Journal Issue

South Africa: Selected Issues

International Monetary Fund
Published Date:
March 2000
  • ShareShare
Show Summary Details

IV. Trade Liberalization and Productivity in South Africa56

A. Introduction

115. The purpose of this chapter is to examine the empirical relationship between trade and total factor productivity (TFP) in South Africa,57 where the hypothesis is that enhanced trade in recent years has improved the efficiency of the economy. The study is important from a policy perspective, as trade liberalization constitutes an important element in the government’s efforts to boost the underlying supply capacity of the economy. In addition, the study briefly reviews the research literature in this area, and discusses whether and how the South African experience differs from that of other countries.

116. South Africa affords the possibility of a rich case study on account of the substantial variation in the degree of openness over time (owing both to the lifting of external sanctions and trade liberalization)58 and in trade policy orientation and productivity performance across sectors. The availability of disaggregated data—on capital stock, employment, and trade policy variables—-permits such questions to be examined in a thorough and comprehensive manner. Thus, a distinctive feature of the paper is that the relationship between trade and TFP growth is examined from both a time-series and a cross-sectional perspective.

117. The main finding of this study is that there is a significant positive relationship between trade and TFP growth both over time and across sectors in South Africa. The relationship holds after controlling for a set of other determinants of TFP growth and is robust to various potential statistical problems.

118. The rest of the paper is organized as follows. Section B briefly reviews some earlier studies related to trade and growth. Section C discusses methodological issues and describes the data, while the results are presented in Section D. Section E offers some concluding observations.

B. Previous Research

119. In theoretical models, the impact of trade liberalization on economic growth is either absent or ambiguous. In a conventional neoclassical growth model, trade does not affect the equilibrium or steady state rate of output growth because, by assumption, growth is determined by an exogenously given technological progress.59 In two-sector models of this kind, trade policy affects the allocation of resources between sectors and, hence, the steady state level of savings and capital accumulation. This relationship can have a one-off effect on the steady state level of output (which can be positive or negative depending on how savings and capital accumulation are affected by trade policy), but not on the long-run rate of growth. Nevertheless, even in the neoclassical model, trade policy can have a transitional impact on growth as the economy converges toward the steady state.60

120. However, in endogenous growth models, the impact of trade liberalization on output growth can be positive or negative, hinging on model-specific assumptions. Increased trade per se can have a number of generalized positive impacts.61 For example, trade enables a country (i) to employ a broader variety of intermediate goods and capital equipment, which could enhance the productivity of its other resources; (ii) to acquire technology developed worldwide, especially in the form of embodied capital goods; (iii) to produce and consume a greater variety of goods; and (iv) to improve the efficiency with which resources are used and thereby help to change market structures, reduce markups, and impart dynamic efficiency benefits. However, as emphasized by Rodriguez and Rodrik (1999), the impact of trade policy changes cannot be unambiguously signed. If the resource allocation effects of trade policy changes promote sectors or activities that generate more long-run growth, the impact is positive, and negative otherwise. The question is then really an empirical one of determining the impact of trade policy in specific cases.

121. The empirical evidence on trade and economic growth includes cross-country as well as within-country studies. The first set of studies has focused on the direct impact of trade on growth in output or in TFP,62 and the broad conclusion is typically that increased trade has a positive impact on economic growth. However, Rodriguez and Rodrik (1999) have questioned the results arguing that the studies (i) examine only whether openness, defined in terms of outcomes, helps growth, rather than whether more liberal trade policy helps growth, and (ii) do not incontrovertibly support the conclusions because they mismeasure trade policy, or because the trade policy variable employed actually is picking up other effects, such as macroeconomic stability or regional dummies.

122. The within-country studies are based on either plant-level data or industry-level data.63 Although it is difficult to summarize the results of this strand of literature, it indicates that the causal link between trade and TFP is less evident in the data. For example, Bernard and Jensen (1998) suggest that, while trade orientation and efficiency are correlated, the causation appears to run from the latter to the former, in the sense that efficient firms tend to self-select into export markets rather than that openness leads to increased efficiency.

C. Methodology and Data

123. As indicated above, some of the empirical cross-country (or cross-sectional) studies have focused on the determinants of growth in TFP rather than in real GDP. The advantage with such an approach is that there is a stronger presumption that growth in TFP is positively related to trade. As discussed above, trade policy might also affect factor accumulation, but in ways that are theoretically ambiguous. Therefore, a study focusing exclusively on output growth would be unable to isolate and capture the effects working through increased efficiency.

124. In addition to using different measures of trade policy to explain fluctuations in TFP growth, previous studies have included various factors that are assumed to be conducive to technological development. These include, for example, investment in machinery and equipment as a share of total investment, research and development (R&D) activities, measures of human capital, terms of trade developments, macroeconomic stability, efficiency of the domestic financial system, and other institutional variables.64 In the current study, we followed a fairly eclectic and pragmatic approach in narrowing the possible determinants of South Africa’s TFP growth. Parsimony in the choice of explanatory variables was also dictated by the relatively small sample size.

