Printer Friendly

Assessing returns to education and labor shocks in mexican regions after nafta.


Mexico's integration with the U.S. economy through capital (foreign direct investment [FDI] and portfolio flows) and labor (legal and illegal migration) makes for a very interesting case study. Presumably, greater integration with the U.S. economy has affected not just output growth but also the supply of labor across regions, changing the regional distribution of wages in Mexico. Previous work by Robertson (2000) based on wage equations for border and interior states of Mexico uses data from 1987 to 1997 and finds support for the proposition that U.S. wage shocks have a stronger effect on the border than on the interior of Mexico. Following a wage shock, wages in Mexican border cities converge to U.S. wages more quickly than wages in the interior of Mexico. Also, within the Mexican border region, cities with more foreign

capital and migration flows experience larger wage shocks and more rapid wage convergence to U.S. wages than other Mexican border cities do. However, his results were less clear-cut about the specific mechanism: "of forces that could integrate labor markets--goods flows, capital movements. and migration--migration may be the dominant mechanism" (Robertson 2000, 743).

We revisit this conjecture in this article along several dimensions. First, the post-North American Free Trade Agreement (NAFTA) period is particularly interesting because, after 1995, no negative shock occurred in Mexico as it did in 1982, 1987, and 1994. There is thus a steady path of rising income. (1) Second, Freeman (2006) suggests that there is no single metric on which to compare the economic importance of flows of people, trade, and capital. We therefore examine two very important markets: goods and services (real output per capita) and labor (real wages), in which the latter typically adjusts--at the national level--with lags to movements in the former. Since the enactment of NAFTA brought not only an increase in trade, but also in capital and labor mobility, we pay particular attention to these effects on (real) wages. We ask whether increasing economic integration between Mexico and the United States implies faster or slower wage convergence at the regional level. Third, we employ several panel data methods to estimate wage equations, including dynamic panels particularly suitable for high persistence patterns and reverse causation channels that address problems encountered with static panel methods.

At least two justifications can be put forward for motivating our approach in the context of a small open economy becoming more and more integrated with its larger neighbor. First, history shows that fundamental determinants may appear more important in some stochastic processes than in others. Studying a sample comprising mostly developed economies, Williamson (1996), for example, finds that the contribution of schooling to gross domestic product (GDP) per worker is never statistically significant during 1870-1913, while it is very significant to real wage growth. Further, his estimates of conditional convergence are higher for real wage equations (1.7% or 1.8% per year depending on the schooling variable used) than for GDP per worker equations (0.5% to 0.7% per year). Second, economic theory suggests that migration in search of higher wages might be expected to increase the speed at which wages converge across regions. Rappaport (2005) looks at the speed at which wages converge to their steady-state as the metric to gauge the direct and indirect effects of labor mobility. For the case of a small open economy with capital intensity below the steady-state, Rappaport (2005) shows that outmigration directly contributes to faster income convergence but also creates a disincentive for gross capital investment, with an ambiguous result. Behrens and Sato (2011) examine the impacts of migration and of labor market integration on the distribution of skills and the wage structure in both the short and long run. It is thus possible that the effects of labor mobility can be identified in wage regressions rather than in output per capita regressions.

Greater economic integration should in theory help reduce the income gap. Studying the "violent process of disintegration" of the world economy during the interwar period (1920-1938). Milanovic (2006) concludes, however, that the period witnessed very fast income convergence. Barro, Mankiw, and Sala-1-Martin (1995) consider cases when there are some types of capital (such as human capital) that cannot be financed by borrowing on world markets, which makes open economies converge only slightly faster than closed economies.

Recent articles have examined the effects of migration on output convergence, with mixed results for industrial economies. Kaufman, Swagel, and Dunaway (2003) study the impact of federal transfer programs on output convergence in Canadian provinces and obtain positive or negative effects depending on the type of transfer. Ostbye and Westerlund (2007) compare Norwegian and Swedish counties over 1980-2000 and conclude that despite Sweden and Norway being similar in many ways, migration has very different effects on convergence in these two countries: net migration has positive but not statistically significant effects on regional output growth in both countries when only migration appears in the regression equations. When their model is re-estimated with a college variable added, the implied rate of convergence increases, particularly in Norway, with the effects of migration changing signs (negative in Norway and positive in Sweden), although never statistically significant.

Some other studies suggest migration, particularly from poorer areas, could have contributed toward output per capita convergence. Mayoral and Garcimartin (2013) suggest that the deep reduction in steady-state disparities during the 1955-2008 period across Spanish regions can be largely attributed to the differences in population growth rates (migration flows). Ozgen, Nijkamp, and Poot (2010) review 12 past studies analyzing the effects of migration on growth regressions and find estimates of about 2.7% per year using neoclassical models of long-run real income convergence. Coates and Gindling (2013) study a panel of 3,101 U.S. counties with observations at five points in time over a 35-year period (1970, 1980, 1990, 2000, and 2005) and find strong support for the proposition that population growth caused by the increase in Hispanics has led to increased economic growth in those small and rural communities whose populations were declining in the 1970s or 1980s.

