The Political Economy of Tariffs and Growth

Transcription

1 The Political Economy of Tariffs and Growth Nathan Nunn Daniel Trefler October 27, 2004 Abstract We consider what the distribution of trade protection across industries tells us about a nation s ability to successfully control rent-seeking behavior. Does a government fall prey to rent-seekers, who are solely interested in barring foreign competitors, or is trade protection used to promote growth, by shielding growth-enhancing industries during their infancy? To investigate this, we turn to data on international trade protection. Two facts emerge. First, countries that focus on protecting low-skilled manufacturing grow slowly. Second, this effect has little to do with the economics of drawing resources into low-growth industries. Rather, it is a statement about the ability of a nation s political process to control rent-seeking behavior. Thus, the place to look for the effects of international trade protection on growth is less in the economic realm than in the political-economic realm. JELL classification: F13; F43; O19; O24 Keywords: Tariffs; Rent-seeking; Institutions; Growth. We thank Daron Acemoglu, Werner Antweiler, Paul Beaudry, Ig Horstmann, Diego Restuccia, Manuel Trajtenberg and seminar participants at the 2003 CEA Annual Meetings for helpful comments and suggestions. Department of Economics and Institute for Policy Analysis, University of Toronto, 140 Saint George Street, Suite 707, Toronto, Ontario, M5S 3G6, Canada. nnunn@chass.utoronto.ca. Rotman School of Management and Department of Economics, University of Toronto, as well as the Canadian Institute for Advanced Research (CIAR), and the NBER, 105 Saint George Street, Toronto, Ontario, M5S 3E6, Canada. 1

2 1 Introduction At the interface of politics and economics is a view that a nation s long-term growth depends on its ability to successfully control rent-seeking behavior. We consider what the distribution of trade protection across industries tells us about a nation s underlying institutions and their ability to prevent rent-seeking. Does a protectionist government necessarily fall prey to rent-seekers, who are solely interested in barring foreign competitors, or is trade protection used to promote growth, by shielding growth-promoting industries during their infancy? We find that countries that focus protection in high-skilled manufacturing industries exhibit high rates of growth, while those that focus protection in low-skilled manufacturing grow slowly. After further investigation, we find that the relationship between tariff structure and growth has relatively little to do with the economics of drawing resources into low-growth industries. Rather, it is a statement about a nation s ability to control rent-seeking behavior. The relationship captures the much stronger link between a country s domestic institutions, their ability to control predatory behavior, and long-run economic growth. It has long been argued that there are benefits to trade protection in dynamic, skill-intensive industries. 1 However, lobbying and other forms of rent-seeking by special interest groups also influence governments decisions. Governments thus faces a trade-off between the costs and benefits of protection in each industry. 2 If one uses the average skill-intensity of an industry as a proxy for the desirability of tariffs in that industry, then protection that is focused in unskilled industries is evidence of the existence of successful rent-seeking behavior. This may come about for a number of reasons. Governments may place little importance on the well-being of its citizens, or they may care about the welfare of its citizens, but be too weak to stop rent-seeking in the economy. Both phenomenon reflect the existence of poor underlying institutions, which are harmful to growth. Conversely, if tariffs are focused in high-skill industries, then this is evidence of a government that is willing and able to implement policies that control rent-seeking. In our empirical analysis, we first demonstrate the existence of a re- 1 This view is exemplified by the import-substitution-industrialization development strategy used extensively during the 1950s and early 1960s. See Bruton (1999) for a comprehensive description and assessment of this strategy. 2 It is unimportant for our analysis whether or not tariffs are beneficial. We only require that tariffs in skilled industries are either more beneficial, or less harmful than tariffs in unskilled industries. 2

3 markably strong relationship between a country s initial tariff structure and subsequent growth. Although the average tariff rate is uncorrelated with growth, variables measuring the distribution of tariffs across industries are strongly correlated with growth. We perform several sensitivity tests and find the relationship to be extremely robust. The magnitude of the estimated effects are large. A one standard deviation change in our measures of the tariff distribution implies a 1.0 to 2.5% increase in the growth rate. In our sample of countries, the average growth rate is 1.8% and the standard deviation is 1.9%. We find strong support for our explanation of the relationship between tariff structure and growth in the raw data. The bivariate relationships between the tariff structure and the quality of domestic institutions (measured using democracy), and between the tariff structure and the pervasiveness of rent-seeking (measured using corruption) are far stronger than the relationship between the tariff structure and growth. We test alternative hypotheses of the relationship, including the possibility that the tariff structure affects growth by channelling resources into certain industries in the economy. To do this, we estimate the relationship between tariffs and output growth at the industry level. We find that tariffs do have a small effect on industry output, but these effects are too small to account for the magnitude of the effects estimated at the country level. In addition, we find the same relationship between the tariff structure and growth found at the country level is found at the industry level. The aggregate tariff structure is positively correlated with output growth across all industries. The magnitudes of the estimated effects at the industry level are consistent with the magnitude of the estimated effects at the country level. This suggests that the tariff structure affects aggregate growth by increasing the growth rate in all industries. This result is consistent with the tariff structure measuring the presence of poor institutions and rent-seeking, both of which are detrimental to growth in all industries. The paper is organized as follows. In Section 2, we describe our data and estimating equations. Section 3 reports the empirical results. In section 4, we provide support for our explanation of the underlying mechanisms behind the tariff structure, growth relationship. In addition, we directly test our explanation for the relationship against a number of alternative explanations. Section 5 concludes. 3