Data used in time-series analysis

125. The time-series variations in the data were examined for the period 1971-97 (Figure 7).65 Two measures of TFP, based on alternative approaches to measuring the factor shares, were used (see Fajgenbaum and others (1998)): one calculates these shares using the national income accounts, while the other (TFP-alt) employs the methodology developed in Sarel (1997).66 Because the latter approach yields consistently smaller capital shares than the former and because capital growth exceeds labor growth, the TFP series resulting from the Sarel methodology is at a consistently higher level than the series based on the national income accounts. In terms of movements over time, however, the two series are fairly similar.

Figure 7.South Africa: Time-Series Data, 1971-97

(Levels (solid lines) on left-hand scale; first-difference (dashed lines) on right-hand scale)

Source: South African Reserve Bank, Quarterly Bulletin and Fund staff estimates.

126. Openness was measured as the ratio of the sum of real imports and real exports of goods and nonfactor services to real GDP.67 The use of this variable is open to the Rodriguez and Rodrik (1999) critique that it measures an outcome and, hence, may not have policy implications. The preferred estimation strategy in this view would be to use direct measures of trade policy. However, it is difficult to compute a reliable series of “trade policy” over the sample period, especially because of the pervasiveness of nontariff barriers until the late 1980s.

127. Time series data for R&D in South Africa are not easily available. However, following DeLong and Summers (1991), the share of investment in equipment and machinery in total investment was used as the proxy for technology. Insofar as South Africa does not undertake significant amounts of R& D activity, it is expected that the bulk of the R&D is embodied in capital equipment, especially that imported from abroad. By looking at total investment in machinery and equipment, the specification implicitly aggregates R&D undertaken at home and abroad and assumes that the two have similar effects on TFP. An alternative approach that could have disentangled the effects of foreign and domestic R&D would have been to use separate measures for domestic and imported capital goods (or even construct an imported R&D variable, a la Coe, Helpman, and Hoffmaister (1997)), but this was difficult to do because of the absence of data on imported capital goods for the entire sample period.68

128. We also tried alternative specifications that included a proxy for human capital, but this variable was dropped subsequently as the proxy was likely mismeasured.69 Similarly, exogenous influences, such as terms of trade developments and the aggregate capital-labor ratio, were initially included in the analysis, but they did not turn out to be important. While recent work in explaining growth in East Asia has focused on the role of the financial sector and the efficiency of its intermediation, this aspect was not explored as it seemed less important in the case of South Africa, which has had well-developed and well-regulated financial institutions for a long time70

Data used in cross-sectional analysis

129. The cross-sectional analysis is based on pooled data for the years 1990-94 and 1994-98 for 24 manufacturing industries (defined at the International Standard Industrial Classification three-digit level). TFP growth was defined in the same way as in the time-series analysis, with the nominal factor shares for each sector—obtained from industry-specific data—used to weight the growth in factors (see Appendix for further details). The trade variable (Tariff) is a policy variable, namely, the sum of all import charges (tariff and import surcharge) for each sector. Data were available for the years 1990, 1994, and 1998, although for three sectors (textiles, clothing, and motor vehicles) the announced tariffs for 2002 were used, rather than the actual 1998 tariffs, in order to capture any forward- looking behavior.71

D. Results

Time-series evidence

130. The time-series properties of the variables were analyzed before any regressions were run. The relatively small number of observations implies that traditional nonstationarity tests do not have great power, especially when several lags are included in the models. Nevertheless, the (augmented) Dickey-Fuller tests indicate that total factor productivity (TFP), share of machinery and equipment investment in total investment (Machlnv), and openness (Open) are all integrated of order 1 (see Table 9); the first differences of TFPand Machlnv appear to be stationary, while the first difference of Open appears to be trend stationary.72 Given these nonstationarity results, the long-run relationships among the variables was estimated using the co-integration tests proposed by Johansen (1988) and Johansen and Juselius (1990).

Table 9.Augmented dickey- Fuller Tests of Unit root,1971/97
Levels (max four large)First differences (max four lags)Additional Regressors
VariableObs.Lags 1/t-value 2/Obs.Lags 1//-value 2/
TFP220-0.16210-3.70Constant and trend
TFP-alt2200.06210-3.60Constant and trend
Open2221.43211-5.23*Constant and trend
Machlnv221-3.38211-3.84*Constant and trend
Capacity224-4.64*214-4.80*Constant and trend
Levels (no lags)First Differences (no lags)Additional Regressors
VariableObs.Lagst-value 2/Obs.Lagst-value 2/
TFP2600.23250-3.95*Constant and trend
TFP-alt2600.38250-4.04*Constant and trend
Open2600.37250-4.47*Constant and trend
Machlnv260-2.65250-4.28*Constant and trend
Capacity260-2.68250-3.72*Constant and trend
Note; See Appendix for difinitions of variables

The lag lenght was chosen by using the Schwarz Bayesian Criterion assuming a maximum of 4 lags

The t- value is the test statistic from the (Augmented) Dickey-Fuller test; *indicates rejection of the null hypothesis of nonstationarity at the 5-percent significance level.

Note; See Appendix for difinitions of variables

The lag lenght was chosen by using the Schwarz Bayesian Criterion assuming a maximum of 4 lags

The t- value is the test statistic from the (Augmented) Dickey-Fuller test; *indicates rejection of the null hypothesis of nonstationarity at the 5-percent significance level.