Recent empirical work on Mexico's regional income has found mixed results. Esquivel (1999) contains evidence of income convergence before NAFTA. Chiquiar (2005) finds that the divergent pattern started in the mid-1980s did not reverse with NAFTA, while Rodriguez-Oreggia (2005) finds evidence in favor of absolute 13-convergence for the period 1985-2000. Cabral and Mollick (2012) also document absolute and conditional real output per capita convergence following the inclusion of Mexico in NAFTA. On wages between 1990 and 2000, Chiquiar (2008) reports that regions more exposed to globalization appear to have exhibited wage increases. Cabral, Mollick, and Faiia (2010) propose a theoretical model of wages and identify a positive effect of FDI (a positive shock to labor demand) on real wages across states. None of these previous works on Mexican wages touches upon the speed of adjustment of wages to the steady-state. Furthermore, Cabral, Mollick, and Faria (2010) do not allow for a key variable in the determination of wages ("education" or years of schooling), nor do they deal with endogeneity in their dynamic panel estimations. We attempt to remedy these two problems in this article.

Building upon the work by Cabral and Mollick (2012), who estimated output per capita convergence, this article examines Mexico's real wage convergence across its 32 national entities for the post-NAFTA years. Assuming years of schooling as the basic fundamental of wages (based on the Mincerian approach, of which Hanson 2003 and Chiquiar 2008 provide evidence for individual wages in Mexico), we empirically allow for state-level forcing variables (FDI inflows and domestic and international migration flows) to endogenously impact the speed of adjustment in labor markets across Mexican regions. Our migration flow variables resemble those in the literature on immigrants joining the labor force, such as Borjas (2003) and Borjas (2006) for the United States, Mishra (2007) for Mexican emigrants, and Schmidt and Jensen (2013) for foreign labor inflow on Danish wages. As a whole, Longhi, Nijkamp, and Poot (2005) find such estimates to be around zero.

Our results show how fixed-effects models (FEM) that allow for state-specific characteristics present counter-intuitive responses. Employing dynamic panel data methods, however, we obtain the following results. First, education (or labor productivity) has slightly higher wage effects in the Border-North region. Second, allowing for foreign capital and labor to respond to wages, returns to education have higher effects in South-Center Mexico, the region with (average) lower education levels. Third, convergence rates become lower with endogenous foreign capital and migration flows: wages move faster in the South-Center region than in the Border-North. Overall, migration flows have greater effects on wages than FDI inflows.

The rest of this article proceeds as follows. Section II describes the dataset and shows a political map of Mexico describing the regional division adopted in this article. Section III introduces the methodology. Section IV discusses the empirical models of real wages employing alternative panel data techniques. Finally, Section V provides concluding remarks.


Our dataset runs from 1996 to 2006 (output) and from 1997 to 2006 (wages) across all of Mexico's 32 subnational entities. The data are compiled from various Mexican government agencies: real GDP per capita (1993 Mexican pesos) is from INEGI; state population figures, international migration rates and domestic migration rates are all from Consejo Nacional de Poblacion; FDI data are from the Ministry of Economy; and average years of schooling are obtained from the Ministry of Education. Wages are for employees enrolled in Mexico's social security system (Instituto Mexicano de Seguro Social [IMSS]). The reason for choosing this wage series relies on its intrinsic quality. Castellanos, Garcia-Verdi:1, and Kaplan (2004) suggest administrative data from the IMSS records offer several advantages over existing surveys for Mexico, such as the Encuesta Nacional de Empleo Urbano: first, with IMSS records it is known with certainty whether a worker remains employed with the same firm over time; second, the definition of the base salary is a comprehensive measure of wages and benefits and is consistently measured over time; and third, given that these wages are those on which employers pay payroll taxes, they are less subject to be measured with error. On the other hand, Aterido, Hallward-Driemeier, and Pages (2011) study Mexico's Seguro Popular (SP) introduced in 2002 to provide health insurance to Mexicans without social security and find that SP has reduced the inflow of workers into formality, which may have impacted wages reported by the social security administration. We therefore analyze, for robustness purposes, two other wage variables (minimum wages and maquiladora wages).

Minimum wages are daily wage rates that vary according to three different categories in which cities are classified. Maquiladora wages consist of average daily wages for white- and blue-collar workers. Both wage variables are expressed in 1993 real pesos. These two alternative variables are not devoid of measurement problems of their own. Minimum wages, for example, are not market-driven and maquiladora wages are only available with an unbalanced coverage across Mexican states.

In 1995, Mexican real GDP fell by more than 6%, which generated distortions we wish to avoid by restricting our analysis to the post-1996 years. Table 1 shows some descriptive statistics. Output per capita is the largest in Distrito Federal (DF), the capital of Mexico. Real wages are also the highest for DF and in five of the six states of the U.S.--Mexico border (the six U.S.--Mexico border states are: Baja California, Baja California Sur, Chihuahua, Coahuila, Nuevo Leon, and Sonora) real wages are higher than the national average.

The (domestic or international) migration rates are calculated as the difference between outflows and inflows of people over total state population: a positive number means a net outflow of people from the state. The ratio of FDI to GDP is the highest in Distrito Federal (8.57%), followed by Baja California (5.10%) and Nuevo Leon (4.82%), two Northern states. Meanwhile, the southern states of Oaxaca (0.03%) and Chiapas (0.02%) display the lowest FDI to GDP ratios.