4 2 Data and Estimation Procedure 2.1 Data Our sample consists of 63 developing and developed countries. Because data availability differs by country, we are unable to consider a single time period for all countries. For 20 of the countries, data are available from 1972 to These countries tend to be the higher income countries of the sample. For 30 of the countries data begin between 1980 and 1983, and for an additional 13 countries, data begin between 1985 to The reason for the imbalance is the unavailability of disaggregated tariff data for developing countries in A list of the countries grouped by geographic region is provided in Table 1. Also shown for each country is the time period for which data are available. Because a detailed description of the data is provided in Appendix A, we only briefly describe the data and their sources here. Income and investment data are taken from the Penn World Tables (PWT) Mark 6.1. The countrylevel human capital measures used are the ratio of workers that completed high school to those that did not. These data are taken from Antweiler and Trefler (2002), and were originally reported in Barro and Lee (1993). The regional dummy variables used are based on the classification used in the Penn World Tables. Industry and country specific tariff data tariff data come from two sources. For the 20 countries with data available for 1972, tariff data are from Antweiler and Trefler (2002). For the other 43 countries, tariff data are from UNCTAD (1994). Production data are from UNCTAD s INDSTAT database. The skill intensity of each industry is measured using the ratio of workers that completed high school to those that did not complete high school in the United States in This measure is from Antweiler and Trefler (2002). 2.2 Estimating Equations We test for a relationship between a country s initial tariff structure and subsequent GDP growth by estimating growth regressions for our sample of countries. Standard specifications for growth regressions estimate the growth rate of per capita GDP as a function of the initial level of per capita 3 Disaggregated tariff data are available for the 30 developed countries in An alternative strategy is to use the 1980s data for all 63 countries, so that the period considered is shorter for this group, but it is consistent across all countries. We have done this, and the results are qualitatively identical to what we report in the paper. 4

5 GDP, as well as initial endowments of human capital and physical capital. Studies often include additional control variables that may affect long-run growth. This approach is problematic for a number of reasons. First, nearly all of the commonly used variables (e.g. black market premium, inflation) are potentially endogenous. Second, there is no clear way of discerning which variables belong in the equation and which do not. This is particularly problematic because studies have found that the estimated coefficients of explanatory variables are very sensitive to the other covariates included in the model. 4 A concern with any growth regression is how the explanatory variables are chosen, whether they are truly exogenous, and how sensitive the results are to the control variables included in the regression. Because of these concerns, the only control variables in addition to initial income, human capital and physical capital that we include are regional dummy variables. These variables do not suffer from the potential endogeneity problem, and this procedure removes the ability to data mine by selectively including control variables that produce the desired results. Our variable of interest is a measure of each country s initial crossindustry tariff structure. We construct three measures of the relationship between tariffs and skill intensities across industries. Our first measure is constructed by partitioning industries into skilled and unskilled groups, and calculating an average tariff rate for each group. We denote the average tariff rates in the skilled and unskilled industries by τ s and τ us, respectively. Our baseline estimating equation is, ẏ c /y c = β 0 + β 1 lny c + β 2 lninv c + β 3 lnhc c +β s lnτ sc + β us lnτ usc + R cβ r + T cβ t + ε c (1) where ẏ/y is the average annual growth of real GDP per capita; y is initial real GDP per capita; INV is initial ratio of investment to GDP; HC is the initial average stock of human capital; R is a vector of regional dummy variables; and T is a vector of two time period dummy variables. One is for countries whose initial period is between 1980 and 1983 and another is for countries whose initial period is between 1985 and The omitted dummy variable is for countries whose initial period is When both skilled and unskilled tariffs are included in the regression equation, the difference between the average tariff rates of the two industries is being estimated. Our second variable, explicitly measures this difference. It is the percentage difference between τ s and τ us : diff ln(τ s ) ln(τ us ). 4 See Levine and Renelt (1992) and Sala-i-Martin (1997). 5