131. The results from the Johansen tests (see Table 10) clearly indicate that there exists one long-run co-integrating vector among TFP, Open, and Machlnv. Moreover, the coefficients of this vector have the expected signs: TFP is positively related to Open and Machlnv73 and all three variables contribute significantly to the co-integrating vector.74 An examination of the speed of convergence coefficients (the alpha matrix) indicates that both TFP and Open are “error correcting,” whereas Machlnvcan be treated as weakly exogenous. The absence of a weak exogeneity result for Open implies that the estimation of a single first- difference equation with TFP as the dependent variable could be problematic. However, as will be discussed below, this apparent absence of weak exogeneity for the openness variable turns out to be a small-sample problem rather than a true simultaneity problem, as various stability tests clearly show that only TFP is error correcting.

132. Hence, in a second step, a single-equation error-correction model was used to examine the annual fluctuations in the variables (see Table 11). The fit of these regressions was remarkably good, considering the small sample size. Moreover, the estimated significant,75 while the estimated coefficient for the lagged error-correction term (EC) is coefficients for both DOpen76 and DMachlnv have the expected positive sign and are negative, as expected, and significant.

Table 11.The Error-Correction Model: TFP Growth and Openness,1971-91
Dependent variable: TFP growth
EC(-1) 1/-0.26-0.25-0.22-0.24-0.22-0.20-0.19
D Tariff-0.17-0.16-0.19-0.17
DW statistic2.
Number of obs.25252525252525
Note: See Appendix for definitions of variables, t- statisitcs in brackets

The error-correction term is derived from the cointergration relation among TFP, Open and MachInv.

Note: See Appendix for definitions of variables, t- statisitcs in brackets

The error-correction term is derived from the cointergration relation among TFP, Open and MachInv.

133. Recursive regressions show that the estimated coefficients in the error-correction model are stable, and no trend breaks could be detected (see top panel of Figure 8). These results tend to support the case for treating the openness variable as weakly exogenous. Indeed, recursive regressions using DOpen as a dependent variable show that the estimated coefficient on the error-correction term is highly unstable and shifts sign over time, indicating that this variable is not really error correcting and thus should be treated as weakly exogenous (bottom panel of Figure 8). Furthermore, when the long-run Johansen equation was estimated using the alternative definition of TFP, weak exogeneity of the openness variable could not be rejected at the 5 percent significance level, and the TFP variable remained error correcting. Taken together, these findings are broadly supportive of the proposition that causation runs from increased openness to higher TFP growth, rather than the converse.

Figure 8.Stability Tests of Error-correction model

(Beta-coefficients 2 standard errors and chow tests)

134. One potentially important problem with the short-run growth regressions is the sensitivity of the measured level of TFP to the business cycle. For example, if it is difficult to adjust the capital stock in the short run, and/or if the labor market is inflexible, leading to labor-hoarding behavior on the part of firms, the measured level of productivity would be higher during booms and lower during recessions. Such an omitted-variable problem could, in turn, generate a simultaneity problem; depending on the magnitude of the export and import elasticities, output fluctuations related to the business cycle could lead to fluctuations in import and export shares of GDP, that is, openness.

135. To deal with this problem, the change in capacity utilization in the manufacturing sector (DCapacity) was added as an independent variable.77 As expected, the estimated coefficient on this variable came out positive and strongly significant, indicating that the growth rate in TFP in a particular year does not necessarily reflect an improvement in technology. Still, the coefficients on DOpen and EC were virtually unaffected by the inclusion of DCapacity. In contrast, the coefficient on DMachlnvdrops sharply and becomes insignificant, suggesting that firms invest less in machinery and equipment during recessions.

136. As emphasized by a number of authors (e.g., Rodriguez and Rodrik, (1999)), openness is somewhat difficult to interpret in a growth regression, as it captures a number of different aspects that contribute to the outcomes; these include not only actual trade policy variables such as tariffs and surcharges, export incentives, and quantitative restrictions, but also variables such as size, geography, foreign demand conditions, transport costs, and preferences. In an attempt to control for some of these aspects, two additional variables were included in the specification: a dummy variable for the period 1985-92, during which South Africa was subject to trade and financial sanctions (Dum8592), and the trade policy variable DTariff, defined as the change in the ratio of import duties and surcharges to import value.

137. These variables are clearly not an ideal measure of the annual change in trade policy in South Africa. Nevertheless, both of their estimated coefficients have, as expected, negative signs, indicating that TFP growth was somewhat lower during the sanctions period and during the years when tariffs were increased. However, the coefficients were in general insignificant or only marginally significant. Moreover, the estimated coefficient on DOpen— which in this context should be interpreted as fluctuations in imports and exports that are not driven by the sanctions or changes in tariff collections—remains positive and strongly significant. Likewise, the coefficient on EC is virtually unaffected by the inclusion of the additional variables.

