1. Introduction

Does trade liberalization increase wage inequality between skilled and unskilled workers? If so, through which channel would trade liberalization affect wage inequality? In spite of traditional theorem inherent in the Heckscher-Ohlin-Samuelson (henceforth, HOS) theory, many empirical studies have shown that trade liberalization has increased wage inequality in Latin American and Caribbean countries (henceforth, LACs) during the last three decades. In this regard, ^{Goldberg and Pavcnik (2007)} provide two important explanations based on an abundant body of empirical research. The first line of explanation for these findings focuses on the inconsistency with the assumptions of the hos theory. Previous research notes that contrary to expectations, the unskilled labor-intensive sectors were in fact the ones most protected, prior to trade liberalization, and that they experienced the largest tariff reductions during the trade liberalization period. Therefore, a rise in wage inequality is exactly what Stolper-Samuelson would predict. In contrast, the second line of explanation focuses on within-industry changes. According to past works, one of the main determinants of increases in wage inequality is the increase in demand for more skilled workers within all industries; that is, a skill-biased technological change (henceforth, SBTC). This increase is attributable to “defensive innovation” (^{Wood, 1995}), which indicates that intensified competition from abroad induces firms to take advantage of the existing new technologies that were unavailable prior to trade liberalization. SBTCs also arise from the incorporation of new technologies embodied in cheaper imported capital goods, such as machines and office equipment, or intermediate goods that are complementary to skilled workers.

With regard to the evolution of wage inequality between skilled and unskilled workers in Chile, the first of the LACs to introduce far-reaching trade liberalization, there seems to be little room for argument; that is, there is, in general, a common agreement that wage inequality increased significantly since the mid-1970s, reached its peak in 1987, then decreased substantially up to the first half of the 1990s, once again increased in the mid-or late 1990s, and decreased afterward over the past decade (that is, during 1998-2007). However, the impacts of trade liberalization on the evolution of wage inequality still remain controversial. ^{Beyer, Rojas and Vergara (1999)} and ^{Gallego (2012)} , in a more recent study, try to answer the aforementioned questions using exactly the same employment survey over the same period while employing a similar methodology. However, the findings are quite different. The former study finds that trade liberalization, measured as the volume of trade over GDP (the ratio of exports plus imports to GDP), and the price reductions of unskilled labor-intensive goods widen the wage premium gap between skilled and unskilled workers during 1960-1996.

Therefore, they conclude that the evolution of wage premiums during this period can be understood within the context of the aforementioned inconsistencies in the assumptions of the HOS theory. In contrast, ^{Gallego (2012)} finds that the wage premium in the United States, which is assumed to be a proxy for SBTC in developed countries, increased the relative demand for skilled workers in Chile, while he finds no statistical significance of the volume of trade over GDP and the prices of unskilled labor-intensive goods during 1960-2000. Therefore, he concludes that SBTC caused by technology transfers from developed countries through imported capital is the crucial determinant of the evolution of relative demand for skilled workers.

Furthermore, the impacts of an increase in the supply of skilled workers on wage inequality are also controversial among previous studies. Theoretically, an increase in the supply of skilled workers should push the wages of skilled workers down relative to that of unskilled workers, thereby contributing to wage equalization. ^{Beyer et al. (1999)} find that an increase in the proportion of the labor force comprising university graduates reduces wage inequality. However, ^{Robbins (1994)} finds that an increase in the relative supply of university graduates does not explain the observed wage inequality during 1975-1992. ^{Gindling and Robbins (2001)} also find that education-price effects, that is, wage premiums for more educated workers, dominate the effects of an increase in the supply of education during 1974-1990.