6 To calculate the two measures, it is necessary to partition the industries into those that are skilled and those that are unskilled. How we choose our cut-off is shown in Table 5. The table reports the skill intensity of each industry. The industries are ordered in descending order from the least skill intensive to the most skill intensive industry. Reported in the column farthest to the right is the percentage increase in skill intensity between each industry. In choosing our cut-off, we partition the data where there is a natural gap in the skill intensity measure. From the reported percent changes, it is apparent that there are two large gaps. When moving from non-metallic mineral products to iron and steel, the measure of skill intensity increases by 32%, and when moving from transport equipment to electrical machinery, the measure of skill intensity increases by 27%. 5 Because these two gaps in the skill intensity measures are the largest jumps between any two industries, they are the cut-offs that we use to construct τ s, τ us, and diff. 6 We call the lower-skill partition cut-off a and the higher-skill partition cut-off b. In Section 3.1 we test the robustness of our results to changes in the cut-off. The third measure that we use more finely quantifies how tariffs vary with the skill intensity of each industry. For each country we calculate a correlation coefficient between the measure of skill intensity in industry i and the average tariff rate in industry i, using 17 observations, one for each industry. We denote this measure ρ (τ,skill). A positive value of ρ (τ,skill) indicates that, across industries, tariff rates are positively correlated with the skill intensity of production. 3 Estimation Results We first estimate (1) without any of our measures of the tariff structure included. The results, reported in column 1 of Table 6, are consistent with what has been found in previous studies of the determinants of growth across countries. 7 Initial levels of human and physical capital both lead to higher 5 There is also a third large jump of 26% when moving from footwear to paper products. Although, this break is also a candidate for a partition, the larger cut-off with the jump of 27% was chosen. This is done not only because the gap is slightly larger, but because this choice provides a cut-off further from the cut-off between non-metallic mineral products and iron and steel. This cut-off could also have been chosen and none of the subsequent results would be affected. This is shown clearly in Section Except for the gap between leather and travel goods and wood products, however partitioning the industries here would leave only one industry in the unskilled group. 7 See for example Barro and Sala-i-Martin (1995). 6

7 growth, and the negative coefficient for initial per capita GDP suggests conditional convergence in the sample. In column 2, we included dummy variables that control for the differences in the time period of observation of each country. When the dummy variables are included, the coefficient for initial income becomes insignificant. Most likely, this is because the countries with data beginning in the 1980s are all developing countries, and therefore the dummy variables are strongly correlated with initial income. 8 Column 4 reports the results for the specification which includes the regional control variables. The omitted categories are the most wealthy countries of Western Europe, North America and Oceana. The addition of regional dummy variables causes the coefficients of all explanatory variables to become insignificant. The regional variables provide more explanatory power than standard variables, such as income, investment and human capital. Columns 3 and 5 report the results when the average tariff rate across all manufacturing industries is included in the regression equation. In column 3, the regional variables are not included, while in column 5 they are. In both specifications, the estimated coefficient for the average tariff rate is not statistically significant. This suggests that average tariff rates do not affect subsequent growth. Although this finding may seem at odds with intuition and the commonly held view that poorer, slower growing countries tend to have much higher tariff rates than richer, faster growing countries, the result is completely consistent with previous empirical work that has tested for this relationship. Past studies have failed to find a robust, negative relationship between average tariffs and growth. A number of studies document correlations between the following variables: a number of alternative (sometimes subjective) indicators of openness and growth rates, 9 trade flows and income levels, 10 trade flows and growth rates, and changes in exports and growth rates. 11 However, a negative relationship between average tariffs and growth has not been established. 12 The first study to test for a relationship between average tariffs and growth was Edwards (1992). Using a sample of 20 countries, Edwards 8 The correlation coefficient for ln y and the dummy variables is.75, while the correlation coefficient for ln y and the dummy is See Edwards (1992), Harrison (1996), Lee (1993), Mosley (1999), Sachs and Warner (1995). 10 See Frankel and Romer (1999). 11 See Feder (1982). 12 Even among these studies, it is not clear that the literature points to a clear unconditional positive relationship between trade openness and growth. See Rodriguez and Rodrik (2000) for a summary, as well as Harrison (1996), Mosley (1999) and Pritchett (1996). 7