138. To summarize, the time-series data indicate that there exists a robust long-run relationship among TFP, the degree of openness (measured as imports plus exports over GDP), and the share of machinery and equipment investment in total investment. In addition, annual growth in TFP is positively (and significantly) related to contemporaneous changes in openness, and temporary deviations from the long-run relationship are restored primarily by adjustments in the level of TFP, rather than through changes in imports and exports or in investment in equipment and machinery. The quantitative effects seem to be quite large. The estimated coefficients indicate that a 10 percentage point increase in openness is associated with an increase in TFP by 5 percent in the long-run. Similarly, an increase in the share of machinery and equipment investment of 10 percentage points is associated with an increase in TFP by about 3 percent in the long run. The coefficient on the error-correction term indicates that nearly one-fourth of a given deviation from the long run equilibrium is adjusted within one year by changes in TFP.

Cross-sectional evidence

139. The evidence from the cross-sectional analysis corroborates the time-series results. The focus is on how variations in TFP growth across 24 different manufacturing sectors are related to tariff reductions during the period 1990-98. There are three advantages with this approach: first, the problem in separating true technological process from aggregate demand- related effects is mitigated, as aggregate shocks affect all sectors; second, the number of observations for measuring the long-run effects is greatly increased; and third, the independent variable is actual trade policy (import tariffs) rather than trade outcomes. As mentioned earlier, it is difficult to measure trade policy—both conceptually 78 and empirically—at the aggregate level. However, in the cross-sectional analysis, there is a fair degree of confidence that the trade policy variable is accurately measured: all the charges on imports (surcharges and tariffs) are included; there is no problem stemming from the effect of quantitative restrictions, as those in manufacturing were virtually eliminated before 1990; and it is possible to control for the impact of the export subsidies.79

140. Figure 9 shows the degree of trade protection—as measured by the level of import tariffs—in the 24 manufacturing sectors in 1990,1994 and 1998. In general, tariffs were reduced substantially during the 1990s, but the magnitude of reduction varied significantly across the sectors. Figure 10 shows the TFP growth in the same 24 manufacturing sectors during the 1990s. It can be noticed that the growth rates tended to be higher after 1994, but also that there was substantial variation in the TFP growth rates across the sectors.

Figure 9.South Africa: Traffic Protection, 1990-98

(In percent)

Sources: Industrial Development Corporation of South Africa; and the World Bank

Figure 10.South Africa: TFP Growth, 1990-98

(Annual percentage change)

Sources: Industrial Development Corporation of South Africa; and Fund estimates

141. Table 12 reports the results from regressions of TFP growth on changes in tariffs (DTariff).80 To ensure that this effect is not picking up the impact of other variables, we included four additional variables: the capital labor ratio (CLR), the share of exports in total domestic production (Exportshare), the share of imports in total domestic sales (Importshare), and the initial level of Tariff. The square values of the levels and changes in tariffs were also included in one specification to test for any nonlinear effects. The regression was pooled over the periods 1990-94 and 1994-98, and all regressors, except forDTariff, were measured at their initial level in 1990 and 1994, respectively. A time-dummy for the second subperiod (Dwm9498) was included, to ensure that the results are mainly driven by cross-sectional variations in the data.

Table 12.Trade Liberization and TFP Growth(Pooled results,1990-94 and 1994-98)
Dependent variable: DTFP
Number of obs.48484848
Note: See Appendix for definition variables. OLS estimtions; the t-statistics (in brackets) are based on heteroscedastic consistent convariance matrix (white (1980))
Note: See Appendix for definition variables. OLS estimtions; the t-statistics (in brackets) are based on heteroscedastic consistent convariance matrix (white (1980))

142. The results show that there is a significant negative relationship between changes in tariffs and TFP growth across the manufacturing sectors. This result is robust to the inclusion of a set of variables that are possibly important for TFP growth. Of these variables, only CLR enters significantly, indicating that more capital-intensive sectors tend to exhibit higher TFP growth rates, possibly reflecting a R&D effect of capital investment. The initial level of the tariff, as well as the degree of export orientation of, and import penetration in, a sector, appears to be less important in explaining TFP growth rates.

143. It is also interesting to note that the impact of the tariff changes on TFP growth seems to be nonlinear, with the marginal effect on TFP growth declining as tariff reductions

144. It is also interesting to note that the impact of the tariff changes on TFP growth seems to be nonlinear, with the marginal effect on TFP growth declining as tariff reductions become larger.81 One possible explanation is that this nonlinear impact simply reflects some exogenous limit to TFP growth within the estimated four-year period. These results are illustrated in Figure 11, where the conditional TFP growth is shown on the y-axis. The figure (and the regression results) also shows that the quantitative effect of trade liberalization is sizable; for example, the results indicate that the annual growth rate in TFP was nearly 3 percentage points higher in sectors where tariffs were reduced by 10 percent (or rather, where prices fell by 10 percent because of tariff reductions), than in sectors where tariffs were unchanged.

Figre 11.Conditional TFP Growth and Traffic Changes

145. Table 13 depicts the results of the estimations for the two different subperiods, 1990-94 and 1994-98. The estimated coefficients on DTariff are negative and significant in both subperiods, but the quantitative effect is somewhat stronger in the latter subperiod. In this subperiod, it was also possible to examine the lagged effects of changes in tariffs on TFP growth. However, the coefficients on these lagged variables were small and insignificant. For the second subperiod, tests were also conducted on whether changes in the export subsidy affected TFP growth.82 The result shows that the reductions in the Generalized Export Incentive Scheme (GEIS) could have adversely affected TFP growth but not significantly.83 More important, the inclusion of the export subsidy variable leaves the coefficient of the tariff change variable.