Overall, Mexico is an unequal country with a less developed, rural south and a more dynamic, industrious north. Since the panel of border states would be too small (with only the six Mexican states bordering the United States), we follow Chiquiar (2008) and merge border states with other Northern states. Under this characterization, we have 13 states in the Border-North panel (6 in the Border and 7 in the North) and 19 states in the South-Center panel. The Center subregion itself comprises 12 states. Inspection of the two bottom rows of Table 1 makes the dissimilar characteristics of the two larger regions clear, with Border-North having lower IMSS wages but higher minimum and maquila wages, higher productivity, more education, twice as much FDI flows as a proportion of GDP, and higher migration. While the Border region (comprised of only six states bordering the United States) has higher figures than the other groups of three regions (except for international migration rate), the sample size of the panel data approach would be very limited if confined to the six states of the Border region. The differences in performance indeed suggest some sort of North-South duality, which to us is more meaningful than a focus of U.S.-Mexico border region versus non-border with respect to the adjustment of wages.

The geographical distribution of our two regions is shown in Figure 1. The bottom lines in Table 1 summarize the descriptive statistics of the four regions and the grouping into two larger regions: Border-North and South-Center. Average real wages (captured by the IMSS measure) are larger in the South-Center ($34.45) than in the Border-North ($33.17) region but output per capita is higher in the latter ($16,185) than in the former ($13,148). This suggests that labor and goods markets provide a mixed view across the two regions. When minimum wages are used, however, the Border-North region has slightly higher wages on average than the South-Center (9.87 versus 9.70, respectively), and the same happens for maquiladora wages (43.01 vs. 40.40, respectively). Data for maquiladora wages decompose labor into skilled (managers) and unskilled (workers) and suggest a skill premium of about 4 on average across states: $120.35/$33.27.

TABLE 1 Descriptive Statistics

State             IMSS  Minimum   Total   White-    Blue-    GDP Per
                  Wage    Wage   Maquila  Collar   Collar     Capita
                                   Wages   Maquila  Maquila
                                            Wages    Wages

Aguascalientes    33.59     9.63    36.02   110.53    31.87   17.918

Baja              39.00    10.75    55.00   159.05    46.87   19.208

Baja California   37.00    10.75    30.20    82.87    26.96   18.397

Campeche          39.72     9.63    26.99   106.23    23.90   23.185

Chiapas           27.01     9.63    44.31   139.15    38.48    6.323

Chihuahua         34.92     9.63                              20.910

Coahuila          34.82     9.63                              20.726

Colima            32.62     9.68    59.34   219.31    44.76   15.123

Distrito          53.98    10.75    45.25    97.71    38.96   36.455

Durango           25.99     9.63    31.42    77.67    28.84   12.912

Guanajuato        30.46     9.63    34.67   138.11    27.65   11.313

Guerrero          30.84     9.65    26.52    60.10    24.49    7.720

Hidalgo           31.59     9.63    42.61   107.05    38.11    8.965

Jalisco           35.75     9.66    57.96   180.43    41.06   14.487

Mexico            31.93     9.63                               8.615

Michoacan         37.07     9.63    37.11   113.49    31.43   13.335

Morelos           40.76     9.71    46.70   154.47    32.89   11.752

Nay ant           27.61     9.63                               8.801

Nuevo Leon        42.91     9.71    58.41   187.99    44.36   26.085

Oaxaca            29.35     9.63                               6.213

Puebla            35.69     9.63    30.55    93.36    27.40   10.004

Queretaro         42.15     9.63    47.88   139.35    37.34   17.277

Quintana Roo      31.83     9.63                              21.801

San Luis          33.13     9.63    44.10   160.27    35.88   11.085

Sinaloa           28.23     9.63    31.38   131.29    23.82   22.829

Sonora            31.45     9.96    49.95   161.70    41.40   17.952

Tabasco           31.42     9.63                               8.910

Tamaulipas        36.09    10.05    57.26   153.16    48.99   15.892

Tlaxcala          32.25     9.63    33.43    72.10    30.12    7.935

Veracruz          32.48     9.68    22.13    45.36    20.18    8.753

Yucatan           27.63     9.63    26.98    68.57    24.73   11.655

Zacatecas         26.42     9.63    28.95    94.54    25.63    8.690

Average           33.93     9.77    40.91   124.30    33.96   14.382

Border           36.532     9.96    55.92   167.77    46.03   20.129

North            30.282     9.79    34.22   111.63    29.27   12.805

Center           36.394     9.74    43.12   129.20    34.74   13.668

South            31.114     9.64    32.24    92.97    28.87   12.285

Border-North     33.166     9.87    43.01   134.37    36.06   16.185

South-Center     34.449     9.70    40.40   120.15    33.27   13.148

State            Population  Years of   FDI to  International
                 Growth      Schooling   GDP      Migration
                  Rate                  Ratio        Rate