8 finds a statistically insignificant negative relationship between average tariffs and growth between 1972 to O Rourke (2000) finds that between 1875 and 1914 average tariffs were positively correlated with growth among the European and European off-shoot countries. Clemens and Williamson (2002) find that prior to 1950, the correlation between average tariffs and growth was positive, but since 1950 the correlation has been negative. Vamvakidis (2002) finds that prior to 1970, there is no evidence of a negative relationship between average tariffs and growth. Instead, either a positive relationship or no relationship is found. He finds evidence of a non-robust negative relationship between average tariffs and growth between 1970 and If one uses a measure of average tariffs from Barro and Lee (1994), a marginally significant negative relationship is initially estimated, but the result is not robust. When import duties as a fraction of imports is used as a measure of the average tariff rate, no statistically significant relationship is found. Yanikkaya (2003), looking at the growth of more than 100 countries between 1970 and 1997, finds that initial tariffs (measured as the ratio of import duties to imports) are positively correlated with subsequent growth. Further, the positive correlation is found to be stronger for developing countries than developed countries. Most recently, DeJong and Ripoll (2004) find that the relationship between tariffs and growth depends on the level of development. They find a negative relationship between average tariff rates and growth among the world s richest countries, and a positive relationship among the world s poorest countries. In all, our results are not at odds with the previous empirical work that has tested for the effect that initial average tariff rates have on subsequent growth. Having shown that average tariffs are uncorrelated with growth, we now test for a relationship between the distribution of tariffs and growth. We first look at the raw data. Bivariate plots of growth and diff a, growth and diff b, and growth and ρ (τ,skill) are shown in Figure 1. For comparison, we also report the bivariate plot for growth and τ avg. Shown in each graph are the estimated regression lines and the correlation coefficients. Consistent with estimates from the regression equations, the raw data shows no relationship between average tariffs and growth. 14 For each of the three measures of the tariff distribution, the correlation coefficient is positive and 13 The study does not report the estimated standard errors of the coefficients, only that their t-statistics are greater than one and less than two. 14 In the graph, Hong Kong is omitted because it is an outlier with zero tariffs in every industry. If we do not omit Hong Kong, then the relationship between average tariffs and growth among the other countries cannot be seen clearly in the graph. The correlation coefficient when Hong Kong is included is.20, which is still statistically insignificant. 8

9 statistically significant at the 5% level or lower. In addition, a positive relationship, although weak, is apparent in the data. This provides initial evidence that countries that focus protection in skilled industries experience faster subsequent growth. To further explore this, we estimate (1) with our measures of tariff distribution included in the estimating equation. The results are summarized in Table 7. The top panel reports results when cut-off a is used and the bottom panel reports the results for cut-off b. In columns 1 and 2, we include individually the average tariff in skilled and unskilled industries. When added alone the coefficients for τ s and τ us are insignificant. Like the average tariff rate across all industries, the average tariff rate across skilled or unskilled industries is uncorrelated with growth. 15 In column 3, we include τ s and τ us together. The coefficients for both variables become highly significant. The inclusion of the two variables increases the R 2 from.60 to.69 when cut-off a is used, and from.60 to.70 when cut-off b is used. The explanatory power of these two variables is surprising, especially since detailed regional dummy variables are included in the regression equations. Recall that investment, human capital, and initial income become insignificant with the inclusion of the geography variables, but this is not true for τ s and τ us. In columns 5, τ avg is also added as an additional explanatory variable; τ s and τ us remain significant and τ avg remains insignificant. Column 4 reports the results when diff a and diff b are used as measures of the tariff distribution. The estimated coefficients are positive and statistically significant. We also estimate (1) with ρ (τ,skill) used as our measure of the tariff distribution. The results are summarized in Table 8. With or without regional dummy variables, the estimated coefficient of ρ (τ,skill) is positive and statistically significant. This is true whether or not one controls for τ avg. Partial correlation plots for each of the three measures of the tariff structure are shown in Figures 2 to 4. The partial regression plots for τ s and τ us from the regressions reported in columns 3 of Table 7 are shown in Figures 2 and 3, respectively. The partial correlation plots for diff a and diff b from column 4 of Table 7, and for ρ (τ,skill) from regression 3 of Table 8 are reported in Figure 4. The partial correlation plots show a clear relationship between the tariff distribution and growth. The plots also show that this relationship is not being driven by a small number of influential observations. 15 This is not surprising given the strong correlation between these variables. The correlation coefficient between τ avg and τ sa, τ sb, τ usa, and τ usb is.88,.93,.83, and.93, respectively. 9

10 Overall, the results show that the average tariff rate across industries does not matter for growth, but that the distribution of tariffs across industries does. Countries that focus protection in skilled industries grow faster. The estimated magnitude of this effect is large. An increase of one standard deviation in diff a and diff b increases growth by 0.8% and 0.9%, respectively. The same increase in ln τ s increases growth by 2.5% for cut-off a, and by 2.3% for cut-off b. The same decrease in ln τ us increases growth by 2.4% for cut-off a, and by 2.2% for cut-off b. The estimated coefficients for ρ (τ,skill) imply that a one standard deviation increase in ρ (τ,skill) will increase the growth rate by.9 to 1.1%. These effects are extremely large, given that the mean growth rate in the sample is 1.8% with a standard deviation of 1.9%. 3.1 Robustness and Sensitivity Analysis We test the robustness of the results to changes in the chosen cut-off. The results of this are summarized in Table 9. The first four columns of the table report the same information that was provided in Table 5. Reported are the skill intensities of each industry, and the industries are again ordered from the least skill intense to the most skill intense. The four columns on the right report for each possible cut-off the estimated coefficients and t- statistics for τ s and τ us. For example, the first set of coefficients reported are the coefficient estimates of τ s and τ us when (1) is estimated and the cutoff is between Leather goods and Wood products. Overall, the coefficient estimates are robust to changes in the cut-off. Although the results from partitions that leave few industries in either the skilled or unskilled groupings are insignificant, for sufficiently balanced partitions, the coefficients for τ s and τ us are the expected signs and significant. It is also interesting that the results are strongest when the partition is chosen where there is a natural gap in the skill measure. We also test the sensitivity of our results to changes in the sample of countries. To ensure that the results are not being driven by one or two influential observations, we omit one observation and re-estimate (1). In addition, we repeat this procedure, but omit pairs of countries, re-estimating (1) using 1,953 different sub-samples. The results of these sensitivity tests are summarized in Table 10. The estimated coefficients for the tariff distribution variables are robust to this procedure. We also re-estimate (1) after omitting observations with the largest studentized residuals. 16 Again, the 16 The standard errors are studentized standard errors. That is, they are calculated from the fit of a regression with the observation in question excluded. This methodology allows one to recognize an outlier that strongly influences the estimated regression line, 10