Table 13.Trade Liberization and TFP Growth; Results for Subperiods
Constant-2.03 [-1.96]-0.69 [-0.17]-4.16 [-2.78]-4.34 [-2.21]-3.99 [-1-84]-4.64 [-2.14]-2.22 [-1.05]
Capital Labor Ratio0.00 [-0.04]0.00 [-0.02]0.01 [3.52]0.01 [3.75]0.01 [3.36]0.01 [3.46]0.01 [2.38]
Exportshare-0.19 [-1.95]-0.20 [-1-93]0.04 [0.50]0.04 [0.48]0.04 [0.51]0.04 [0.57]0.03 [0.45]
Importshare0.02 [0.52]0.03 [0.67]0.03 [0.56]0.03 [0-71]0.03 [0.56]0.03 [0.55]0.03 [0.84]
Tariff-0.11 [-0.87]0.07 [0.43]
Tariff-sq0.00 [1.56]0.00 [-0.89]
DTariff-0.43 [-2.57]-0.46 [-1.75]-0.63 [-2.77]-0.51 [-2.02]-0.28 [-4.62]-0.63 [-2.18]-0.74 [-3.57]
DTariff-sq0.02 [2.45]0.63 [1.92]0.02 [1.69]0.00 [0.08]0.02 [1.10]0.02 [2.08]
DTariff (-1)-0.03 [-0.38]-0.05 [-0.17]
DTariff-sq (-1)0.00 [0.07]
DGEIS0.11 [1.39]
Number of obs.24242424242424
Note: OLS Estimtions; the t-statistics(in brackets) are based on a heteroskedastic consistent convariance matrix(see White (1980)).
Note: OLS Estimtions; the t-statistics(in brackets) are based on a heteroskedastic consistent convariance matrix(see White (1980)).

146. The robustness of the results was examined in several ways. First, to test the sensitivity of the results to individual sectors, 24 additional regressions were run in which the observations from a single sector were dropped alternatively.84 The estimated coefficient on DTariffalways remained negative and significant at the 5 percent level, except in one case where it remained significant at the 10 percent level. Second, to test whether the impact of trade liberalization was confined to the import-competing sector, the observations for the two most export-oriented sectors were excluded; again the results remained broadly unaffected. Also, various measures of the extent to which a sector is a net exporter were included in the regressions. In every instance, this variable was added separately (as an alternative to Exportshare and Importshare), but also interacted with DTariff. In no regression did these coefficients turn out to be significant, but the estimated coefficient on DTariff remained negative and significant. Finally, the average capacity utilization of individual sectors was included in the regressions to capture the possibility of idiosyncratic shocks affecting TFP growth differently across sectors. This variable was not significant, and it did not affect the importance of the tariff change variable.

147. While the results appear strong, it is possible that they are driven by the impact of trade liberalization on employment. If this impact is negative, TFP growth may have increased because firms fired less productive workers as tariffs were reduced in order to stay competitive. This is an important issue to clarify in the case of South Africa because employment fell almost continuously during the 1990s; in the manufacturing sector, employment fell in 18 of the 24 sectors examined in this study between 1990-98. However, the data do not lend any support to this hypothesis.

148. Table 14 reports regression results similar to those discussed above, but in which the dependent variable is employment growth, capital growth, or the growth in capital intensity (C/L), rather than TFP growth. There is no evidence for the hypothesis that the tariff reductions are positively related to die employment decline across the manufacturing sectors. In fact, the coefficient on DTariff is negatively signed, indicating that, if anything, employment has fallen less in the sectors where tariffs have been reduced more aggressively.85 However, it can be noted that capital growth is positively related to changes in tariffs. This result suggests that, in sectors that have experienced larger tariff reduction, firms have tended to use the existing capital stock more efficiently rather than adding more capital; to some extent, this might also have had an indirect effect on the relative improvement in TFP growth in these sectors. Taken together, the data reveal that capital intensity increased more in the sectors that remained relatively highly protected (i.e., where tariffs were reduced less) during the 1990s, rather than the opposite.

Table 14.Trade Liberalization and Factor Accumulation(Pooled results, 1990-94 and 1994-98)
Dependent Variable
Employment growthCapital growthGrowth in C/L
Number of obs.484848484848
Notes: See Appendix for definition of variables. OLS estimations; the t-statistics (in brackets) are based on a keteroscedastic consistent covariance matrix (see White (1980)).
Notes: See Appendix for definition of variables. OLS estimations; the t-statistics (in brackets) are based on a keteroscedastic consistent covariance matrix (see White (1980)).

E. Discussion and Conclusions

149. The proposition that trade is beneficial to dynamic efficiency (and not just to static economic welfare) is theoretically ambiguous, and the empirical evidence supporting it has been questioned. In this paper, we tested this proposition for South Africa using an aggregate time-series approach (covering the period 1970-97) and a cross-sectional approach covering the manufacturing sector for the period 1990-98 when South Africa witnessed major trade reform. Both approaches validate the proposed correlation between trade and TFP growth with a remarkably high degree of statistical reliability.