Aguascalientes    2.07          8.16    0.01         0.44

Baja              3.33          8.40    0.05        -0.30

Baja California   3.07          8.56    0.04         0.33

Campeche          1.52          7.43    0.00         0.39

Chiapas           1.48          5.70    0.00         0.22

Chihuahua         1.38          7.96    0.03         0.68

Coahuila          1.42          8.63    0.01         0.39

Colima            1.58          7.91    0.00         0.54

Distrito          0.28          9.81    0.09        -0.05

Durango           0.68          7.56    0.01         0.96

Guanajuato        0.86          6.66    0.01         1.17

Guerrero          0.32          6.45    0.00         1.20

Hidalgo           0.85          6.92    0.00         1.05

Jalisco           1.17          7.81    0.02         0.60

Mexico            0.02          6.52    0.00         1.66

Michoacan         1.15          7.96    0.01         0.84

Morelos           1.79          8.31    0.02         0.30

Nay ant           0.54          7.50    0.02         1.35

Nuevo Leon        1.67          9.09    0.05         0.30

Oaxaca            0.34          5.99    0.00         1.12

Puebla            1.38          7.02    0.02         0.45

Queretaro         2.22          7.91    0.01         0.30

Quintana Roo      4.64          8.08    0.01        -0.79

San Luis          0.82          7.22    0.01         0.90

Sinaloa           0.50          7.92    0.00         0.93

Sonora            1.36          8.42    0.01         0.53

Tabasco           0.93          7.44    0.01         0.42

Tamaulipas        1.60          8.30    0.02         0.58

Tlaxcala          1.76          7.88    0.01         0.33

Veracruz          0.54          6.73    0.00         0.58

Yucatan           1.55          7.09    0.00         0.01

Zacatecas         0.18          6.74    0.00         1.51

Average           1.34          7.63    0.02         0.59

Border            1.79          8.47    0.03         0.36

North             1.12          7.66    0.01         0.92

Center            1.13          7.62    0.02         0.65

South             1.54          6.88    0.00         0.37

Border-North      1.43          8.03    0.02         0.66

South-Center      1.28          7.35    0.01         0.55

State            Domestic

Aguascalientes       -0.49

Baja                 -1.11

Baja California      -1.57

Campeche             -0.22

Chiapas               0.35

Chihuahua            -0.22

Coahuila             -0.02

Colima               -0.55

Distrito              0.93

Durango               0.18

Guanajuato           -0.05

Guerrero              0.41

Hidalgo              -0.27

Jalisco              -0.02

Mexico                0.08

Michoacan            -0.39

Morelos              -0.31

Nay ant              -0.26

Nuevo Leon           -0.28

Oaxaca                0.23

Puebla               -0.01

Queretaro            -0.67

Quintana Roo         -1.78

San Luis              0.06

Sinaloa               0.32

Sonora               -0.10

Tabasco               0.39

Tamaulipas           -0.49

Tlaxcala             -0.27

Veracruz              0.37

Yucatan              -0.07

Zacatecas             0.11

Average              -0.18

Border               -0.37

North                -0.24

Center               -0.10

South                -0.10

Border-North         -0.30

South-Center         -0.10

Numbers in hold indicate the national averages of the 32
suhnational entities and the averages of the two regions,
Border-North and South-Center, upon which the analysis
of this study is based.

Population not only grows faster in the Border-North (1.43%) than in the South-Center (1.28%), but is also more educated (8.03 vs. 7.35 average years of schooling, respectively). With respect to capital and labor mobility, the Border-North receives on average more FDI relative to GDP (2%) than the South-Center (1%). (2) The Border-North also shows higher international migration than the South-Center (net international migration of 0.66 and 0.55, respectively) and higher net domestic migration inflows (-0.30 and -0.10, respectively). Owing to the nature of international and domestic migration, more people decide to move abroad than into other states of Mexico, with only three states (Baja California, Distrito Federal, and Quintana Roo) having larger international inflows than outflows due to tourism and business activities.(3)

As Table 2 indicates, the three measures of wages are positively correlated, varying from a low of 0.273 between minimum wage and maquila wages to 0.559 between minimum wage and IMSS wages. High correlation coefficients between years of schooling and GDP per capita (0.822) and between FDI/GDP and GDP per capita (0.736) are observed, as well as a more moderate correlation coefficient between years of schooling and FDI/GDP (0.642). The empirical models will take into account these properties. For the explanatory variables to be used in the regression models for wages below, we do not have correlation coefficients greater than 0.50. In particular, each fundamental of the empirical model (either education levels or labor productivity) is postulated to affect real wages, jointly with labor supply shifts (either international or domestic migration rates) and labor demand shocks caused by more inflows into Mexico as measured by the FDI/GDP ratio at the state level. Also, the real exchange rate (which does not vary by state and is thus omitted from Table 2) captures international forces and the overall competitiveness of the Mexican peso.