In addition to considering the changes in labor supply, we must take into account other possible explanations, such as labor market conditions, when analyzing the evolution of wage inequality between skilled and unskilled workers (^{Gindling and Robbins, 2001}). Among the labor market conditions, minimum wages, unemployment, and union bargaining should be considered the relevant factors for the Chilean case.² However, the findings of previous studies are also ambiguous. ^{Gindling and Robbins (2001)} find a negative relationship between minimum wages and wage premiums: an increase in the minimum wages contributes to wage equalization, but the correlations are not statistically significant at 10%. ^{Gallego (2012)} too finds a negative relationship between minimum wages and the relative demand for skilled workers but these correlations are not statistically significant at 10%. ^{Gindling and Robbins (2001)} find a positive relationship between unemployment and wage premiums: a decrease in unemployment contributes to wage equalization; however, the correlations are not statistically significant at 10%. Thus, these previous studies find that the evolution of wage inequality cannot be explained by changes in labor market conditions, although they do recognize the potential role that they play. In contrast, ^{Reinecke and Valenzuela (2011)} argue that the weaknesses of unionization and collective bargaining are the key determinants of the limited improvement found in wage inequality after 1990.

We assume that there are at least two important reasons for the afore-mentioned unsettle debate on the distributional effects of trade liberalization. First, the two studies mentioned above, ^{Beyer et al. (1999)} and ^{Gallego (2012)} , cover not only the post-1974 trade liberalization period but also the pre-1973 import substitution industrialization period. The problem with including this period is that Chile had experienced political and economic turmoil during this period, especially from 1970 to 1973; that is, Chile passed through the “socialist-populist Chilean experience” (^{Larraín and Meller, 1991}) era. While various previous studies show that wage inequality was at its lowest level during this period, the determinants of wage equalization comprise the temporal results of distributional policies typically inherent in “economic populism” (Dornbusch and Edwards, 1991); thus, wage equalization should be interpreted as in a separate mechanism from the trade liberalization perspective.

This problem is also applicable in the case of labor market condition variables. For example, a decrease in wage inequality chronologically coincides with temporary increases in real minimum wages; this is one of the typical “economic populism” policies. Thus, the aforementioned negative relationship between minimum wages and wage premiums could disappear if we exclude this period from our study. Therefore, we should only cover the post-1974 period (i.e., the 1974-1989 military government period and the post-1990 democratic government period) when analyzing the impact of trade liberalization on wage inequality; this study extends the analysis to 2007.

Further, we argue that all the trade-related variables used by ^{Beyer et al. (1999)} and ^{Gallego (2012)} are not necessarily appropriate measures of trade liberalization. First, the volume of trade over GDP has flaws as a measure of trade liberalization, as pointed out by ^{Rodriguez and Rodrik (2001)} . In particular, when one considers the specific Chilean context, trade volumes have crucial flaws as a measure of trade liberalization. This is because exports from Chile still depend heavily on a limited number of commodities, including copper; thus, trade volumes can be approximately determined from the price of copper and the real exchange rate, neither of which is related to trade liberalization by itself.³ Second, changes in the prices of unskilled labor-intensive goods occur not only because of trade liberalization, that is, tariff reductions or elimination of non-tariff barriers, but also because of other factors, such as decreases in domestic demand. Therefore, we cannot deny the fact that the prices of unskilled labor-intensive goods are endogenous variables with respect to wage inequality. Third, the channel through which wage premiums in the United States would contribute to SBTC in Chile is ambiguous, although wage premiums in the United States are certainly exogenous variables with respect to wage inequality in Chile.

In this regard, ^{Gallego (2012)} argues that the imported capital from developed countries plays a crucial role in technological transfer. However, he does not show any direct evidence linking technological transfers and imported capital. Therefore, we are rather skeptical of the findings of the two studies and therefore, argue that we should use tariffs, which are the most direct measures of trade liberalization (^{Rodriguez and Rodrik, 2001}). Moreover, tariffs have been not only appropriate proxies for trade liberalization but also exogenously determined with respect to wage inequality in the case of Chile, because Chile applied flat tariffs to most goods since 1979.⁴

Therefore, the objective of this study is to empirically analyze whether trade liberalization increases wage inequality between skilled and unskilled workers and identify the channel through which it would affect wage inequality by using tariffs that are the most direct measure of trade liberalization, and covering only the post-1974 Chilean trade liberalization period. In addition, this paper tries to identify the impacts of the labor supply of skilled workers and labor market conditions on wage inequality, which vary in previous studies. With regard to the analytical methodology, this paper takes advantage of the methodology developed by the relevant labor economics literature, as will be discussed in greater detail in the following section.