150. It is generally agreed that the South African economy needs to boost its supply capacity—through increases in factor accumulation and in TFP growth. The results reported in this section indicate that trade liberalization has contributed significantly to the growth process through increases in TFP. For example, the openness ratio increased on average by about 3.2 percent per year during the period 1990-97, which, according to our long-run results, contributed to TFP growth of about 1.6 percent per year. The actual annual growth in TFP during 1990-97 was 1.8 percent, implying that increased openness accounted for close to 90 percent of the actual TFP growth in that period.86 The cross-sectional analysis yields similar results. The average price reduction in the 1990s due to the tariff changes was about 14 percent, which translates to higher TFP growth of nearly 3 percent per year. In other words, the typical manufacturing industry exhibited higher TFP growth per year of almost 3 percent because of the trade liberalization.

151. The time-series results pointing to the joint importance of the openness and the technology variable draw attention to two key and complementary channels of influence on the economy’s productivity. While R&D, as embodied in investment in machinery and equipment, augments productivity, it also appears to be important to provide an open or liberal environment in which the gains from R&D can be maximized. A policy corollaiy of this finding could be that emphasis on increasing an economy’s access to foreign capital goods—by, say, selectively liberalizing imports of capital goods—might be insufficient to harness the benefits from technology absorption. By the same token, the results suggest that an open environment needs to be complemented by appropriate avenues for the creation and absorption of technology.

152. The high level of unemployment is, arguably, the most serious macroeconomic problem in South Africa. A concern among policymakers and analysts has been that trade liberalization could aggravate the unemployment problem, as firms might reduce the size of workforce to remain competitive. However, the results in this study indicate that this concern is unfounded; employment has tended to fall less in the sectors where tariffs have been reduced most aggressively.

153. A comparison of the “footwear” and “chemical” sectors vividly illustrates this point. The footwear sector employed 33,000 people in 1990, and was relatively highly protected by an import tariff of 47 percent. The sector remained quite protected during the 1990s, as the tariff was reduced only to 34 percent by 1998. Despite this continued protection, employment fell on average by 5 percent per year to 22,000 by 1998. In addition, TFP fell on average by 1.9 percent per year, and value added fell on average by 5.1 percent per year. In contrast, the sector “other chemical products” employed 64,000 people in 1990, and the tariff was 29 percent. By 1998, the tariff had been slashed to 5 percent. Nevertheless, employment had increased on average by 1 percent per year to 68,000, and, at the same time, the sector had improved its efficiency: TFP increased on average by 1.3 percent per year, while value added grew on average by 2.6 percent per year.

154. The results in this paper are encouraging, but there remains considerable scope for refining and deepening the research agenda. In particular, it would be interesting to explore the impact of liberalization at the plant level. Plant-level data exist for the manufacturing sector (in the form of the manufacturing census) for 1991 and 1993, and those for 1996 are expected to be released in early 2000. This would constitute a rich data set for examining issues related to trade, concentration, and efficiency, as has been done for Turkey (Levinsohn, 1993) and Cote d’lvoire (Harrison, 1994).

155. Although significant strides have been made in opening up the economy, three significant problems remain with the South African tariff regime: its complexity, the continuing high protection for selected sectors, and the enduring problem of discretionary tariff changes. Addressing these problem could further increase the efficiency gains that can be reaped from greater openness.

APPENDIX: Data Description and Sources
A. Time-Series Analysis
A. Time-Series Analysis
TFPIndex of growth in private nonagricultural GDP minus growth in capital and labor, weighted by their respective shares in output; factor shares based on national income accounts.Fajgenbaum and others (1998)
TFP-altIndex of growth in private nonagricultural GDP minus growth in capital and labor, weighted by their respective shares in output; factor shares based on Sarel (1997).Fajgenbaum and others (1998)
OpenReal imports and real exports of goods and nonfactor services divided by real GDP.South African Reserve Bank (SARB), Quarterly Bulletin, 1998
MachlnvShare of investment in machinery and equipment in total gross fixed capital formation.SARB, Quarterly Bulletin, 1998
CLRReal private nonagricultural capital stock divided by private nonagricultural employment.SARB, Quarterly Bulletin, 1998
TariffSum of tariff revenues and import surcharges divided by value of imports.SARB, Quarterly Bulletin, 1998
DTarijfChange in tariff divided by 1 plus initial level of tariff.
DCapacityChange in capacity utilization in manufacturing.SARB, Quarterly Bulletin, 1998
Diwi8592Sanctions dummy taking a value of 1 for the period 1985-92 and 0 otherwise.
TFP growthAnnual average of growth in real value added in a sector minus the factor share weighted growth in capital stock and employment; factor share is in nominal terms.Industrial Development Corporation of South Africa (IDC)
ExportshareExports divided by production (in current prices).IDC
ImportshareImports divided by domestic consumption (in current prices).IDC
TariffSum of tariff revenues and import surcharges divided by value of imports.Belli, Finger, and Ballivian (1993) for tariff data for 1990; IDC for tariff data for 1994 and 1998; and GATT (1993) for import surcharge data.
DTarijfChange in tariff divided by 1 plus initial tariff.
Dwm9498Dummy variable that takes a value of 1 for the period 1994-98 and 0 otherwise.
Generalized Export Incentive Scheme-Export subsidy.Belli, Finger, and Ballivian (1993)
C/LCapital stock in constant prices divided by employment.IDC (1999)
1/ The data refer to the following 24 International Standard Industrial Classification (ISIC) three-digit subsectors within the manufacturing sector: food processing, beverages, textiles, clothing, leather, footwear, wood and wood products, furniture, paper and paper products, printing and publishing, basic chemicals, other chemical products, rubber products, plastic products, glass and glass products, other nonmetallic minerals, basic iron and steel, basic non-ferrous metals, metal products, machinery and equipment, electrical machinery, motor vehicles, transport equipment, and other manufacturing.

