nonfinancial public enterprises.

As recorded by the Salvadoran Social Security Institute (ISSS).

Average for the consolidated central government.

Table 11.

El Salvador: Real Wages by Sector

article image
Sources: Salvadoran Social Security Institute (ISSS); Ministry of Finance; and Fund staff estimates.

January-October for private sector wages.

Includes nonfinancial public enterprises.

As recorded by the Salvadoran Social Security Institute (ISSS).

Average for the consolidated central government.

Table 12.

El Salvador: Nominal Minimum Daily Wages 1/

(End of period)

article image
Sources: Ministry of Labor, and Salvadoran Social Security Institute.

In colones per 8-hour day.