This paper is organized as follows. Section 2 describes a stylized theoretical model to analyze wage inequality between skilled and unskilled workers and explains how to apply the model to the Chilean context. Section 3 describes the data and method of estimation. Section 4 provides the findings of econometric analysis. The conclusions are summarized in Section 5.

Source: ^{Ffrench-Davis, Leiva and Madrid (1992)}, ^{Ffrench-Davis (2002)}, and World Integrated Trade Solution (WITS).

Figure 1 Evolution of average nominal tariff rates, 1974-2007 (%) 

2. Model

In this section we describe a stylized theoretical model to analyze wage inequality between skilled and unskilled workers and explain how to apply this model to the Chilean context. This model was presented by ^{Katz and Murphy (1992)} and then developed by ^{Card and Lemieux (2001)} and many other previous studies. This model assumes that wage inequality between skilled and unskilled workers is attributable to the evolution of the relative supply and relative demand for skilled workers. While ^{Card and Lemieux (2001)} developed a model incorporating imperfect substitutability between not only different skill groups but also different age groups, this study uses the simplest model presented by ^{Katz and Murphy (1992)},⁵ because this study uses not wages in themselves but “pure” skill premiums estimated by controlling for observable workers’ characteristics, as will be discussed in greater detail in the next section.

Therefore, we assume the following CES production function for aggregate output Y with two production factors, skilled workers and unskilled workers:

(1)

where Ls_t and Lu_t are the total quantities of skilled and unskilled workers used in production during period t; α_t and b_t are the factor-augmenting technological parameters in period t, meaning that technological changes cause an increase in the productivity of skilled and unskilled workers; and σϵ[0,∞) is the time-invariant elasticity of substitution between skilled and unskilled workers.

If we assume that the wage rate of unskilled workers is equal to the marginal product of unskilled workers and that the wage rate of skilled workers is equal to the marginal product of skilled workers, we obtain the following wage inequality function as the natural log of the ratio of the wage rate of skilled workers to that of unskilled workers:

(2)

where Ws_t and Wu_t stand for the wage rate of skilled and unskilled workers, respectively, in period t. This equation decomposes the evolution of wage inequality between skilled and unskilled workers into the first term, which represents the relative demand for skilled workers, Inatbt, and the second term, which represents the relative supply of skilled workers,  InLstLut . What is relevant from the theoretical assumption regarding the impacts of an increase in the supply of skilled workers on wage inequality can be given as follows:

(3)

Therefore, this model shows that an increase in the relative supply of skilled workers reduces the wage inequality between skilled and unskilled workers.

Since we assume that σ is a time-invariant production parameter, technological change is skill neutral as long as the productivity of skilled and that of unskilled workers increase at the same rate. However, this model usually assumes the existence of SBTC, that is, atbt>1 . Since α_t and b_t are not directly observable parameters, the literature assumes that there is a log linear increase in the relative demand for skilled workers over time coming from technology; that is, the evolution of the relative demand for skilled workers can be captured by a linear time trend (^{Acemoglu and Autor, 2011}). Therefore, we make the following assumption on the first term, which represents the relative demand for skilled workers:

(4)

where t is a linear time trend. If we substitute equation (4) into equation (2), we obtain

(5)

Apparently, this model assumes that SBTC takes place at a constant rate. However, there is no special reason for SBTC to always take place at a constant rate; therefore, we do not have to necessarily abide by the aforementioned assumption. In fact, ^{Autor, Katz and Kearney (2008)} allow the term representing the relative demand for skilled workers to have more flexible specifications; they include not only a linear time trend but also a quadric time trend and a cubic time trend. Moreover, it is natural to assume that SBTC takes place in Chile at a varying rate, because trade liberalization in Chile is very rapid and comprehensive as compared to other developing countries, including the LACs. Therefore, we assume that a linear time trend can capture a SBTC that is not necessarily related to trade liberalization as mentioned in the usual labor economics literature; thus, we include trade-related variables that can capture the SBTC inherent in trade liberalization instead of a quadric or a cubic time trend. Under this assumption, the term representing the relative demand for skilled workers can be described as follows:

