Liberalization in Mexico and the determinants of regional
wages.
Richt Dijkstra
Thesis Msc. International Economics & Business Rijksuniversiteit Groningen
1. Introduction
For decades, Mexico was a closed economy with strong trade protection policies. In 1985, it radically changed its policy and initiated a liberalization. With its membership to the General Agreement on Tariffs and Trade (GATT), it started to reduce these protections and altered its strategy towards a more open economy. Tariffs, quotas, and import licensing were reduced and trade started to make its rise. Back in 1994, when Mexico joined the North American Free Trade Agreement (NAFTA), trade intensified and international investments started to increase. Characteristic of this agreement was a fierce increase of foreign direct investment (FDI) and maquiladoras in Mexico coming from the United States (U.S.). Especially the states in Mexico that share a border with the U.S., started to witness a fast and large increase in international operating firms. This led to an uneven development in economic growth in different regions in Mexico.
Today, the wage levels in the Mexican regions differ significantly, and regional inequality is high. The liberalization influences the wage level through different mechanisms. In this paper I will test which determinants have an influence on regional inequality in Mexico since the liberalization in the early 1980s. It discusses the effects of increased trade and reduced trade barriers on wages. The focus, particularly, is on wage differences between the border region and the interior region. Research on this topic argues that liberalization alters the demand and supply of labor. Mexico is a special case in this topic, because it does not follow the pattern predicted by trade theory. Simple trade theory predicts a convergence of wages of skilled and unskilled workers in a developing country as an effect of trade liberalization. In turn, convergence of wages reduces the income inequality. Mexico, contrarily, experienced an overall increase in income inequality. The GINI coefficient, which measures the extent of unequal distribution of income, increased from 46.26 in 1984 to 48.11 in 20061. Remarkable, however, is the fierce increase to 51.89 in 1994 and a decrease after 1994.
1
Overall inequality in Mexico has been studied broadly, that is why this paper discusses the regional inequality, which contributes to the discussion of the effects of liberalization on wage dispersion. In particular, it focuses on determinants of wages in Mexico, which, in turn, gives an insight in factors that influence regional wage inequality. The main data used in this paper is the Mexican household survey Encuesta Nacional De Ingresos y Gastos De Los Hogares (ENIGH) and is conducted by the Instituto Nacional de Estadística y Geografía (INEGI). It covers the 1984-2006 period and supplies data of individuals‟ wages, living locations, education levels and working sectors. This dataset provides the opportunity to investigate the determinants of regional wage differences in Mexico.
The rest of the paper is organized in the following way. Section 2 shows an overview of previous research on this topic. Section 3 lays down the related theory. Section 4 and 5 provide an insight into the data and method used to test the hypotheses. In turn, these hypotheses, are given in section 6, together with the results. Finally, section 7 concludes.
2. Literature Review
Since the early 1980s, Mexico has changed its trade policy dramatically. Previous to the trade reform, Mexico protected its domestic industries through high trade barriers. Before 1985, the year that Mexico announced its membership to the GATT, tariffs and products covered by import licensing where high. The average percentage of tariff protection in 1985 was 23.5% of which the manufacturing sector was most protected: i.e. 50%.2 The average percentage of products covered by import licensing was 92.2%, indicating that almost the whole industry was controlled and protected against foreign imports. The membership to the GATT and the reforms in trade policy reduced these numbers dramatically. The share of products covered by tariffs and import licensing decreased to 12.5% and 19.9% respectively in 1990. In 1994, Mexico continued with its trade reforms towards more openness by entering the NAFTA. Especially for the United States and Mexico, this agreement has caused a large increase in trade and investment between these
2
countries. Mexico‟s total foreign trade, the total value of imports and exports, increased from 26% of GDP in 1985 to 38% in 1994, but after the NAFTA membership this growth intensified strongly, up to 65% in 2000.3
These radical changes in trade policy have affected earnings in Mexico and in particular its regional distribution. Several authors describe an increase in regional disparities in employment and wages after the Mexican liberalization.4 Particularly, they find an increase in economic growth in the border area and overall higher wages in this region compared to the rest of Mexico. Why is it that these trade reforms have heterogeneous effects on different regions?
There is one main explanation to why liberalization has altered the relative positions of the regions in Mexico. That is, trade liberalization has led to a change of the optimal location choice for firms. Previously, Mexico City functioned as the most important economic centre of Mexico. When trade was liberalized, however, a large share of firms and production processes were relocated to the border region, where there is better market access to the United States. Whereas the shares of national manufacturing employment in Mexico City and the border regions were fairly the same in 1985, the border regions‟ shares were more than three times as high as in Mexico City in 2003.5
The preferable border region attracted large capital inflows from abroad in the form of FDI and maquiladoras. Maquiladoras are manufacturing or assembly factories in Mexico who import materials and equipment from the U.S. and export back the final product. These maquiladoras do not have to pay tariff or have other duties on the imported and exported goods. The interaction with the foreign market and the externalities in the new industry centre caused a raise in productivity and efficiency which induced higher wages.6 Furthermore, foreign firms have better technology, which leads to a higher premium paid by these firms. Thus, suggesting that the border region has more foreign-owned firms and FDI, the wages in the border should be relatively higher.
3
Esquivel and Rodríguez-López, 2003
4
Chiquiar, 2003, Feenstra and Hanson, 1997, Hanson, 1997, Faber, 2007
5
Faber, 2007
6
Since the reduction in trade restrictions, the border region has attracted firms because of its superior location: i.e. near the U.S. market. Several authors note benefits of the border region in terms of location strategy.7 A location near the U.S. offers benefits such as increasing returns and higher productivity. Reasons for this are lower transportation costs, spillovers from U.S. firms and agglomeration effects. Besides, the border possesses relative large endowments in communication, transportation and energy infrastructure and human capital, which makes it an attractive location and provides the possibility for a rapidly growing industry. Trade liberalization in Mexico has led to a shift from firms producing for domestic markets to firms producing for foreign markets. A consequence of this shift is a spatial deconcentration of economic activity which leads to differences in the development of regions. The border became a very attractive location for national and international firms. Thus, the location close to the U.S., the good infrastructures and externalities led to a new industrial centre in the border after liberalization. Clearly, the border region has attracted firms after liberalization and it developed into an important industry centre. Hanson (1997) confirms the importance of such industry activity on the level of wages and finds evidence of a positive correlation between nominal wages and proximity to industry centers. The empirical results in his paper show a decrease in wage with distance from the two most important industry centers of Mexico, i.e. Mexico City and the U.S.