11 results remain robust. We also test the robustness of our results to changes in the industries that are included in the calculations of the tariff distribution variables. We omit one industry at time, recalculate τ s and τ us, and re-estimate (1). The results of this test are summarized in Table 11. This serves not only as a robustness check, but it also provides insight into which industries drive the results. In all but one case the coefficients for τ s and τ us are robust to the exclusion of industries. The one exception occurs when cut-off b is used and industry 140 is excluded. This result is being driven by one outlying observation, Singapore. When industry 140 is excluded, this leaves industries 232, 211, 213 and 249 as high skilled industries. In all four industries Singapore has zero tariffs, leading to a very low τ s relative to τ us for a country with above average growth. In these industries tariffs appear to be set at zero because non-tariff barriers (NTBs) are used instead. 17 This causes Singapore to be a significant outlier. Omitting Singapore results in estimates that are similar to the baseline estimates. As an additional robustness check, we use an alternative measure of cross-industry skill intensity. Because of a lack of data availability, we are forced to use the skill in each industry in the United States as our measure of skill intensity for all countries. If there are factor intensity reversals in some industries, then using the United States factor intensities for other countries is inappropriate. To address this concern, we test whether our results are robust to an alternative measure of skill intensity taken from a developing country. Table 12 reproduces the estimates of Table 9, except that the measure of skill intensity is taken from South Africa in Although this measure is from a more recent time period, because South Africa in 1997 is less developed than the United States in 1972, this measure of skill is more representative of the production technologies used by developing countries. 19 When South African skill intensity measures are used, tariffs in skilled and unskilled industries continue to have differing effects on growth. As before, countries that focus protection in skilled industries grow faster. The estimated coefficients suggest that a one standard deviation increase in τ s or decrease in τ us increases growth by as much as 2.4%. These magnitudes and thus has a small standard error. See Belsley, Kuh and Welsch (1980). 17 The NTB coverage ratios for Singapore are: between 56.3 and 95.9% in industry 211, 18.3% in 232, 2.1% in 249 and 0% in 213. The average NTB coverage ratio among all industries is 12.9%. 18 The measure is taken from Alleyne and Subramanian (2001). 19 Per capita real GDP for the United States in 1972 is $17,700, while per-capita real GDP for South Africa in 1997 is $7,500 (from PWT Mark using rgdpch). 11

12 are almost identical to the magnitudes from the estimates when the United States skill intensities are used. 4 Tariff Structure and Rent-Seeking Our view is that the relationship between a country s initial tariff structure and subsequent growth reflects a much stronger relationship between domestic institutions, the pervasiveness of rent-seeking behavior, and economic growth. Protection focused in low skilled industries may be due to government preferences that are unaligned with aggregate welfare, or a weak government that is incapable of controlling rent-seeking in the economy. Poor institutions and rampant rent-seeking have first-order effects on growth, and it is this relationship that is being captured in our estimates. 4.1 What is the Tariff Distribution Correlated With? In more democratic countries, the government s preferences are more aligned with the welfare of its citizens, and as a result, tariffs should be focused in skill-intense industries. We test for the relationship between democracy and tariff structure using data from the Polity III database. Using the average democracy score from 1970 to 1994, we find that there is a very strong relationship between democracy and our measures of tariff structure. An alternative test is to see whether measures of the tariff distribution are correlated with measures of the pervasiveness of corruption. To measure corruption we use the corruption index from the Political Risk Services. This measure captures the extent to which high government officials are likely to demand and accept bribes connected with special privileges such as import and export licenses, tax breaks, loans, etc. Rather than measuring the government s concern for citizen welfare as the democracy index does, this measure directly measures the existences of corruption and similar forms of rent-seeking in the economy. We show the bivariate relationship between the two variables and diffa in Figure 5. The bivariate relationship between the variables and the measure of the tariff distribution is very strong, and is significantly stronger than the bivariate relationship between the tariff structure and growth, which is shown in Figure 1. The correlation coefficient between diff a and democracy is.61, and between diff a and corruption is.60. The correlation coefficient between diff a and growth is only.33. Although we do not show the bivariate graphs for diff b and ρ (τ,skill), the graphs are similar. The correlation coefficients between diff b and the 12