(6)

where trade stands for the trade-related variables in period t. We include the average nominal tariff rates of each year as the measure of trade liberalization, as discussed in the introduction. We also separately include the ratio of imports of consumption goods, capital goods, and intermediate goods to each year’s GDP (Consumption M/GDP, Capital M/GDP, and Intermediate M/ GDP) to identify the channel through which trade liberalization would affect SBTC. This is because SBTC is attributable to both import competition in final goods and the incorporation of new technologies embodied in capital and intermediate goods, as discussed in the introduction; thus, we need to identify which channel is relevant to the case of Chile. Therefore, this is the estimation strategy in which we apply this model to the Chilean context. If we again introduce equation (6) into equation (2), we obtain

(7)

In addition, we add possible control variables that could have different effects on skilled and unskilled workers. First, we need to control for labor market conditions, such as minimum wages and unemployment, as discussed in the introduction. Theoretically, if the wages of unskilled workers are affected by the minimum wages while the wages of skilled workers are not, an increase in minimum wages can be expected to increase the wages of unskilled workers; thus, it contributes to a reduction in wage inequality (^{Gindling and Robbins, 2001}). Concerning unemployment, if unemployment requires some additional qualification, an increase in unemployment would be expected to lead to an increase in the wages of skilled workers, thus contributing to an increase in wage inequality (^{Larrañaga, 2001}). Second, we consider the export impacts predicted by the HOS theory: if the non-traditional natural resource-based exports, e.g., fish, fruit, and forestry products, expanding after trade liberalization comprises unskilled labor-intensive goods, these changes would favor unskilled labor, thus contributing to a reduction in wage inequality.⁶ In our empirical analyses, we estimate equations (5) and (7) without and with control variables, as will be discussed in greater detail in the next section.

3. Data and estimation method

In this section, we describe the data used in this study and the method of estimation. In the estimation of equations (5) and (7) mentioned above, the evolution of wage inequality between skilled and unskilled workers,  InWstWut , and that of the relative supply of skilled workers,  InLstLut , are observable variables. Therefore, the most relevant issue in the estimation method is how to calculate the aforementioned two variables.

The data used for calculating the variables come from the Employment and Unemployment Survey for Greater Santiago (Encuesta de Ocupación y desocupación en el Gran Santiago), conducted by the University of Chile. This survey covers the Greater Santiago area, which includes roughly 40% of total Chilean population, and is conducted in June every year. Each survey covers fixed 3,600 households,⁷ and approximately 10,000 and 14,000 individuals during 1974-2007. The data are repeated cross-section, and the sample is fully representative of the Greater Santiago area.

We define “wages” as the net (after tax) remuneration from paid employment excluding the income from self-employment, assets, pensions, or other sources, that is, the sum of wage income, family allowance, bonuses, and allowance for overtime work, eliminating income taxes, the complementary global tax, and other legal deductions.⁸ As such, wages include the impacts of monetary transfers. The samples used in this study comprise the working age (14-65 years) population who report positive income and positive work hours. The sample includes only salaried workers-that is, white-collar (“empleados”) and blue-collar (“obreros”) workers-who work on a full-time basis (more than 30 hours a week). Therefore, self-employed workers-that is, employers and independent workers-are not included. Domestic servants, unpaid family workers, and military personnel are also excluded from the sample because their wages are not likely to be determined by market forces. The samples whose variables are not answered in at least one survey question are also eliminated in advance.

To calculate the evolution of wage inequality between skilled and unskilled workers and the relative supply of skilled workers, we define skilled workers as college equivalents and unskilled workers as high-school equivalents, because there is a notable wage gap between people who completed university education and those who completed only high school in Chile (^{Ffrench-Davis, 2010}). Therefore, the evolution of wage inequality between skilled and unskilled workers and that of relative supply of skilled workers are calculated as the skill premiums between college equivalents and high-school equivalents (wp) and that of the relative supply of college equivalents (relative supply). Then, we apply the methodology of ^{Autor et al. (2008)} to our study to calculate the evolution of wage inequality college equivalents and high-school equivalents.