Faber (2007) highlights another development among firms after liberalization. That is, the reduction of trade barriers between Mexico and the U.S. did not only lead to a pull-effect for firms towards the border. There was also a push-pull-effect away from the border to the interior region, where firms are naturally protected against foreign competition due to lower transport costs. The author states that export and intermediate supply have positive effects on the employment in the border regions, but that the increased import competition has an opposite effect. National firms which are threatened by foreign firms prefer a location further away from the U.S. border, because transportation costs for foreign firms are higher in that case. The push and pull effect, that is caused by the liberalization, affect sectors in different ways. Export sectors in the border regions have
7
benefitted more from liberalization compared to export sectors in the interior regions. For import sectors the opposite is true.
Additionally, research finds an increase in the demand for skilled workers since the trade liberalization.8 This can be explained by several reasons and has its implication for the relative wage levels in the border states compared to the rest of Mexico. First the tradable sector is experiencing higher growth rates and a larger increase in the share of skilled workers compared to the non-tradable sector.9 Tradable sectors are highly present in the border region, thus if the skilled/unskilled labor employment rate is growing faster in tradable sectors than in non-tradable sectors, the border region is becoming more skill intensive. Additionally, the tradable sector has a less fragile position than the non-tradable sector.10 It has better access to international financial markets which gives it access to credit, whereas, non-tradable firms have difficulties becoming credit. This has caused a more stable environment for the tradable sector since the liberalization. Furthermore, the depreciation during the Peso Crisis has caused the prices of non-tradables to fall, making the input prices for the tradable sector cheaper. This improves their relative position. Verhoogen (2006) linked this depreciation to the productivity of exporting and domestic firms.11 He argues that the depreciation benefitted exporters, which led to a quality upgrade in their manufacturing plants. In turn, this led to an upgrade in the skill of the workforce. Furthermore, exporters introduced more new technology and invested more in training than non-exporters.12 Because of these investments, the skill share of the workforce in the border increased.
A second reason for the increase in the skill of workers is the large increase in FDI, and in particular in the border region. Feenstra and Hanson (1997) state that over 50 percent of the increase in skilled labor share can be explained by the growth in FDI. The arrival of foreign firms increased the demand for more capital, new technologies and logistic
8
Feenstra and Hanson, 1997, Chiquiar, 2004
9
Cragg and Epelbaum, 1996
10
Tornell et al., 2004
11
From Goldberg and Pavcnik, 2007
12
operations for these companies‟ assembly production.13
These assembly operations require relatively more skilled labor than other Mexican manufacturing plants.
Another aspect that influenced the inequality in Mexico is the pre-trade reform protection levels. Protection levels of almost all industries were very high in Mexico before the membership with the GATT. When an industry trade barrier is reduced, competition from abroad increases which, in turn changes the price levels of the products in that industry. This alters conditions in this industry, which results in lower wages. The larger the trade barrier reductions, the more it impacts wages. This is confirmed by Hanson and Harrison (1999), who find a negative impact of initial tariffs on prices, which suggests that prices fell more in sectors with higher initial protection. Contrary to what one would expect, Mexico‟s protection levels were higher in unskilled industries than in skilled industries. At the same time, Mexico is assumed to be abundant in unskilled labor. When trade was liberalized, the skilled sector had less tariff reductions than the unskilled sector. According to the Stolper-Samuelson theory, this should lead to larger price reductions in unskilled industries. Robertson (2000) confirms this and shows that trade liberalization causes the relative price of skill-intensive goods to rise. Moreover, he finds a positive and increasing relationship between relative prices and relative wages between 1987 and 1995, implying that relative wages increased during this period.
In sum, the economic changes, caused by liberalization are the following: First, the border region experienced a relatively higher economic growth than the other regions in Mexico after liberalization. Second, Liberalization led to an increase in the share of skilled workers in the border region compared to the other regions. Third, the reduction in trade reforms led to uneven development of wages in several industries. All three arguments are able to explain the higher wage levels in the border region. The basic facts in this paper support the first and the third argument, but, however, contradict the second. It is true that the border region witnessed a large increase in economic activity after liberalization. FDI figures show a large absolute increase in FDI, which can be seen as an indicator for economic growth. Later on, it will be confirmed that this indeed can explain
13
the regional wage differentials. As well, the border region industries were less affected by trade barrier reduction compared to the rest of Mexico. However, the data about the share of skilled workers shows other facts. The share of skilled workers did increase in the border region, but surprisingly not compared to the rest of Mexico. Whole Mexico witnessed an increase in the share of skilled workers, in which the border region shows no difference.
This raises the question why it is possible that previous research finds other results than what the data in this paper reveals. There are two possible explanations for this, one is the type of data and the other is the time period. The data in this paper is a household survey, while most papers use manufacturing firm-level data. Using household data yields that sectors that are less affected by liberalization, are also included. The manufacturing sector, clearly, is highly influenced by trade liberalization. While other sectors, like some service sectors, have had a substantial smaller impact from liberalization.
That the industries are affecting inequality differently is confirmed by Cragg and Epelbaum (1996), who also use a household survey. They conclude that if there is an explanation of the wage dispersion in Mexico, it is the higher skill intensification in the manufacturing sector. In their research, they use only full-employed workers, from a data sample of only urban areas. My research differs from that of Cragg and Epelbaum, in a way that it includes part-time workers and workers in rural areas as well. This can influence the data because these workers may have a lower education on average. Thus the inclusion of all sectors and all type of workers can be an explanation why the relative share of skilled workers in the border region compared to the rest of Mexico in this paper differs from that of other papers.
paper, and which is confirmed by Esquivel and Rodrígues-López (2003) and Bineli and Attanasio (2009), the period before and after NAFTA show different characteristics. Both papers find that before 1994, the wage inequality increased. The gap between skilled and unskilled wages increased fast. After 1994, the fast increase in wage inequality stopped, and relative wages stayed at a fairly equal level.