TABLE 2 Correlogram

                IMSS   Minimum   Total   White-Collar  Blue-Collar
                Wage     Wage   Maquila    Maquila       Maquila
                                  Wages      Wages        Wages

1MSS wage           1

Minimum wage    0.559        1

Total Maquila   0.480    0.273        1

White-collar    0.286    0.094    0.846          1
Maquila wages

Blue-collar     0.460    0.352    0.956      0.735           1
Maquila wage

GDP per         0.772    0.557    0.359      0.224       0.363

Population      0.140    0.295    0.195      0.140       0.213
growth rate

Years of        0.765    0.556    0.488      0.332       0.463

FD1 to GDP      0.662    0.582    0.295      0.099       0.326

International  -0.505   -0.398   -0.289     -0.130      -0.314

Domestic       -0.078   -0.282   -0.214     -0.202      -0.221

               GDP per  Population  Years of   FDI to  International
               capita     Growth    Schooling    GDP       Migration
                            Rate                Ratio       Rate

1MSS wage

Minimum wage

Total Maquila

Maquila wages

Maquila wages

GDP per              1

Population       0.153         1
growth rate

Years of         0.822     0.231          1

FD1 to GDP       0.736     0.201      0.642         1

International   -0.476    -0.594     -0.435    -0.469           1

Domestic        -0.039    -0.843     -0.235    -0.053       0.192


1MSS wage

Minimum wage

Total Maquila

Maquila wages

Maquila wages

GDP per

growth rate

Years of

FD1 to GDP


Domestic           1


Theoretical models along the lines of Barro. Mankiw, and Sala-l-Martin (1995) suggest channels upon which output (or "income" in most studies) convergence may be affected. For the case of a small open economy with capital intensity below the steady-state, Rappaport (2005) shows that outmigration directly contributes to faster income convergence but also creates a disincentive for gross capital investment. For low income levels, the latter effect dominates and labor mobility actually slows down the speed of income convergence.(4) The net effect is ambiguous.

Real wages must always converge provided there is sufficient labor market flexibility across regions, while real output per capita may either converge (because of diminishing returns assumed in neoclassical models) or not converge (according to the endogenous growth theory). Support for open-economy versions of the neoclassical growth model can be found in Caselli, Esquivel, and Lefort (1996), who identify a rate of output convergence of countries at about 10% in dynamic panels. Evidence for wage regressions is more mixed. An important dimension we explore in this article is that "income convergence" referred to by studies may actually indicate "per capita output convergence" or "wage convergence." These are not conceptually the same since wages and profits determine national income, which is equal to output in the expenditure approach. Empirically, there are two sources for the asymmetry of convergence across regions: there are the FDI inflows to the Border-North and there is more domestic and international migration, especially after NAFTA and the currency crisis of 1994-1995.

We test initially for absolute convergence in wages. Wages are a function of state-specific effects and lagged wages:

(1) [] = + [u.sub.i] + [[gamma] - 1] + []

where [u.sub.i] is state (indexed by i) fixed effects and [] is an idiosyncratic error term. Following Islam (1995) the implied rate of convergence, X. is calculated as the negative of the log for the lagged per capita wages coefficient, y. Absolute convergence exists if 0 < [gamma] < 1 ; < [gamma] > 1 would suggest divergence.(5)

For wage equations, we consider two basic fundamentals: human capital (years of education) and real per capita output (a measure of labor productivity). Works grounded on the Min-cerian tradition for Mexican individual wages include Hanson (2003) and Chiquiar (2008). The alternative model looks at real per capita output, which is a measure of labor productivity, thus connecting goods and labor markets. As productivity increases, wages should increase. At the individual level, Card (2001) assumes for U.S. cities that the city-specific productivity effect is a function of common occupation effect, a city effect, and an occupation and city-specific productivity term. At the sector level, Mollick and Cabral (2009) find that productivity has a positive effect on employment for 25 industries of Mexican manufacturing from 1984 to 2000. At the country level, Felbermayr, Hiller, and Sala (2010) allow for the stock of immigrants and geographical factors to be endogenous to per capita GDP in income regressions. Using the labor productivity at the state level in this article, wages paid should be higher when productivity increases.

It is important to employ as well a number of state control variables in order to minimize problems of "omitted variables.- The focus in this article is on FD1 to capture foreign capital inflows and state migration to capture labor dynamics. Studies using these two forces separately include Cabral, Mollick, and Faria (2010) for FDI effects on wages in Mexico and DiCe-cio and Gascon (2010) for income convergence combined with U.S. state migration. Since large currency fluctuations may have exposed Mexico to international markets more than trade reforms, we control--as in Verhoogen (2008)--for the competitiveness of the Mexican peso as measured by the real exchange rate. Robertson (2003) finds that real exchange rates correlate with real wages in Mexico. The models employed to estimate conditional convergence (omitting the constant term) have the following forms:

(2) [] = [[gamma]W.sub.1t - 1] + [alpha] [] + [[beta].sub.j] [] + [[member of]] and

(3) [] = [[gamma] - 1] + [alpha][(Y / N)] +[E[beta].sub.j][] + [[zeta]]

where EDU stands for years of schooling, YIN stands for GDP per capita, and Xi, is the group of j = 1 to k forcing variables: FDI/GDP, migration rates (domestic and international), and other controls. Several other regressors, such as the share of government expenditures to GDP, the proportion of rural population, the number of telephone lines, and the share of manufacturing GDP on total state GDP, are excluded due to their lack of statistical significance as additional controls. We exclude population growth from the estimations due to the strong negative correlation of this variable with foreign and domestic migration rates (-0.83 and -0.73, respectively). The interpretation on [gamma] is as discussed next to Equation (1).