The methodology is as follows. First, we construct six education categories (workers with up to primary school education, high-school dropouts, high-school graduates, vocational high-school dropouts and graduates, some college,⁹ and university graduates) up to 1997, five education categories (workers with up to primary school education, high-school dropout, high-school graduate, some college, and university graduates) after 1998,¹⁰ and four potential experience categories¹¹ (0-9, 10-19, 20-29, and 30+ years), and then construct 24 or 20 education × experience groups. Second, we estimate the skill premiums for high-school graduates and university graduates from the following wage equation, limiting the sample to males:¹²

(8)

where i indexes an individual; W stands for hourly wages obtained by dividing monthly wages by four weeks of working hours and are then deflated by the national consumer price index (December 1998 = 1); vector educateg represents the aforementioned education categories;¹³exp is years of potential experience;¹⁴ and vector X contains a head of household dummy that is 1 for workers with a head position, a public sector dummy that is 1 for workers employed in the public sector, and 8 industry indicators.¹⁵ Third, using the results of the aforementioned regression, we calculate the predicted log wages for high-school graduates and university graduates evaluated at the relevant experience level (5, 15, 25, or 35 years, depending on the aforementioned experience groups) and at the base categories with regard to all variables in vector X. Finally, we calculate the difference between the predicted log wage for the group of university graduates, InWs^t , and high-school graduates, InWu^t using the average share of total monthly hours worked by education × experience groups in each year as weights.

To calculate the evolution of the relative supply of college equivalents, we apply the methodology followed by ^{Card and Lemieux (2001)} and ^{Ciccone and Peri (2005)} . The model assumes that there are only two kinds of production factors, skilled workers and unskilled workers. Therefore, we need to classify all male and female workers into two groups-that is, college equivalents and high-school equivalents-. In this regard, we calculate the quantities of high-school equivalents, Lu_t , in period t as follows:

(9)

where LtPS LtHSD LtHSG and LtSC are the total monthly working hours of workers who have completed up to primary school education, are high-school dropouts, are high-school graduates, and have attended some college, respectively; W-tPS W-tHSD W-tHSG W-tSC and W-tUG are the average wages of each year for workers who have completed up to primary school education, are high-school dropouts, are high-school graduates, have attended some college, and are university graduates, respectively. Therefore, we assume perfect substitution between the less educated workers up to high-school graduates, and precisely weigh their working hours according to the ratio of their actual hourly average wages to those of high-school graduates. We distribute the working hours of the workers with some college education, which are “split” between college equivalents and high-school equivalents on the basis of their relative wages, as shown in equation (9). The quantities of college equivalents, Ls_t , in period t are calculated in exactly the same way:

(10)

Where LtUG is the total monthly working hours of workers who are university graduates.

Although ^{Gallego (2012)} also weighs the working hours of workers according to their education level, he applies fixed weights to all years: for example, the working hours of high-school dropouts are half of those of high-school graduates. However, the relative wages substantially fluctuated during his analysis period, rendering his assumption unrealistic and his calculations of the relative supply of skilled workers biased.

Regarding the trade-related variables, the data of average nominal tariff rates (Tariff) are from ^{Ffrench-Davis, Leiva and Madrid (1992)} for 1974-1989, and ^{Ffrench-Davis (2002)} and World Integrated Trade Solution (WITS) for 1990-2007 (shown in percentage terms). We do not consider preferential tariffs given the enforcement of numerous preferential free trade agreements that Chile has actively pursued since the 1990s. The data of the imports of consumption goods, capital goods, and intermediate goods are from ^{Banco Central de Chile (2002)} for 1974-1999 and its web page for 2000-2007 (shown in percentage terms). As for the control variables stated earlier, we use the unemployment rate of males (Unemployment) from our own calculations based on data from the aforementioned employment survey (shown in percentage terms), real minimum wages (Real minwage) from ^{Banco Central de Chile (2002)}, and the web page of Biblioteca del Congreso Nacional de Chile ¹⁶ deflated by the national consumer price index (December 1998 = 100). As a proxy for non-traditional natural resource-based exports, we use the ratio of non-copper exports to GDP (Non-copper EX /GDP) from ^{Banco Central de Chile (2002)} for 1974-1999 and its web page for 2000-2007 (shown in percentage terms).