Unfortunately, these papers do not research the regional wage differences, nevertheless, they prove that the wages and labor market characteristics differed pre and post NAFTA. The data in my paper shows that, pre NAFTA, the share of skilled workers in the border is slightly higher than the national share of skilled workers. Post NAFTA, this difference is abolished. In addition the trade barrier reduction influence is also different pre and post NAFTA. Before NAFTA, the share of people working in industries heavily affected by barrier reduction, were far lower in the border region compared to the rest of Mexico. After NAFTA this share is fairly equal. Because the effects before and after NAFTA are included in the results in this paper, it neutralizes the effects pre and post NAFTA. This leads to other findings in this paper, compared to the papers discussed above.
Apparently, the data in this paper does not suggest that the higher wage level in the border is determined by a higher share of skilled workers. However, some researchers mention an increase in the skill premium in Mexico after liberalization. An increase in the skill premium means that the wages for skilled workers increase relative to the wages of unskilled workers. If that is the case for the border region, more than the other regions, it could explain the higher wage level. Next, an overview of the evidence and reasons of higher relative wages.
unskilled workers by looking at production and non-production workers. Production workers are associated with unskilled labor and non-production workers with skilled labor. They found an increase in the wage gap in real wages of 27% between 1988 and 2000. Feleciano (2001) shows that the largest contribution to the increased wage gap, is an increase in the gap between highest wages and the middle wages, i.e. between the 90th and 50th percentile. Feenstra and Hanson (1997) also find, as a result of the highly concentrated maquiladora activity, that the border regions had the largest increase in relative wages and non-production wage share between 1985 and 1988.
The most important reason of the increased relative wages is the increase in the return to education. The increase in the return to education between 1987 and 1993 can be explained by both portable skills associated with a particular task and the rise in return to general education.14 The increased return to education was also reported in Binelli and Attanasio (2010). They state that the return to education has increased between 1987 and 1995. The returns decreased between 1995 and 2000, although with a significantly smaller amount. So overall, they conclude, there is an increase between 1987 and 2000. Next to the return to education, Revenga (1997) discusses that firms with relatively more skilled employees, share a larger part of their rents than firms with relatively more unskilled employees. The author argues that there could be a scarcity of skilled labor and so a stronger rent sharing due to the bargaining position of skilled workers. Esquivel and Rodríguez-López (2003) find that technological change is an important factor of the wage inequality between 1988 and 2000. They even conclude that without technological change, trade liberalization would have led to a reduction in the wage gap.
In sum, possible determinants of wages, based on literature, are; the level of education, economic activity, the reduction in trade barriers and the skill premium. I will test the effect of these determinants on the wages in Mexico and their implications on regional differences.
14
These determinants are mainly based on research dealing with one particular subject. And few papers discussed the regional differences in wages. Most papers focused on overall wage inequality. Besides, almost all findings are based on research conducted using firm level manufacturing data. My research, however, uses household data, that encompasses a random sample of the whole population as a representative for the whole country. Especially the use of data on all industries, can lead to different conclusions than if only manufacturing data is used. Data on part-time and full-time workers of all age categories are included, which enables me to test wage determinants between larger sample groups. I will use a time period between 1984 and 2006, which gives a good opportunity to compare pre and post NAFTA effects. The individual data is averaged per state, so regional differences become measurable. Furthermore, the household data is combined with a FDI dataset. With this data I will test previously found hypotheses on this topic in a different way which is not been done before. Further on, it becomes clear that this data and method gives new insights on the discussion about wage inequality.
3. Theory
skilled labor or the relative supply of unskilled labor has increased in Mexico, which contradicts the Heckscher-Ohlin theory. This can be explained using the following theory described by Barba-Navaretti and Venables (2004). The next figure illustrates what happens to the wage-rental ratio in a developing country when domestic firms start producing a stage of the production coming from multinationals in developed countries.
The framework for this theory is a 2-country, 2-sector and 2-factor model. The manufacturing sector (M-sector) has fixed factor intensities and uses factors Kim for
capital and Lim for labor. The rest of the economy (Y-sector) is numeraire and uses the
rest of the factor endowments, Ki- Kim and Li- Lim. Capital and Labor can be assumed as
skilled and unskilled labor respectively.
Figure 1 – Relocation and factor prices
Source: adapted from multinational firms in the world economy, G. Barba-Navaretti and A.J. Venables, 2004
In figure 1, the lines OY1 and OY2 represent the skilled-unskilled endowment in the
Y-sector of the developed and developing country respectively (Ki- Kim /Li- Lim). The
isoquant represents the technology of the Y-sector and the slope where this isoquant crosses the endowment lines, indicates the factor price ratios of country 1 and 2.
labor intensive. This stage of production can also be relatively unskilled labor intensive to the developing country, say with capital labor ratio lying below lines OY1 and OY2
(interval A). In this case the same happens as described in the Heckscher-Ohlin theory. The demand for unskilled labor in the M-sector increases, which means that the supply of skilled labor in the Y-sector increases as well. This indicates that the relative wages of unskilled workers in developing countries increase compare to those of skilled workers. Thus the unskilled/skilled ratio (ui/si) increases. Alternatively the relocated stage of
production can be relatively skilled labor intensive, with skilled-unskilled ratio lying below line OY1 and above OY2 (interval B). All cases in interval B leads to a fall in the
unskilled/skilled ratio (ui/si), meaning an increase of relative wages in the developing
country. This is the case because the demand for skilled labor in the M-sector and the supply of unskilled labor in the Y-sector both increased. The activity gained by the M-sector is relatively skill intensive, increasing the demand for skilled labor, while the supply of unskilled labor in the Y-sector increases because it must employ the rest of the unskilled endowments of the M-sector. This is visible in the downward movement of the line OY2 indicated by the arrow IIB.
This theory can be applied to Mexico. Researchers assume that the relocated parts of production from the U.S. to Mexico are relatively skill intensive for Mexico.15 This could explain the increase in inequality instead of a convergence of wages.
An increase in regional wage differences can be explained by the large shares of FDI and maquiladoras. Most of it is coming from the U.S and is located in the border states. So when the above theory is related to the Mexican case, the prediction is that the demand for skilled labor in the border region has increased relative more to the rest of Mexico. This implies that the relative returns to skilled labor are higher in the border region.