Our migration variables embedded in the vector X of controls resemble those in the literature on immigration joining the labor force. Borjas (2003) uses decennial U.S. census data for 1960-1990 and annual supplement of the CPS for pooled 2000 wage data of men aged 18-64 and finds that immigration lowers wages: a 10% increase in supply reduces wages by 3% to 4%. Mishra (2007) looks at the Mexican case using data from the Mexican and U.S. censuses from 1970 to 2000 and finds that a 10% decrease in the number of Mexican workers due to emigration in a group formed by schooling and experience characteristics increases the average wage in that skill group by 4%. See also Hanson (2003) and Chiquiar (2008) for Mincerian-type studies for the Mexican case.(6)

The Generalized Method of the Moments (GMM) estimator solves the consistency problem of ordinary least squares (OLS) estimators in Equation (1), taking first-differences, removing the state effects (pi), and producing an equation that is estimable using instrumental variables in which endogenous explanatory variables are instrumented with suitable lags of their own. Blundell and Bond (1998) propose a model in which lagged differences are employed in addition to the lags of the endogenous variables, producing more robust estimations when the autoregressive processes becomes persistent. GMM estimators are said to be consistent if there is no second order autocorrelation in the residuals by the Areliano-Bond (AB(2)) test and if the instruments employed are valid by the Sargan or Hansen's J-test.

The identifying assumption for the reverse causation patterns in this article is that wages are allowed to impact FDI inflows and migration flows; all else is assumed to be exogenous. In order to avoid overidentification problems, the instrument set is constrained in two different ways. First, we limit the number of lags of the dependent variable and the endogenous regressors to three.(7) In addition, we restrict our specification to one instrument for each lag distance and instrumenting variable. Roodman (2009) provides details on this procedure.


A. Main Findings

Table 3 presents the estimates of absolute convergence corresponding to Equation (1) using several econometric methods. For all wage measures, the lagged per capita wage coefficient ([gamma]) is statistically significant and < 1, which implies a positive rate of convergence. In columns I, 5, 9, and 13 we employ first a pooling of time series and cross-sectional observations estimated through OLS. Accounting for state unobservable fixed effects in columns 2, 6, 10, and 14, the rate of convergence (implicit A) becomes higher in all cases, more than doubling in case of IMSS wages and more than tripling in maquiladora wages. With dynamic panels, these figures vary from 4.3% to 6.8% for IMSS wages and from 17.1% to 10.6% for maquila wages, depending on whether we employ dynamic or system GMM techniques (DGMM or SGMM, respectively). The estimates of the rate of convergence of minimum wages are very stable (around 19% per year), reinforcing the nature of staggered rises in minimum wages or even wage stickiness. We also report in the last four columns the results for output per capita convergence for comparison purposes. Output per capita across states moves in between the speeds of convergence of IMSS and maquila wages: from 7.8% in SGMM by column 16 to 10.1% in DGMM by column 15.

TABLE 3 Absolute Convergence

                             IMSS Wage

                OLS        FE       DGMM      SGMM      OLS
Variables       (1)       (2)       (3)       (4)       (5)

Lagged real   0.981***  0.957***  0.958***  0.934***   0827***
wages          (0.009)   (0.015)   (0.013)   (0.014)   (0.002)

Lagged per

Constant      0.093***  0.175***  0 172***  0.259***  0.690***
               (0.033)   (0.051)   (0.047)   (0.049)   (0.006)

Implicit         0.019     0.044     0.043     0.068     0.189

No. of             288       288       256       288       288

No. of                                  21        30

Hansen                             31.8751   31.9790
                                   [0.045]   [0.275]

AB(2)                              -3.8806   -3.8315
                                   [0.000]   [0.000]

                 Minimum Wage                         Maquiila

               FE          DGMM      SGMM      OLS        FE
Variables      (6)           (7)       (8)       (9)     (10)

Lagged real   0.825***  0 825***  0.824***  0.921***  0.756***
wages          (0.001)   (0.001)   (0.002)   (0.027)   (0.037)

Lagged per

Constant      0.696***  0.697***  0.698***  0.807***  2.378***
               (0.005)   (0.005)   (0.006)   (0.258)   (0.356)

Implicit         0.191     0.191     0.192     0.082     0.276

No. of             288       256       288       166       166

No. of                              17        25

Hansen                         31.9998   31.9998
                               10.006]   10.1001

AB(2)                           4.1644    4.1703
                               [0.000]   [0.000]

                                             Output Per Capita

                DGMM      SGMM      OLS       FF.       DGMM
Variables       (11)      (12)      (13)      (14)      (15)

Lagged real   0.842***  0.899***
wages          (0.042)   (0.040)

Lagged per                        0.997***  0.831***  0.904***
capita                             (0.004)   (0.025)   (0.026)

Constant      1.557***  1.011***     0.050  1.616***  0.931***
               (0.402)   (0.380)   (0.037)   (0.239)   (0.244)