4. Estimation results

Figure 2 shows the evolution of skill premiums between college equivalents and high-school equivalents, and Figure 3 shows the evolution of the relative supply of college equivalents during 1974-2007. The evolution of skill premiums basically coincides with that seen in the previous studies: it increased significantly since the mid-1970s, reaching its peak in 1987; it then decreased substantially up to the first half of the 1990s, and once again increased in the mid-1990s; after that, it slightly decreased, although fluctuating year by year. This is as discussed in the introduction. The evolution of the relative supply of college equivalents basically shows an increasing trend, although this too fluctuated year by year.

Source: Author’s calculations based on the data from Encuesta de Ocupación y Desocupación en el Gran Santiago.

Figure 2 Evolution of skill premiums between college equivalents and high-school equivalents in Greater Santiago, 1974-2007 

Source: Author’s calculations based on the data from Encuesta de Ocupación y Desocupación en el Gran Santiago.

Figure 3 Evolution of the relative supply of college equivalents in Greater Santiago, 1974-2007 

The results of the Augmented Dickey-Fuller (ADF) unit roots tests without and with time trend are presented in Table 1. The lag length is determined by the sequential t test on the highest-order lag, beginning with higher-order lags and stopping when the coefficient is statistically significant. The results show that we cannot reject the null hypotheses of unit roots at the 5% significance level in without and with time trend, except for the relative supply of college equivalents and the ratio of the imports of intermediate goods to GDP, indicating the existence of unit roots in dependent and all independent variables. Therefore, we perform the Engel-Granger cointegration tests in the estimation equations.

Table 1 Results of the ADF unit root tests 

** and * indicate significance at the 5% and 10% levels, respectively.

Table 2 reports the results of estimation equations (5) and (7). The coefficients on the relative supply of college equivalents are not statistically significant at all. Moreover, they show positive in some models-that is, they do not show theoretically the “correct” signs-. Therefore, we cannot reject the perfect substitution between college equivalents and high-school equivalents; we find that an increase in the relative supply of skilled workers does not contribute to wage equalization during 1974-2007. The findings are quite different from both ^{Gallego (2012)} , who rejects the perfect substitution between college equivalents and high-school equivalents during 1960-2000,¹⁷ and ^{Beyer et al. (1999)}, who find that an increase in the proportion of labor force with university graduates reduces wage inequality during 1960-1996, but agree with ^{Robbins (1994)} .

Table 2 Regression results of estimation equations (5) and (7) 

***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.

Note: Numbers in parentheses are Newey-West HAC standard errors.

Regarding trade-related variables, the coefficients on tariffs are negatively significant in models without and with control variables (reported in columns [3], [4] and [8]), indicating that tariff reductions contribute to increases in wage inequality between skilled and unskilled workers. The coefficients on the ratio of the imports of consumption goods and capital goods to GDP are positive but not statistically significant (reported in columns [5] and [6]). The coefficients on the ratio of the imports of intermediate goods to GDP are negatively significant at 10% level (reported in column [7]). However, the significance of the ratio of the imports of intermediate goods to GDP disappears after controlling for tariffs, while the significance of tariffs is found to persist (reported in column [8]). Therefore, the results show that tariff reductions in themselves affect wage inequality, while the observed negative relationship between the ratio of the imports of intermediate goods to GDP and wage inequality is lack of robustness; that is, tariffs correlate with both wage inequality and the ratio of the imports of intermediate goods to GDP. In fact, a regression of the ratio of the imports of intermediate goods to GDP on tariffs yields a positive and statistically significant coefficient of 0.074 (with a T-statistic of 4.46) during 1974-1989.