Another explanation for the regional inequality is given by the theory of Hanson (1997). It states that the border region attracted firms, leading to agglomeration effects. The theory explains the increase of a firm‟s returns resulting from agglomerated markets. It
15
assumes that each good is produced by a single, monopolistic firm and that, all else equal, fixed costs and transport costs make firms prefer to serve consumers from a single location near to large markets. This yields benefits for firms located in this industry centre compared to firms who are not. Once firms are agglomerated in an industry centre, it attracts other firms and new entrants. They prefer this centre because, in addition to the location, there are external effects to benefit from. It offers higher quality infrastructure, better forward and backward linkages and knowledge spillovers. These externalities lead to an enhancing of productivity of all firms in the industry centre. So both internal and external benefits yield higher returns for firms in agglomerated markets which can be reflected in wages.
Additionally, from the perspective of workers, there are congestion costs related to living in the industry centre as compared to living in outlying locations, like higher land rents. Therefore in the industry centre, firms are willing to pay relative higher wages to compensate workers for these congestion costs and so to attract them to the industry centre. Contrary, the outlying regions face higher transport costs and thus pay their workers relative lower wages to compensate for this. Congestion costs, thus, is an explanation for the determinants of wages. However, when looking at overall inequality, the higher living costs neutralize the higher wages. On average, the employees in the border are not better off than employees in the interior region. When looking at congestion costs, wage levels can be explained, but inequality not. Previous to the liberalization in Mexico, Mexico City functioned as the industry centre for Mexico. However after trade was liberalized, the border attracted firms into its region due to its proximity to the U.S. So since trade reform, firms started to agglomerate in the border region creating a new industry center in Mexico. This would mean that wages in the border region increased relatively to the other regions in Mexico after liberalization.
explained by the Stolper-Samuelson theorem. Assuming that trade barrier reductions are larger in initially high-protected industries compared to initially low-protected industries, this theory states that trade liberalization increases the price of goods in industries with initially low-protection relatively to goods in industries with initially high-protection. The change in goods prices reduces the labor demand in high-protected industries and increases the demand in protected industries. The increased demand in the low-protected industries causes the relative employment and wages to rise in these industries, compared to high-protected industries. A second effect of a decrease in industry protection is the reduction in the available rents in that industry. This reduces the size of rent that is available for wages in firms where workers are able to obtain a share of that rent. In sum, trade liberalization affects the wages and employment in industries depending on the size of trade barrier reductions. If trade barrier reductions are high, the negative effect on wage and employment are larger compared to when trade barrier reductions are low.
The three main determinants of differences in regional wage levels researched in this paper and based on literature and theory are the following.
Figure 2 – Real wages per hour
Source: own calculations from ENIGH database
The first determinant is the share of skilled workers in a region. The prediction is that the coefficient of the variable „SKILL_SHARE‟ shows a positive sign, which means that the larger the share of high skilled workers, the higher is the real wage. If the share of skilled workers is higher in the border region, it can explain the wage difference.
The second determinant concerns the reduction in trade barriers of protected industries. Before the liberalization, the protections of particular industries were tremendously high. After Mexico joint the GATT most of the trade barriers were reduced. As theory predicts, trade barrier reduction increases competition from abroad, which forces prices to go down. These price reductions and fierce competition are reflected in wages. This is what this determinant is about to test. The variable REDUCED_BARRIERS indicates the share of people working in industries with the largest trade barrier reductions. The
prediction for the sign of the coefficient is negative, because that would mean that large trade barrier reductions have a negative influence on wages.
FDI is the final determinant for real wage in Mexico. FDI is an indicator for industrial activity and as described before wages are supposed to be higher in agglomerated markets. This leads to better back and forward linkages, spillovers, lower transport costs, higher productivity and more efficiency for all firms in the agglomerate market. Beside, foreign firms, in general, possess better technologies, more human capital and more efficient product processes. All these factors can raise the wages paid by firms. So regions with large shares of FDI have higher wages compared to regions with no or small shares of FDI. This means that the predicted sign for the coefficient of the variable „FDI‟ is positive.
Using these three variables as the determinants of Real Wage in Mexico, it gives the following regressions. The panel variable, i, indicates STATE and the time variable, t, indicates YEAR.
(1) Real Wageit = α0 + α1SKILL_SHAREit + α2REDUCED_BARRIERSit + α3FDIit +
α4PER_1996_1998it + α5PER_2000_2006it + eit
This model will be used to estimate the effect and significance of the determinants on wages. With the estimated coefficients, it is possible to test if these determinants can explain the regional differences in wages.
4. Method
Before testing wage differences between regions, I will do a basically univariate test for equality of the mean wage across states. This means to test if there are significantly differences in real wages across the states of Mexico. Therefore, a dummy, δi, for each
H0 : δ1 = δ2 = … = δ32
H1 : the δi are not all equal
The F-statistic for 31 observations and 278 (320-32-10) degrees of freedom is 14.80. This result indicates that at a significance level of 99%, the H0 hypothesis is rejected, which
means that at least one of the coefficients is not equal to another.
Next to consider is the right panel data model to use for the regressions. The possible models are the pooled regression, the seemingly unrelated regression, the fixed effects model or the random effects model.16 The characteristics of these models are explained by Hill et al. (2008). The pooled regression assumes that the parameters are fixed for all years and the same for all states. The seemingly unrelated regression (SUR) jointly estimates the equations for all states, accounting for different error terms and correlated error terms. The fixed effects model assumes that all the behavioral differences between states over time are captured by the intercept. The random effects model assumes that the differences in states are captured by the intercept, but threat the state differences as random.
A pooled regression can be useful to estimate the model in a simple way, but it does not estimate the effects of the determinants for each state. The SUR model is not of practical value because there are too many units, N=32. This leaves either the fixed-, or the random-effects model. To test which model is best, a Hausman test can be done.17 This test compares the coefficient estimates from the random-effects model to those from the fixed-effects model. I find that the p-value (Prob > Chi2) is smaller than 0.05 for all regressions. This indicates that the null hypothesis is rejected and that the coefficient estimates are not equal to each other. This means that the random-effect estimator is inconsistent, so a fixed-effects model is preferred over a random-effects model.