Implicit         0.171     0.106     0.003     0.185     0.101

No. of             141       166       288       288       256

No. of              16        22                            21

Hansen         23.5227   23.9234                       3).2764
               [0.052]   [0.246]                       [0.052]

AB(2)          -0.8436   -0.8073                        0.0161
               [0.399]   [0.420]                       [0.987]

Variables       (16)

Lagged real

Lagged per      0.925***
capita           (0.027)

Constant      1.0.726***

Implicit           0.078

No. of               288

No. of                30

Hansen           31.9459

AB(2)             0.0176

Notes: One-step robust estimators are reported for DGMM and SGMM
estimations as proposed by Blundell and Bond (1998). The Hansen test
reports that under the null the overidentified restrictions are valid.
AB<2) corresponds to the Arellano-Bond test for serial correlation,
under the null of no autocorrelation. Robust standard errors are
reported in parentheses.

*Significant at 10%; **significani at 5%: ***significant at 1%.

For the DGMM equations in 'MSS wages the Hansen's J-test rejects the null of no misspecifi-cation and the AB second-order autocorrelation tests reject the null of no serial correlation in the idiosyncratic error for both real wage equations in columns 3 and 4. These estimates in Table 3 illustrate considerable variability in convergence rates and suggest misspecification problems overall, possibly caused by "omitted variables."

Owing to the strong persistence of wages and output per capita ([gamma] - 1) inferred from Table 1, SGMM will be used throughout the rest of the article in order to better assess conditional convergence. This choice of SGMM as the preferred estimator is based on diagnostic tests to our dataset of Mexican states, with the causation mechanisms to be specified below.(8)

In Table 4 we estimate static versions of Equation (2) by FEM as a preamble to the dynamic panels.(9) While these results provide no insight on convergence, they allow us to observe the short-run impact of our control variables and compare them later on with their long-term impacts under dynamic panel methods. Panel A contains the Border-North region and Panel B the South-Center region. In this model, state unobservable characteristics are allowed to covary with the (RHS) regressors but strict exo-geneity of regressors is assumed. Because of the 0.642 correlation coefficient between education and FDI/GDP ratio, we omit the latter for sensitivity checks in the right side of the table. The results are very much preserved but, by omitting FDI/GDP, we know that they are not driven by this positive correlation. We find that returns to education are positive as expected, but the coefficients are always greater than one, except for minimum wages. Across regions, the coefficients of years of schooling do not vary much for IMSS wages (from 1.789 in Border-North to 1.691 in South-Center). Similar values for all the maquiladoras vary from 4.137 in Border-North to 3.066 in South-Center

TABLE 4 Conditional Convergence Under Static Panel Data Methods

                                                 White-     Blue-
                                       Total    Col lar     Collar
                  IMSS      Minimum    Maquila   Maquila    Maquila
Variables         Wage       Wage       Wages     Wages      Wages

Panel A.

Log of years    1.789***   0.217***   4.137***  6.465***   7.680***
of schooling     (0.105)    (0.044)    (0.392)   (1.013)    (0.469)

FD1 to GDP        -0.404    -0.408*      0.261     1.972     -0.200
ratio            (0.269)    (0.222)    (0.525)   (1.450)    (0.690)

International   -0.103**      0.005      0.051    -0.185     -0.037
migration        (0.046)    (0.035)    (0.095)   (0.221)    (0.145)

Domestic        -0.113**     -0.035     -0.174   -0.536*     -0.054
migration        (0.053)    (0.039)    (0.129)   (0.283)    (0.192)

Real exchange  -0.002***  -0.001 **  -0.007***    -0.002  -0.014***
rate             (0.000)    (0.000)    (0.001)   (0.003)    (0.001)

Constant          -0.103   1.844***      1.292    -3.001  -6.110***
                 (0.234)    (0.101)    (0.822)   (2.231)    (0.979)

No. of               130        130         79        79         79

[R.sup.2]          0.904      0.321      0.852     0.695      0.948

[R.sup.2]          0.706      0.539      0.285     0.123      0.272

[R.sup.2]          0.728      0.477      0.403     0.175      0.412

Panel B.

Log of years    1.691***   0.337***   3.066***  2.578***   6.029***
of schooling     (0.070)    (0.037)    (0.377)   (0.436)    (0.320)

FDI to GDP        -0.065     -0.199      1.877    1.869*    2.513**
ratio            (0.147)    (0.150)    (1.139)   (0.978)    (1.188)

International  -0.107***  -0.055***  -0.759***  -0.604**  -0.668***
migration        (0.027)    (0.019)    (0.243)   (0.234)    (0.225)

Domestic       -0.068***     -0.022     -0.563    -0.263   -0.631 *
migration        (0.026)    (0.020)    (0.361)   (0.314)    (0.357)

Real exchange  -0.001***  -0.001***  -0.006***    -0.003  -0.013***
rate             (0.000)    (0.000)    (0.002)   (0.002)    (0.002)

Constant          0.308*   1.682***   4.194***  5.977***   -1.884**
                 (0.177)    (0.054)    (0.963)   (0.965)    (0.799)

No. of               190        190        112       112        112

[R.sup.2]          0.902      0.486      0.651     0.423      0.899

[R.sup.2]          0.471      0.124      0.146     0.051      0.129

[R.sup.2]          0.510      0.133      0.173     0.054      0.175

                                                  White-       Blue-
                                       Total      Collar       Collar
                 IMSS      Minimum    Maquila    Maquila    Maquila
Variables        Wage       Wage       Wages      Wages      Wages

Panel A.