13 two measures are.54 and.55, and the correlation coefficients for ρ (τ,skill) are.57 and.67. Again, these correlations are significantly higher than the correlation between each of these two variables and growth. For diff b the correlation coefficient is.28, and for ρ (τ,skill) it is.26. Overall, these results suggest that the measures of the tariff distribution are much more closely related to measures of rent-seeking than to economic growth itself. 4.2 Testing Alternative Explanations of the Relationship Our explanation for the relationship between tariff structure and growth is not the only possible explanation. In this section we consider two compelling alternative explanations and test for these. One alternative explanations is that the results are simply picking up a relationship between the industrial mix of a country and growth. If a country s level of protection in an industry increases with the size and importance of the domestic industry, then countries that produce mainly high skilled goods will have higher tariffs in these industries. Conversely, countries that produce primarily low skilled goods will have higher tariffs in these industries. If countries producing high-skilled goods grow faster than countries producing low skilled goods, then one may estimates a spurious correlation between the tariff distribution and growth. A second explanation is that there are real benefits to protecting highskilled industries. Protection of high-skilled industries allows the development of these industries, which through learning-by-doing, linkages and/or the endogenous accumulation of physical and human capital leads to a path of sustained development and high growth. In the following sections we test each of these alternative explanations, and find that neither are not supported by the data Does the Distribution of Tariffs Simply Reflect the Distribution of Production? We test whether our results are being driven by differences in the structure of production between countries. The distribution of production across industries may affect both the tariff distribution and growth, resulting in a spurious correlation between the two variables. Much of the evidence that we have already presented rules out this explanation. In our regressions we control for a number of variables, such as initial income and regional dummy variables, that are highly correlated with 13

14 the level of development and production of high-skilled goods. In addition, when one looks at the partial regression plots shown in Figures 2 to 4, it is clear that the high growth, high tariff differential countries are not the countries with high income levels that primarily produce high-skilled goods. The countries with tariffs focused in skilled industries include Bolivia, Chile, Ghana, Haiti, Mauritius, and Singapore. With the exception of Singapore, all of the countries are low or middle income countries where production is not focused in high-skilled industries. At the other end of the spectrum, the low growth countries with tariffs focused in low-skilled industries include Australia, Austria, Denmark, New Zealand and the United States. All of these countries produce primarily high-skilled goods. We formally test this explanation by controlling for the initial distribution of production. The results of these tests are summarized in Table 13. We report the results when cut-off a is used in the top panel and the results from cut-off b in the bottom panel. Our core specification with skilled and unskilled tariffs is reported in column 1. In column 2, we include log value added manufacturing as a percentage of aggregate GDP. In the third column, we include the value added in the skilled and unskilled industries as a percentage of GDP. In both regressions, variables that control for production structure are statistically insignificant and the estimated coefficients for each tariff distribution variable remains robust. This suggests that the tariff differential is not just a proxy for the distribution of production, which is correlated with growth Do Tariffs Have Real Effects? The second argument is that tariffs in skilled industries have beneficial effects that lead to higher growth. Differences in the average tariff rates between skilled and unskilled industries causes resources to be drawn into skilled industries, and this results in higher aggregate growth. This explanation rests on the existence of two causal relationships. 1. Higher protection in an industry causes increased resources to be channel into the industry, resulting in increased production. 2. Increased production (i.e. faster growth) in skilled industries results in faster aggregate GDP growth. Schematically, these two relationships can be represented as follows: We estimate the first relationship using industry-level production and tariff data. We test whether cross-industry tariffs are correlated with crossindustry differences in the growth rate of production. To do this, we estimate 14

15 cross-industry differences in growth rates aggregate tariff distribution across industries growth the following equation. q ic /q ic = β 0 + β 1 ln ( τic τ c ) + D cγ + D iδ + ε ic (2) where q ic /q ic is the average annual growth rate of real output for country c in industry i; τ ic is the initial tariff rate of country c in industry i; τ c is the initial tariff rate across all industries; D c is a vector of dummy variables that control for country fixed effects; and D i is a vector of controls for industry fixed effects. The results are summarized in Table 14. The first column reports the estimates of (2) without country or industry fixed effects, and the second column reports the estimates controlling for industry and country fixed effects. Because there are a number of observations with extremely high or low rates of growth, we re-estimate (2) after omitting these potentially influential observations. The results are reported in the lower panel of the table. Overall, the results do provide some evidence for a relationship between initial tariff rates and subsequent output growth at the industry level. In both specifications the estimated coefficient for the tariff variable is positive, but statistically insignificant. We include our measures of the aggregate tariff distribution across countries. This is reported in columns 3, 4 and 5 of the table. The estimated coefficients are statistically significant and the effects mirror the findings at the aggregate level. Not only does the distribution of tariffs matter for aggregate GDP growth, but it is also important for efficiency and output growth at the industry level. 20 Countries with tariffs focused in skilled industries experience faster growth in all industries. This is consistent with our hypothesis that the tariff distribution is a proxy for the prevalence of rent-seeking in the economy, since economy wide rent-seeking is expected to cause under-investment and efficiency loss in all industries. The estimated coefficient for industry-specific tariffs range from.001 to.005. This estimated effect is too small to account for the large aggregate effects found in the growth regressions. A one standard deviation increase in ln( τ ic τ c ) increases the industry-specific growth rate by 0.14 to 0.7% per year. The mean growth rate across all manufacturing industries is 9.0%. This can 20 Although we do not report it here, the results similar if one includes the full set of aggregate explanatory variables from (1) as explanatory variables in (2). 15