16
Hill et al., (2008)
17
To continue, heteroskedasticity is tested. In this study for example, it would be possible that at higher education levels, the variances in real wages is higher than for lower education levels. When there is heteroskedasticity, an ordinary least squares (OLS) model is not useful anymore. Instead a generalized least squares model must be used to estimate the linear model, when the errors are heteroskedastic. To test for heteroskedasticity with panel data, an LR test can be done.18 Testing for heteroskedasticity results in a p-value smaller than 0.01. This means that the regressions have to be controlled for heteroskedasticity.
Finally, an autocorrelation test is done. If autocorrelation is present, it means that the mean error term is correlated. This can be due to an immeasurable shock, which correlates the error term of one year to a previous year. To test autocorrelation for panel data, Wooldridge (2002) derives a test for panel-level autocorrelation and Drukker (2003) developed a user-written program which enables to perform this test.19 Testing for autocorrelation, I find the presence of serial correlation at a significance level of 99%. This means that the regressions have to be controlled for autocorrelation.
Concluding from the above findings, I use a feasible generalized least squares (FGLS) with heteroskedasticity across panels and autocorrelation within panels. To see what happens to significance levels and coefficients, I will run several different tests in the next section.
5. Data
For this research I will use a panel data model. These kinds of data consist of a group of cross-sectional units which are observed over time. In this case the observed units are the 32 states of Mexico and the timeline is between 1984 and 2006. The most important dataset used in this research is secondary data that is made available by the National Institute of Statistics and Geographic‟s (Instituto Nacional de Estadística y Geografía,
18
Stata: http://www.stata.com/support/faqs/stat/panel.html
19
INEGI). The data consist of a household survey of incomes and spendings in Mexico between 1984 and 2008. I have chosen to drop the data of 2008, because it uses other variables and indicators than the data of other available years. Because there are some gaps between the several years of data that is made available, the observed years are only: 1984, 1989, 1992, 1994, 1996, 1998, 2000, 2002, 2004 and 2006, so that T=10. The observations in the database are those of individuals. However, because I am interested in the differences between regions, the individuals are grouped together for each state. This leads to a dataset with states averages. This implies that for each year the number of observations is N=32, so that the total dataset contains 320 observations. The time period of this dataset in particular measures the regional inequality after the major trade reforms of Mexico, since its start in the early 1980s.
For the variable that indicates the education level some adjustments are necessary as well. The coding for the level of education differs for almost all the years of observation. The codes indicate which level of education has been completed by the individual. I equalize the coding for all the years so that they will correspond to each other by using codes from 0 to 9. When using this variable in a linear regression, it would imply that, for example, the difference between 0 and 1 is the same as between 8 and 9, whereas this is not the case. Therefore, I group the individuals in skilled and unskilled workers. When an individual has an education level of 6 or higher, this person is assumed to be skilled. In this survey an education level of 6 is equal to 12 years of education or more. I use an education level of 6 and higher for high skilled workers, because skilled workers then represent approximately 15-25% of the population. These percentages reflect a plausible division of skilled and unskilled workers of the population.
Subsequently, from the individual dataset I calculate the share of skilled workers per state, so that it is a continuous variable. Furthermore, the coding for industry differs among the surveys. Especially from 2000 onwards, other coding is used. I equalize the coding from before and after 2000 using the six-digit NAICS codes & titles.20 This website provides indications of which code stands for which part of the industry. Using the codes for the different industry I could make a distinction between industries which experienced relatively larger trade barrier reductions and industries with relative small trade barrier reductions. In the individual data a person who works in an industry with large trade barrier reductions was indicated with a 1 and a person working in one of the other industries with a 0. For the dataset which I want to use for the regression, I could thus calculate for each state the percentage of people working in the 1-industry and the 0-industry. A limitation related to this method is that it only allows me to use the manufacturing industries. Only these industries have available data on trade barrier reductions from right after the membership GATT in the early 1980s. Following Hanson and Harrison (1999), the manufacturing sector is divided in 9 industries of which 3 witness the largest trade barrier reductions. These industries are wood products, paper and printing, and stone, clay and glass. A consequence is that it narrows the sample size.
20
For some small states, the number of individuals who were working in the manufacturing sector is too small to use as a reliable representation of the population. This is why I generalize the average of a region for all the states in that particular region, so there is no distinction between the states in one region.
One limitation of my data is that it uses secondary data that I needed to edit so it could be used for this research. However, I could not solve all problems related to the fact that scaling and categories changed overtime. Within my capabilities, I have equalized the scaling of the variables for all years. Nevertheless, a dataset where all years use the same scaling would lead to better results.
To complete the dataset, I add a variable for FDI. Data about FDI is not included in the survey of the INEGI, but is made available by the Secretaría de Economía. The data from the secretary of economics represents total foreign direct investments in millions of dollars in both directions. This leads to the possibility that some states have negative FDI numbers. I merge this dataset with the dataset described above, so I can include FDI in the regression. Unfortunately, this data was only available from 1994 onwards, so there are missing values in years 1984, 1989 and 1992. Furthermore, to control for the size of the states I divide the FDI numbers by the relative share of inhabitants for each state, where the whole population is equal to 100. In this way, FDI figures are no longer biased by the size of the state or region. These adjustments in the individual household survey, create a dataset with state averages which leads to a usable dataset for this research. I will shortly describe which dependent and independent variables I use in this paper.
working in industries that witnessed the largest trade reductions. The third measures the presence and influence of foreign firms, „FDI‟. State data has been available since 1994, which means that the relationship between FDI and real wages can only be researched in the post-NAFTA period.
Furthermore, I will use time dummies as independent variable for three time periods in order to capture yearly effects. I use periods instead of years of observation, to increase the degrees of freedom. The first period is from 1984 until 1994, the start of the liberalization. The second period captures the peso crisis, from 1994 until 1998. The final time dummy is the period between 2000 and 2006, in which Mexico recovered from the crisis. By using time dummies, events that occur in a particular period but are relevant to the whole country, are not visible in the coefficients of the other variables. Finally, I create an interaction variable between „FDI‟ and „Skill Share‟. This variable takes the value of FDI x Skill Share for each observation. I use it to test the theory that increased openness and trade increases the demand for skilled workers. The measurement for increased openness used here is FDI.