The null hypotheses of no cointegration are also rejected, as shown in columns [4] and [8]; thus, the relationship between wage inequality and tariffs is not spurious. Moreover, the impacts of tariffs on wage inequality are practically large; a coefficient of -0.0090 presented in specification [8] shows that a 50% tariff reduction would lead to a 0.45-point increase in the log wage difference between college equivalents and high-school equivalents. The results indicate that the explanatory factor of tariff for the total variance of dependent variable (i.e., log wage difference between college equivalents and high-school equivalents) during 1974-2007 is as large as 39.4%, and thus, it is the most important factor among the explanatory variables except for the time trend.¹⁸ A logical examination of these findings would directly show what ^{Beyer et al. (1999)} imply by using the prices of unskilled labor-intensive goods; the unskilled labor-intensive sectors that were protected with the highest tariffs prior to trade liberalization experienced the largest tariff reductions.

In fact, ^{Hachette (2000)} shows that the two unskilled labor-intensive sectors, textiles, and footwear and apparel, which were protected with the highest effective tariffs among 23 industrial sectors in 1975, experienced the largest tariff reductions from 1975 to 1979.¹⁹ Therefore, tariff reductions contribute to increases in wage inequality through the Stolper-Samuelson effects.

A possible explanation for the insignificance of the ratio of the imports of capital and intermediate goods to GDP is the effect of “imports-de-substitution” (^{Ffrench-Davis, 2008}, ²⁰¹⁰) after trade liberalization in Chile. ^{Ffrench-Davis (2008}, ²⁰¹⁰⁾ point out that the abandonment of the import substitution of final consumption goods reduced the imports of intermediate goods and capital goods, such as machinery and equipment, while it increased the imports of consumption goods (particularly non-food goods), leading to bankruptcy in the remaining import substitution sectors, such as the auto-motive industry. Therefore, the findings show that SBTC is not attributable to the incorporation of new technologies embodied in capital and intermediate goods in the case of Chile. Moreover, the findings are unique to Chile in that they are in contrast with some other LACs like Mexico, which experienced a growing importance of trade in intermediate goods; thus, the import catalyzed technological change is the main cause for SBTC.²⁰ Therefore, the findings are quite different from both ^{Gindling and Robbins (2001)} and ^{Gallego (2012)} , who conclude that skill-intensive technologies embodied in imported capital goods caused SBTC in Chile. The findings can be also interpreted as the experience of a country with a weak manufacturing sector and comparative advantage in natural resources after rapid and far-reaching trade liberalization, that is, “premature de-industrialization” (^{McMillan and Rodrik, 2011}).

Another interesting finding is the impact of minimum wages. The coefficients on minimum wages are negatively significant in cointegrated models (reported in columns [2], [4], [5], [6] and [8]). The finding shows that a decrease in minimum wages certainly contributes to increases in wage inequality between skilled and unskilled workers, even if we exclude the “socialist-populist Chilean experience” era from the analysis period; thus, it is quite different from the ambiguous results of the previous studies. This is a reasonable finding, because the minimum wages were sufficiently low under the military government; it is just the same as the per capita poverty line income of an urban area in Chile.²¹ Thus, the finding coincides with the theoretical assumption that if the wages of unskilled workers are affected by minimum wages while the wages of skilled workers are not, an increase in minimum wages would be expected to decrease wage inequality. Therefore, we conclude that the weak performance of labor markets under the military government is also an important determinant of increasing in wage inequality, while the improvement in labor market performances, especially an increase in the minimum wages, contributed to wage equalization during the period of the democratic governments. On the other hand, the coefficients on unemployment are positive, except for column [7]; that is, they are the expected signs, but not statistically significant.

The coefficients on the ratio of non-copper exports to GDP are positive except for in columns [4]. However, one possible explanation for the unexpected positive sign is that this employment survey covers only the Greater Santiago area-an urban area that includes part of a metropolitan region-; that is, only about 1% of all the workers surveyed there were employed in the agriculture, forestry, or fishing sectors, excluding for the natural resource-based manufacturing sectors. Therefore, it is not appropriate to empirically analyze the impacts of natural resource-based exports on wage inequality using this employment survey.