16 be compared to the results found in the cross-country analysis, where a one standard deviation increases in diff and ρ (τ,skill) increases growth by 1.0% and a simultaneous one standard deviation increase in lnτ s and decrease in lnτ us, increases growth by approximately 2.3 and 2.4%, respectively. The mean growth rate of aggregate per capita GDP is 1.8%. The estimated effects of tariffs at the industry level are too small to generate the effects that tariffs have at the aggregate level. As a second test of whether tariffs have industry specific effects on output, we test whether the differences in the tariff rates in skilled and unskilled industries results in differential rates of output growth in the two industries. To test this, we estimate the following equation, g sc g usc = β 0 + β s lnτ sc + β us lnτ usc + ε c (3) where g s and g us are the average annual rate of output growth across the skilled and unskilled industries. Results are reported in Table 15. In columns 1 and 3, we estimate (3) with and without industry and country fixed effects. In columns 2 and 4, we use diff as our measure of the tariff distribution. Overall, the results do not indicate that initial differences in tariff rates between skilled and unskilled industries result in differential growth rates between the two industries. The coefficient for each variable of interest is statistically insignificant and often of the wrong sign. Last, we test the second relationship: whether the differential growth rate between skilled and unskilled industries is correlated with aggregate growth. We estimate the following equation ẏ c /y c = β 0 + β s g sc + β us g usc + ε c (4) where as before y is real per-capita GDP, and g s and g us are the average annual rates of output growth in skilled and unskilled industries. The results are summarized in Table 16. In columns 1 and 2, we include the output growth rate in the skilled and unskilled industries individually in the regression equation. The growth rate in either sector is positively correlated with aggregate growth. Next, we include the two growth rates together. This is reported in column 3. The results show that the differential growth rate between the two industries is not important for aggregate growth. This is confirmed when we use the differential growth rate, reported in column 4. The coefficient for this variable is statistically insignificant. Overall, the results indicate that production growth in both skilled and unskilled industries is correlated with faster aggregate income growth. The evidence does not support the hypothesis that countries with faster output growth in skilled relative to unskilled industries have faster aggregate output growth. 16

17 As a final test of this alternative hypothesis, we estimate (1) with the differential output growth rate variable g s g us included as an additional explanatory variable. The results are reported in Table 17. In columns 1 and 3, the differential growth rates are included. The estimated coefficients for the variables are insignificant. In columns 2 and 4, we include the tariff distribution measures. The estimated coefficients for these variables are the expected signs and statistically significant. The relationship between the tariff structure and aggregate growth is robust to the inclusion of measures of the differential growth rates in the two industries. After controlling for the possible effect of tariffs on the distribution of production in the economy, the tariff distribution continues to be highly correlated with GDP growth. 5 Conclusions We find that across countries the initial tariff structure is correlated with subsequent economic growth. Countries that focus protection in skilled industries grow faster than countries that focus protection in unskilled industries. The estimated growth effects are very large. A one standard deviation change in the measured tariff structure, results in an increase in the growth rate by 1.0 to 2.5%. We test for the underlying causes of this relationship, and find that the magnitude of these effects cannot be explained by tariffs drawing resources into key growth promoting industries. Instead, we find that a country s tariff distribution reflects the pervasiveness of rent-seeking in the economy. Where rent-seeking is prevalent, tariffs are focused in low skilled industries. It is the much stronger relationship between rent-seeking and growth that is being captured in our estimates. References [1] Alleyne, Trevor and Arvind Subramanian What Does South Africa s Pattern of Trade Say About Its Labor Market? IMF Working Paper, WP/01/148. [2] Antweiler, Werner and Daniel Trefler Increasing Returns and All That: A View From Trade. American Economic Review: 92: [3] Barro, Robert and Jong-Wha Lee International Comparisons of Educational Attainment. Journal of Monetary Economics, 32:

18 [4] Barro, Robert and Jong-Wha Lee Data Set for a Panel of 138 Countries. Harvard University. [5] Barro, Robert and Xavier Sala-i-Martin Economic Growth. New York: McGraw-Hill. [6] Belsley, David A., Edwin Kuh and Roy E. Welsch Regression Diagnostics: Identifying Influential Data and Sources of Collinearity. New York: John Wiley & Sons. [7] Bruton, Henry J.& A Reconsideration of Import Substitution. Journal of Economic Literature, 36: [8] Clemens, Michael A. and Jeffrey G. Williamson Why did the Tariff-Growth Correlation Reverse after 1950? NBER Working Paper [9] Cook, R.D Detection of Influential Observations in Linear Regression. Technometrics, 19: [10] DeJong, David N. and Marla Ripoll Tariffs and Growth: An Empirical Expoloration of Contingent Relationships. University of Pittsburgh, Mimeo. [11] Dollar, David Outward-Oriented Developing Economies Really Do Grow More Rapidly: Evidence from 95 LDCs, Economic Development and Cultural Change, 40: [12] Edwards, Sebastian Trade Orientation, Distortions, and Growth in Developing Countries. Journal of Development Economics, 39: [13] Feder, Gershon On Exports and Economic Growth. Journal of Development Economics, 12: [14] Frankel, Jeffrey and David Romer Does Trade Cause Growth? American Economic Review, 89: [15] Harrison, Ann Oppenness and Growth: A Time-Series, Cross- Country Analysis for Developing Countries. Journal of Development Economics, 48: [16] Heston, Alan, Robert Summers and Bettina Aten. Penn World Table Version 6.1. Center for International Comparisons at the University of Pennsylvania (CICUP). October

19 [17] Lai, Huiwen and Daniel Trefler Gains from Trade with Monopolistic Competition: Specification, Estimation and Mis-Specification. NBER Working Paper [18] Lee, Jong-Wha International Trade, Distortions, and Long-Run Economic Growth. International Monetary Staff Papers, 40: [19] Levine, Ross and David Renelt. A Sensitivity Analysis of Cross- Country Growth Regressions. American Economic Review, 82: [20] Pritchett, Lant Measuring Outward Orientation: Can it be done? Journal of Development Economics 49: [21] Rodriguez, Francisco and Dani Rodrik Trade Policy and Economic Growth: A Skeptic s Guide to the Cross-National Evidence. In Ben Bernanke and Kenneth S. Rogoff (eds.), Macroeconomics Annual Cambridge, MA: MIT Press for NBER. [22] Sachs, Jeffrey and Andrew Warner Economic Reform and the Process of Global Integration. Brookings Papers of Economic Activity, 1: [23] Sala-i-Martin, Xavier I Just Ran Two Million Regressions. American Economic Review, 87: [24] Temple, Jonathan The New Growth Evidence. Journal of Economic Literature, 37: [25] UNCTAD Directory of Import Regimes. Part I: Monitoring Import Regimes. New York: United Nations. [26] Vamvakidis, Athanasios How Robust is the Growth-Openness Connection? Historical Evidence. Journal of Economic Growth, 7: [27] Welsch, Roy E Influence Functions and Regression Diagnostics. In Modern Data Analysis, ed. R.L. Launer and A.F. Siegel, New York: Academic Press. [28] Yanikkaya, Halit Trade Openness and Economic Growth: A Cross-Country Empirical Investigation. Journal of Development Economics, 72:

20 A Data: Description and Sources Per capita real GDP growth (ẏ c /y c ): Calculated as average log change in real per capita GDP. The measure of real per capita GDP used is rgdpch from the Penn World Tables Mark 6.1 (PWT 6.1). The initial period of the measured growth varies by country and is reported in Table 1. For all countries, except for Cyprus, Sierra Leone and Singapore, the end period is Because of the unavailability of the Penn World Tables income data for these three countries, the final year is Initial GDP per capita (y c ): The initial value of real per capita GDP. The measure of real per capita GDP used is rgdpch from the PWT 6.1. Investment to GDP ratio (INV): Gross Investment (Private plus Public) divided by GDP, measured in 1985 international prices, from PWT 6.1. Tariffs by industry (τ ic ): From Antweiler and Trefler (2002) and the UNCTAD s Directory of Import Regimes (1994). Skilled and unskilled average tariff rates (τ s ), (τ us ): Constructed by aggregating the industry level tariff data to form skilled and unskilled average tariff rates. Each country s industry output in 1972 is used as weights. Human capital (HC): From Antweiler and Trefler (2002), with the original data from Barro and Lee (1994). Skill intensity by industry: From Antweiler and Trefler (2002). It is the ratio of workers that completed high school to those that did not complete high school in industries in the United States in Antweiler and Trefler scale the measure so that the skill intensity in the electricity industry is unity. Output and value added by industry (q ic ): From UNIDO s INDSTAT production database. Corruption Index: An average of the months of April and October of their monthly corruption index between 1982 and The index is based on the opinion of experts, and intends to capture the extent to which high government officials are likely to demand special payments and illegal payments are generally expected throughout lower levels of government in the form of bribes connected with import and export licences, exchange controls, tax assessments, police protection, or loans. From Political Risk Services. Democracy Index: The average democracy score over the years The Scale is from 0 to 10, with lower values indicating a less demo- 20