Tables 1 to 5 represent descriptive statistics of the dataset. The first table provides a small overview of the amount of states per region and the share of the population living in these regions. The other tables give summary statistics of the 4 most important variables.
Table 1 – General Region Statistics
REGION STATES % of POPULATION
Table 2 – Real Hourly Wage in Mexican Peso
YEAR STATISTICS
REGION 1984 1994 2006 MEAN ST.DEV MIN MAX
NATIONAL 17651 17599 20814 17522 4880 7022 36674 BORDER 22783 21216 24528 21476 4114 14071 33392 NORTHERN CENTRAL 17096 16310 21336 17452 4413 9805 29072 SOUTHERN CENTRAL 14387 15601 19391 15683 3670 9093 27348 MEXICO CITY 23921 32484 24501 24076 5326 14176 36674 SOUTH 16679 14387 18088 14959 3374 7022 21850
Table 3 – Share of skilled workers
YEAR STATISTICS
REGION 1984 1994 2006 MEAN ST.DEV. MIN MAX
NATIONAL 0.156 0.157 0.243 0.200 0.065 0.034 0.387 BORDER 0.164 0.171 0.261 0.206 0.058 0.074 0.321 NORTHERN CENTRAL 0.166 0.144 0.259 0.212 0.069 0.080 0.387 SOUTHERN CENTRAL 0.135 0.135 0.220 0.176 0.056 0.034 0.299 MEXICO CITY 0.188 0.292 0.308 0.270 0.084 0.132 0.384 SOUTH 0.159 0.151 0.227 0.198 0.055 0.052 0.308
Table 4 – Share of people working in industries highly affected by trade barrier reductions
YEAR STATISTICS
REGION 1984 1994 2006 MEAN ST.DEV MIN MAX
Table 5 – Total FDI in millions of Dollars, adjusted for the size of the regions
YEAR STATISTICS
REGION 1994 2000 2006 MEAN ST.DEV MIN MAX
NATIONAL AVERAGE 64.08 126.62 164.90 108.11 226.89 -17.46 1850.51 BORDER 107.44 305.31 298.07 201.45 146.52 43.40 608.17 NORTHERN CENTRAL 14.70 68.59 185.49 75.90 159.47 -14.56 909.68 SOUTHERN CENTRAL 33.95 49.62 64.07 41.51 56.12 -17.46 194.93 MEXICO CITY 443.32 530.14 619.41 571.57 619.21 25.34 1850.51 SOUTH 10.99 26.17 44.33 23.04 42.74 -0.03 246.11
Note for Table 1-5: Source: Own calculations form ENIGH database.
means that, relatively, the border region is converging to the Mexico City region in terms of total FDI.
The summary statistics of the data can tell a lot about the differences between the regions. However, next I will estimate the model to test if these variables are determinants of real hourly wages. Then I will use the data observations to see how the determinants can explain differences in regional wage levels.
6. Results
The results from running regression (1) are presented in the following 2 tables. I will show the results of the estimation of the pooled regression and the FGLS regression controlled for heteroskedasticity and autocorrelation.
In both tables the first column shows the results for the time period 1984-2006 and the second column for 1994-2006. This leads to some different values of the coefficients and significance of the variables. The first table shows the estimation of the model using a pooled regression.
Table 7 - GLS regression controlled for heteroskedasticity and autocorrelation
Note for table 1-4: * = statistical significance at the 90% level, ** = statistical significance at the 95% level, *** = statistical significance at the 99% level.
These results show that all three variables are statistically significant at the 99% level. The expectation of a positive sign for Skill Share and FDI and a negative sign for Barrier Reduction is met. These signs do not change in table 7. However, the significance of the variable Barrier Reduction changes in the second model.
In the first column it is significant at the 90% level, and in the second column it is not significant anymore. This is probably because the effects of the barrier reductions were the heaviest right after GATT. These are the late 1980s and early 1990s. The second column does not capture this period, so in this column the estimation shows a coefficient that is not significant. This shows that before 1994, the effect of the trade barrier reduction increased the regional wage differences. However, after 1994, this is no longer the case. This, again, can explain the different trends before and after 1994. To explain the effect of trade barrier reductions on wage I will use the 1st column, for the other two I will use the 2nd column. The estimation of the model shown in table 7 will be used in this paper to research the effects of the variables. However, the first table gives important information about the goodness-of-fit measure. This measure is only possible for the pooled regression method because it uses the ordinary least squares concept. The FGLS
model instead uses the generalized least squares concept. The coefficient of determination, the R2, is circa 0.5, which is the proportion of variation in real wage explained by the independent variables within the regression model. An R2 of 0.5 means that the model has a good explanatory power. This is important when interpreting the effects of the determinants on wage, because this means that the determinants explain a large part of the variation in wage. This measurement of goodness-of fit is not given in the last table, nevertheless that is the model that needs to be used for this dataset.
Use of the Model to explain regional wage differentials
The tables above are giving a good first impression about a positive or negative influence of the determinants and the statistical significance of them. In this section I will describe the effect of each determinant on the Real Wage and its implications for the wage inequality between the border region and the rest of Mexico.
Skill Share
border. In later years, however, the national share was much higher. Previous literature based on the period before NAFTA, finds larger shares for skilled workers in the border. However looking at the whole period, there is no prove of a higher share of skilled workers in the border.
Figure 3 – Share of skilled workers
Source: Own calculations form ENIGH database.
The significant positive coefficient of skill share indicates that the share of skilled workers in a region is a determinant of real wage, but the above illustrates that it is not a determinant for the wage inequality between the border and the other regions. This contradicts the expectations and in the next section I will discuss possible explanations for this.
Barrier Reduction
As expected the sign of the coefficient of trade barrier reductions is negative. This means that when trade barrier reductions are high, it negatively influences real wage. So when
one region experienced larger trade barrier reductions than another region, the average real wage in that region is lower, all else equal. The measurement for trade barrier reductions is the share of people working in industries which witnessed the largest barrier reductions. When this share increases with 0.01, the real wage in a region decreases with $58,05. Again it is useful to study the presence of trade barrier reductions for all the years of observation. This is done in figure 4.
Figure 4 – Share of people working in industries highly affected by trade barrier reductions
Source: Own calculations form ENIGH database.