The data refer to the following 24 International Standard Industrial Classification (ISIC) three-digit subsectors within the manufacturing sector: food processing, beverages, textiles, clothing, leather, footwear, wood and wood products, furniture, paper and paper products, printing and publishing, basic chemicals, other chemical products, rubber products, plastic products, glass and glass products, other nonmetallic minerals, basic iron and steel, basic non-ferrous metals, metal products, machinery and equipment, electrical machinery, motor vehicles, transport equipment, and other manufacturing.

The data refer to the following 24 International Standard Industrial Classification (ISIC) three-digit subsectors within the manufacturing sector: food processing, beverages, textiles, clothing, leather, footwear, wood and wood products, furniture, paper and paper products, printing and publishing, basic chemicals, other chemical products, rubber products, plastic products, glass and glass products, other nonmetallic minerals, basic iron and steel, basic non-ferrous metals, metal products, machinery and equipment, electrical machinery, motor vehicles, transport equipment, and other manufacturing.

    AndersonJames E. and J. PeterNeary1994“Measuring the Restrictiveness of Trade Policy,”World Bank Economic ReviewVol. 8 (May) 151-69.

    BarroRobert J. and Jong-WhaLee1997“Schooling Quality in a Cross Section of Countries,”NBER Working Paper No. 6198 (Cambridge, Massachusetts: National Bureau Of Economic Research).

    BelliPedroJ.M.Finger and AmparoBallivian1993“Schooling Quality in a Cross Section of Countries,”World Bank Southern African Department Discussion Paper No. 4 (Washington: World Bank)

    BernardAndrew B. and J. BradfordJensen1998“Exporting and Productivity,”NBER Working Paper No. 7135 (Cambridge, Massachusetts: National Bureau of Economic Research).

    BhagwatiJagdish N.1971“The Generalized Theory of Distortions and Welfare,”in Trade Balance of Payments and Growth: Papers in International Economics in Honor ofCharles P.Kindleberger do by Jagdish N.Bhagwati and others (Amsterdam: North Holland) 69-90.

    CoeDavid T.ElhananHelpman and Alexander W.Hoffinaister1997“The Generalized Theory of Distortions and Welfare,”Economic Journal107January99134.

    CollinsSusan M. and Barry P.Bosworth1996, “Economic Growth in East Asia: Accumulation Versus Assimilation,”Brookings Papers on Economic Activity: 2 Brookings Institution135-91.

    CollinsSusan M. and Barry P.Bosworth“Economic Growth in East Asia: Accumulation Versus Assimilation,”Brookings Papers on Economic Activity: 2 Brookings Institution199691135.

    De LongJ. Bradford and Lawrence H.Summers1991“Equipment Investment and Economic Growth,”Quarterly Journal of EconomicsVol.106 (May) pp. 445-502.

    DollarDavid1992“Outward-Oriented Developing Economies Really Do Grow More Rapidly: Evidence from 95 LDCs, 1976-1985,”Economic Development and Cultural ChangeVol.40 (April) 523-44.

    EdwardsSebastian1998“Openness, Productivity and Growth: What Do We Really Know?”Economic JournalVol.108 (March) 383-383.

    FajgenbaumJosé and Others 1998“Growth Accounting,”in South Africa: Selected Issues IMF Staff Country Report No. 98/96 (Washington: International Monetary Fund).

    General Agreement on Tariffs and Trade1993Republic of South Africa: Trade Policy Review (Geneva: General Agreement on Tariffs and Trade)

    GrossmanGene N. and ElhananHelpman1991Innovation and Growth in the Global Economy Cambridge Massachusetts: MIT Press)

    HarrisonAnn E.1994“Productivity, Imperfect Competition and Trade Reform: Theory and Evidence,”journal of International EconomicsVol.36 (February) pp. 53-73.

    JohansenSoren1988“Statistical Analysis of Cointegrating Vectors,”Journal of Economic Dynamics and ControlVol.12 (June-September) pp. 231-73.

    JohansenSoren and K.Juselius1990“Maximum Likelihood Estimation and Inference on Cointegration—With Applications to the Demand for Money,”Oxford Bulletin of Economics and StatisticsVol.52 (May) pp. 169-210.

    LevinsohnJames1993“Testing the Imports-as-Market Discipline Hypothesis,”Journal of International EconomicsVol.35 (August) pp. 1-22.

    NehruVikrammEricSwanson and AshutoshDubey1995“A New Database on Human Capital Stock in Developing and Industrial Countries: Sources, Methodology, and Results,”Journal of Development EconomicsVol.46 (April) pp. 379-401.