5. Conclusions

This study empirically analyzed whether trade liberalization increases wage inequality between skilled and unskilled workers and identifies the channel through which it would affect wage inequality in Chile. The findings show that tariff reductions contributed to increases in wage inequality by causing price reductions of unskilled labor-intensive goods protected with the highest tariffs prior to trade liberalization. In contrast, we found no evidence that new technologies embodied in capital and intermediate goods caused SBTC. In addition, this study shows that an increase in the relative supply of college equivalents did not contribute to wage equalization. In contrast, the minimum wages did matter-that is, a decrease in minimum wages contributed to increases in wage inequality during the period of the military government, while an increase in the minimum wages contributed to wage equalization during the period of the democratic governments. Therefore, this study found that under the military government, unskilled workers faced two major issues. First, the price reductions of unskilled labor-intensive goods after trade liberalization decreased their relative wages. Second, they were also affected by a squeeze on the real minimum wage due to the government’s repressive labor policy. In this regard, this study shows quite different findings from previous studies by using tariffs, which are the most direct measures of trade liberalization, and by covering only the post-1974 trade liberalization period in Chile.

This study does not deny that trade liberalization would improve the efficiency of resource allocation and enhance productivity thereby contributing to economic growth. There is little doubt that unskilled workers who temporally experienced losses, eventually benefited from the rising average income derived from sustainable economic growth since the mid-1980s. Moreover, active social policies and labor reforms including the increase in minimum wages, while avoiding running a fiscal deficit, under the democratic governments wound not have been carried out without the sustainable economic growth during this period. In sum, this study does not argue that Chile should not have opened its economy.

However, the findings that across-the-board, indiscriminate, and unilateral trade liberalization made unskilled workers employed in industrial sectors previously protected by the highest tariffs certainly temporally worse off could provide the following policy implications: the widening gap between the rich and the poor might lead to inter-group conflicts, which might under-mine the progress of economic liberalization. However, there is no optimal level of wage inequality per se and we cannot judge whether the observed distributional change in itself is economically good or bad. In the case of Chile, various repressive policies under military government seem to play crucial role in continuing with the implementation of economic liberalization in its early stage. Although the targeted social policies, including monetary transfers such as family allowance, partially introduced during that period too might contribute toward alleviating inter-group conflicts, the progress herein was insufficient to improve the situation of low-income groups (^{Marcel and Solimano, 1994}).

Therefore, the findings of this study at least may suggest the following policy lessons for other LACs, which liberalized their economies later in the 1980s and 1990s under democratic governments: they need structural measures to remedy unfavorable conditions or income transfers in order to avoid inter-group conflicts, because the gains from economic liberalization take considerable time to materialize, especially in terms of distributional aspects. At the same time, they need to be implemented without harming economic efficiency. Although one possibility according to our empirical findings is an increase in the minimum wages, this study falls short of identifying what kind of specific measures should have been taken in the Chilean context; accordingly, further research needs to be done from a policy perspective.

Moreover, much work also needs to be done related to empirical area. First, we have not mentioned the reason why the increase in the relative supply of college equivalents does not contribute to wage equalization. One possible explanation is the quality of education. If new private universities established after 1980 are not compatible with traditional universities, our calculation of the relative supply of college equivalents could be biased, although both ^{Gindling and Robbins (2001)} and ^{Gallego (2012)} argue using cohort effects analyses that the difference in the quality of education between the new universities and the traditional universities does not influence the skill premiums for college equivalents. Second, if we were to obtain more detailed data on the components of consumption, capital, and intermediate goods, we could identify more precisely the channel through which trade liberalization affects wage equality in Chile. For example, if we were to classify consumption goods into non-food and food goods and capital goods into machinery and equipment and other goods, we could obtain the precise impact of trade liberalization on SBTCs. However, no such empirical analyses regarding these issues were undertaken in this study; this area will be an interesting subject for future research.