Right after the liberalization in the early 1980s, the share of people working in industries affected by large trade barrier reductions in the border, was significantly lower than the national average. In 1989 and 1992 the difference was 0.1 or larger. A difference of 0.1 in the share of people working in affected industries means a difference of $580.51 in real wage. The smaller amount of industries affected by trade barrier reductions in the border region has positively influenced the average wage level of that region. Especially after the membership of the GATT which was the start of large reductions in trade barriers. This also explains the change in significance when the time period is reduced to 1994 to
2006. As the figure shows, in that period the difference is much smaller compared to between 1984 and 1994. That the regions were differently affected by barrier reductions has only determined regional wage difference until the mid 1990s. Thereafter, the variable Barrier Reduction can no longer explain regional wage differences.
Concluding, because the share of people working in industries affected by large trade barrier reductions is much lower in the border region, the average wages are higher in the border compared to the other regions in Mexico. Therefore trade barrier reduction is a determinant of the wage inequality between the border and the rest of Mexico.
Foreign Direct Investments
The effect of FDI on real wage is positive and significant. The coefficient of 7.72 means that when FDI goes up with 100, the real wage increases with $772,-. From the descriptive statistics it already becomes clear that the level of FDI is substantially higher in the border compared to the national level. The next figure illustrates the development of FDI in the border and at national level.
Figure 5 – Total Foreign Direct Investment, adjusted for region size
Source: Own calculations form ENIGH database.
All three variables are considered to be determinants of the level of real wage. Only the last two can explain the difference between the wage level in the border and the rest of Mexico. These variables can explain a gap of ca. $1350. This is about 25% of the average wage gap between the border and the national level. The skill level in the border is not substantially higher than the rest of Mexico, and it therefore cannot explain differences in real wage. However the high demand for skilled labor in the border region can lead to higher wages for the skilled workers, which increases the overall wages. Previous literature and theory found that liberalization and increases trade increased relative wages. Thus, in regions with large investments from abroad, the demand for skilled workers should increase, which in turn pushes up the skilled wages. In this way, the skill level might be equal in the border and the rest of Mexico, but the returns to skill might be higher in the border. Using an interaction variable between FDI and Skill Share, it is possible to test the influence of foreign firms on the wage premium.
Interaction variable
An interaction variable captures the interaction effect of two variables, in this case between FDI and the share of skilled workers. Theory suggests that FDI affects real wages in two ways. First effect is the clustering of firms and industries in regions with large shares of FDI. Businesses and organizations are interested in locating near active and large markets. These agglomerated markets have benefits over other markets which lead to lower production costs and higher productivity, which in turn are reflected in wages. So regions with much FDI have in general higher wages. This first effect has been confirmed in the previous section. However, FDI does also affect real wages indirectly through the demand for skilled labor. As is described in previous literature, foreign firms increase the demand for skilled labor, which leads to higher wages for skilled workers. Thus large values of FDI increase the relative wages for high skilled workers as compared to the low skilled workers. The interaction variable captures this second effect. The regression with and without the interaction variable looks as follows, where Real Wage is Y, Skill Share is X and FDI is Z.
(1) Y = β0 + β1X + β2Z + e
(2) Y = β0 + β1X + β2Z + β3XZ + e
The marginal effect of Skill Share on real wages in the first regression is a constant effect equal to ∂Y/∂X = β1, while in the second regression the marginal effect of Skill Share
depends on FDI and is equal to ∂Y/∂X = β1 + β3Z.
Table 8 – Regression (1) Table 9 – Regression (2) Coefficient Coefficient Constant Skill Share FDI Number of observations 11126*** 39893*** 7.27*** 224 Constant Skill Share FDI Skill*FDI Number of observations 9827*** 43631*** 23,5*** -51.1*** 224
From the results above it can be concluded that the expectation that an increase in FDI has a positive effect on real wages through higher wages for skilled workers is not the case here. This is visible in the negative sign of the interaction variable and becomes clearer when looking at the following figure.
Figure 6
This figure illustrates the marginal effect of Skill Share for different levels of FDI. The range for FDI is based on the range of total FDI in one region. Some states show larger values of total FDI, but the highest regional total value of FDI is about $600. You can see that for the first regression, the blue line, the marginal effect of Skill Share is constant for all values of FDI. However when introducing an interaction variable the marginal effect is no longer constant but is now dependent on FDI.
The red line represents the marginal effect of Skill Share for the second regression. It illustrates that when FDI is low, the marginal effect of Skill Share is high and when the value of FDI increases, the marginal effect of Skill Share decreases.
7. Conclusion
Trade liberalization affected the wage conditions in Mexico. Larger volumes of trade and the rise in FDI and in particular, maquiladoras all altered the wage levels. The trade developments in Mexico were affecting regions differently, which led to different wage levels in the 5 Mexican regions. These differences can be explained by several reasons. The increase in skill share and FDI, and the reduction in trade barriers all have a significant influence on the wage level. However, this paper shows that some of these determinants have a different effect on regional wages than is expected by theory and previous research.
Previous findings are, in general, that the wages in the border region are higher compared to the interior region. The reasons for these higher wages are the following: First they find that wages are higher in regions where economic activity is high. That is where industrial centers are located. Mexico is characterized by much FDI, of which much is in the form of maquiladoras. The border region is one of those industrial centers who experienced the largest increase in FDI. Second they find that the share of skilled workers increases. Especially foreign orientated en foreign owned firms tend to invest more in knowledge and education of their workers. Besides they tend to pay higher wages for their workers and in particular their skilled workers. Finally, some researchers find a decrease in wages in industries where there are large trade barrier reductions. The border region was relatively less affected by these reductions compared to the other regions.
from our estimated model it can be concluded that if FDI increases, the marginal return to skill decreases. Thus, in the border, where FDI is high, the relative wages for skilled workers decreased instead of increased. It also did not lead to a higher share of skilled workers. The fact that previous research did find a link between increased FDI and increased demand for skilled labor, is not been confirmed in this paper. To find out why relative wages increased, other than the increase in FDI, is an important topic for further research.