    RodriguezFrancisco and DavidRodrik1999“Trade Policy and Economic Growth: A Skeptic’s Guide to the Cross-National Evidence,”NBER Working Paper No. 7081 (Cambridge, Massachusetts: National Bureau of Economic Research).

    SachsJeffrey D. and AndrewWerner1995“Economic Reform and the Process of Global Integration,”Brookings Papers on Economic Activity: 1 Brookings Institution pp. 1-118.

    SarelMichael1997“Growth and Productivity in ASEAN Countries,”IMF Working Paper 97/97 (Washington: International Monetary Fund).

    TyboutJames R.1992“Linking Trade and Productivity: New Research Directions,”World Bank Economic ReviewVol.6 (May) pp. 189-211.

    WhiteHalbert1980“A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity,”EconometricaVol.48 (May) pp. 817-838.

Prepared by Gunnar Jonsson and Arvind Subramanian.

Throughout this paper, the term “trade” encompasses two distinct concepts: the first, trade openness, will refer to trade outcomes, while the second, trade liberalization will denote explicitly the reduction of domestic trade policy barriers.

See Section III for a detailed description of developments in South Africa’s trade regime.

In static models without market imperfections (such as monopolistic market structures, internal and external economies of scale, or other distortions), trade restrictions reduce the level of real GDP (equivalent to welfare when measured at world prices). The presence of imperfections opens up a plethora of possibilities in which the effects of trade policies are typically indeterminate, depending on the prior distortion (see Bhagwati (1971))

The distinction between the transitional path and the steady state is well-defined in theory but less easily applied empirically. If transitions are sufficiently long, the actual data could exhibit growth effects from trade policy changes

See Grossman and Helpman (1991) and the references therein.

See Appendix for data description and sources.

Sard’s (1997) methodology involves computing sector-specific capital shares based on data for a cross section of OECD and developing countries, and then using these to compute the economy-wide capital share. Under this approach, capital shares vary across countries only to the extent of differences in the sectoral composition of output.

As alternatives, we used this ratio in nominal terms, as well as the ratio of exports and imports of goods alone to GDP; the results were similar but less robust.

These data were only available from 1979

The Nehru-Swanson-Dubey (1995) human capital stock series does not cover South Africa. The Barro-Lee (1997) series does cover South Africa but exhibits anomalous movements, which raise doubts about its quality. In private correspondence, the authors agreed that this series required further refinement.

Macroeconomic policy could also have been considered as a possible determinant of TFP growth, but this variable was ignored as it is, in general, more important in influencing capital accumulation than TFP growth (see Collins and Bosworth (1996)).

As explained in Section III, under the Uruguay Round commitments, South Africa announced tariff reductions for these three sectors that would extend to the 2002.

Broadly similar results were obtained when the Johansen procedure was used to test for the order of integration of the variables.

One lag was included in the co-integration models.

Using the alternative measure of TFP (TFP-alt) generated qualitatively the same results (bottom panel of Table 10).

The first lags of the variables were insignificant and therefore dropped.

All variables beginning with the operator “D” refer to the change in the underlying variable.

The level of capacity utilization, a business cycle indicator proposed by the Economics Department of the South African Reserve Bank, and fluctuations in the terms of trade were used as alternative measures. The results were qualitatively the same, in the sense that the estimated coefficients on DOpen and DMachlnv were virtually unaffected by the choice of the proxy for cyclical fluctuations.

There are well-known problems relating to finding a scalar measure that successfully aggregates protection across sectors. One exception is the measure developed by Anderson and Neary (1994), but its data requirements are fairly onerous.

Although data on effective protection are available, they were not used for three reasons: first, the data were based on statutory tariffs alone and did not incorporate the impact of the import surcharges, which varied substantially across sectors; second, the effective protection data series contained a few outliers, which raised doubts about its accuracy; and third, nominal protection has a more natural metric and is, therefore, more easily interpretable.

The variable DTariff is measured as the change in tariff divided by 1 plus the initial tariff and, hence, reflects the change in domestic price owing to the tariff reduction.

More precisely, given the normalization that occurs when calculating DTariff\ the marginal effect on TFP growth tends to decline as the price reductions caused by tariff changes become larger.

The export subsidy remained broadly unchanged during 1990-94

One point on the measurement of the export subsidy should be noted. On the one hand, the subsidy provided effective protection to those sectors that received it; on the other hand, insofar as the subsidy was linked to the use of locally produced inputs, its effect was diluted (on the reasonable assumption that the local content requirement was binding). It is not clear that the manner in which the subsidy is measured adequately captures the latter effect.

Thus, the number of observations dropped from 48 to 46 in these regressions.

The regressions in Table 14 are not structural equations for factor accumulation and should therefore be interpreted with caution. However, even after controlling for variables such as nominal and real wage growth and labor productivity, the basic conclusion with regard to the relationship between employment growth and tariff reductions remains robust.

Real output was virtually flat between 1990 and 1997, as the growth in TFP, together with capital accumulation of 0.9 percent per year, was offset by a reduction in labor input of 2.3 percent per year.

Other Resources Citing This Publication