The contributions of this paper to the discussion of regional wage differences are the following. Firstly, the time period on which my conclusions are based. Most of the previous literature on this topic covers the period before the mid 1990s. However, after NAFTA, there were some radical changes that have affected wages. Most of the papers so far did not include the years after NAFTA in their research. The most important basic fact after NAFTA is that the rapid increase in relative wages stopped. The income inequality, measured by the GINI coefficient even decreased. The peak of this income inequality was in 1994 and according to this paper the regional wage differences were also the highest in 1994. It is therefore important to look at the period pre-NAFTA and post-NAFTA. This paper shows that pre-NAFTA the regional wage inequality between the border region and the rest of Mexico is due to a smaller impact of trade barrier reductions. Post-NAFTA this regional wage inequality is due to large values of FDI and the rise of an economic center in the border region.
workers than the rest of Mexico. However when looking at all the industries there is no substantial difference between the share of skilled workers in the border and the rest of Mexico.
When looking at the contributions of this paper, one has to keep in mind the limitations it has. The data used in this paper is secondary data that I have edited so it could be used for this research. However, I could not solve all problems related to the fact that scaling and categories changed overtime. Besides, this paper looks only at the wage determinants and their regional impact. It does not look at the living costs, so these conclusions cannot say anything about the real inequality between regions. If wages and living costs are both high, overall inequality stays the same. In addition, this paper only focuses on determinants as a consequence of liberalization. Of course, other reasons like political, geographical or juridical ones, can also determine wage levels in Mexico. Other papers discuss the effect of labor unions, agricultural developments or migrations. Aspects like these have not been discussed in this paper. Finally, there are some large gaps in the observed years between 1984 and 1992. The first gap is 5 years and the second is 3. Developments during these years cannot be measured and used in the estimation of the model.
In sum, this paper concludes that the higher wages in the border compared to the interior region, can be explained by larger values of FDI and a smaller decrease in trade barriers in the border region. An increase in FDI leads to a lower marginal effect of the share of skilled workers on average real wages. This might imply that the FDI investments from the U.S. to Mexico are relatively unskilled intensive for Mexico, which contradicts the previous idea that, it is relatively skilled intensive. If the former is true, an increase in FDI, leads to a decrease in the relative wages, leading to a convergence of skilled and unskilled wages in Mexico. With this idea in mind, the wage inequality in Mexico should have a different cause than an increase in FDI. Suggestions for further research on the effects of FDI could focus on the increase of wages for unskilled workers and the determinants for the reduction in relative wages.
8. References
Anderson, E. “Openness and inequality in developing countries: A review of theory and recent evidence” World Development Volume 33, Issue 7, July 2005, Pages 1045-1063
Barba-Navaretti, G and Venables, A.J. “Multinational firms in the world economy” 21 january 2004, Chapter 4.
Binelli, C., Attanasio, O. “Mexico in the 1990s: The Main Cross-Sectional Facts” Review of Economic Dynamics Volume 13, Issue 1, January 2010, Pages 238-264
Hill, R.C., Griffiths, W.E., Lim, G.C. “Principles of econometrics” Third edition. John Wiley & Sons, 2008
Chiquiar, D. “Why Mexico's regional income convergence broke down” Journal of Development Economics Volume 77, Issue 1, June 2005, Pages 257-275
Cragg, M.I., Epelbaum, M. “Why has wage dispersion grown in Mexico? Is it the incidence of reforms or the growing demand for skills?” Journal of Development Economics Volume 51, Issue 1, October 1996, Pages 99-116
Drukker, D.M. “Testing for serial correlation in linear panel-data models.” Stata Journal 3, 2003, Pages 168–177.
Esquivel, G., Rodríguez-López, J.A. “Technology, trade, and wage inequality in Mexico before and after NAFTA” Journal of Development Economics Volume 72, Issue 2, December 2003, Pages 543-565
Feenstra, R.C., Hanson, G.H. “Foreign direct investment and relative wages: Evidence from Mexico's maquiladoras” Journal of International Economics Volume 42, Issue 3-4, 1 May 1997, Pages 371-393
Feliciano, Z.M. “Workers and trade liberalization: The impact of trade reforms in Mexico on wages and employment” Industrial and Labor Relations Review Volume 55, Issue 1, 2001, Pages 95-115
Goldberg, P.K., Pavcnik, N. “Distributional effects of globalization in developing
countries” Journal of Economic Literature Volume 45, Issue 1, March 2007, Pages 39-82
Hanson, G.H. “Increasing returns, trade and the regional structure of wages” Economic Journal Volume 107, Issue 440, 1997, Pages 113-133
Hanson, G.H. “Regional adjustment to trade liberalization” Regional Science and Urban Economics Volume 28, Issue 4, 1 July 1998, Pages 419-444
Hanson, G.H. “What has happened to wages in Mexico since NAFTA? Implications for hemispheric free trade” NBER Working Paper No. 9563, March 2003 JEL No. F1, J3
Hanson, G.H., Harrison, A. “Trade liberalization and wage inequality in Mexico”
Industrial and Labor Relations Review Volume 52, Issue 2, January 1999, Pages 271-288
Harrison, A., Hanson, G. “Who gains from trade reform? Some remaining puzzles” Journal of Development Economics Volume 59, Issue 1, June 1999, Pages 125-154
Revenga, A. “Employment and wage effects of trade liberalization: The case of Mexican manufacturing” Journal of Labor Economics Volume 15, Issue 3, July 1997, Pages S20-S43
Robertson, R. “Trade liberalisation and wage inequality: Lessons from the Mexican experience” World Economy Volume 23, Issue 6, June 2000, Pages 827-849
Stata, Data Analysis and Statistical Software “How do I test for panel-level heteroskedasticity and autocorrelation?”
http://www.stata.com/support/faqs/stat/panel.html, September 2010
Tornell, A., Westermann, F., Martinez, L. "NAFTA and Mexico's Less-Than-Stellar Performance," NBER Working Paper Series, Vol. w10289, February 2004, pp. -, 200
Wood, A. “Openness and wage inequality in developing countries: The Latin American challenge to East Asian conventional wisdom” World Bank Economic Review Volume 11, Issue 1, January 1997, Pages 33-57
Wooldridge, J. M. “Econometric Analysis of Cross Section and Panel Data” Cambridge, MA: MIT Press. 2002
World Bank, Development Research Group “GINI Index”