Log of years    1.788***   0.215***   4.120***   6.339***   7.693***
of schooling     (0.105)    (0.046)    (0.391)    (1.020)    (0.462)

FD1 to GDP

International   -0.100**      0.007      0.047     -0.211     -0.034
migration        (0.048)    (0.038)    (0.095)    (0.221)    (0.142)

Domestic         -0.104*     -0.025     -0.183   -0.610**     -0.046
migration        (0.054)    (0.043)    (0.127)    (0.271)    (0.186)

Real exchange  -0.002***   -0.001**  -0.008***     -0.002  -0.014***
rate             (0.000)    (0.000)    (0.001)    (0.003)    (0.001)

Constant          -0.102   1.845***      1.329     -2.717  -6.139***
                 (0.235)    (0.106)    (0.818)    (2.248)    (0.963)

No. of               130        130         79         79         79

[R.sup.2]          0.903      0.286      0.851      0.690      0.948

[R.sup.2]          0.715      0.505      0.280      0.110      0.273

[R.sup.2]          0.734      0.445      0.400      0.164      0.413

Panel B.

Log of years    1.689***   0.332***   3.178***   2.690***   6.179***
of schooling     (0.069)    (0.037)    (0.386)    (0.441)    (0.344)


International  -0.106***  -0.050***  -0.856***  -0.701***  -0.798***
migration        (0.025)    (0.019)    (0.214)    (0.204)    (0.218)

Domestic       -0.067***     -0.019   -0.684**     -0.383   -0.792**
migration        (0.025)    (0.020)    (0.333)    (0.292)    (0.365)

Real exchange  -0.001***   -0.001**  -0.006***    -0.004*  -0.014***
rate             (0.000)    (0.000)    (0.002)    (0.002)    (0.002)

Constant          0.309*   1.684***   4.085***   5.868***   -2.030**
                 (0.177)    (0.051)    (0.986)    (1.000)    (0.859)

No. of               190        190        112        112        112

[R.sup.2]          0.902      0.480      0.638      0.409      0.892

[R.sup.2]          0.476      0.173      0.149      0.070      0.133

[R.sup.2]          0.514      0.167      0.172      0.069      0.170

Notes: Newey-West heteroskedasticity and autocorrelation robust

standard errors are reported in parentheses.
*Significant at 10%: **significant at 5%; ***significant at 1%.

In general, shift factors such as capital inflows and migration do not have statistically significant effects, except for a negative effect of FDI/GDP in minimum wages at 10% in Border-North and positive effects for skill types of maquiladora workers in South-Center that are not preserved for the column of total maquila wages. More importantly, migration has negative effects against the theoretical predictions, particularly international migration in the South-Center region: with more outflows and all else constant, wages fall in that region rather than moving up. The explanatory power ranges from 40.3% in total maquila wages to 72.8% in IMSS wages in Panel A and from 17.3% in total maquila wages to 51% in IMSS wages in Panel B. Overall, the magnitude of the returns to education coefficient seems too high, and none of the shift factors of demand and supply of labor makes sense. Perhaps the only variable that behaves as expected in the FEM models of Table 4 is the one that does not vary across states: a real depreciation of the peso implies lower real wages.(10)

In Table 5 we present the results of estimating Equations (2) and (3) under dynamic panel data methods while considering the possibility of reverse causation from wages to FDI and to migration flows. In order to verify in detail the role of reverse causation we proceed gradually. The first two columns assume that all regressors are exogenous; the following two columns assume that the foreign capital and migration regressors are endogenous. The latter set of estimates allows for the

possibility that, for example, when wages are low in a given state people move away from that state in search of higher wages. For the wage equations in the upper part of Table 5 we use model (2) above with years of schooling as the basic economic fundamental. For the wage equations in the lower part of Table 5 we use (3) with GDP per capita as the economic fundamental. We will report in the article only the results for IMSS wages, but the estimates for minimum wages and maquiladora wages are available upon request to the authors.

TABLE 5 Dynamic Panel Data (SGMM) Models for IMSS Wages

               Exogenous             Endogenous

Variables       Border-     South-       Border-     South-
               North (1)    Center     North (3)     Center
                             (2)                      (4)

Panel A. With education

Lagged real     0.602***   0.658***    0.836***   0.765***
wages            (0.064)    (0.062)     (0.065)    (0.066)

Log of years    0.805***   0.516***     0.255**    0.297**
of schooling     (0.128)    (0.119)     (0.104)    (0.128)

FDI to GDP        -0.052     -0.072      -0.176      0.407
ratio            (0.092)    (0.240)     (0.200)    (0.455)

International      0.023     0.029*      -0.006   0.020***
migration        (0.034)    (0.016)     (0.012)    (0.007)

Domestic          -0.004     -0.016      -0.007      0.008
migration        (0.034)    (0.024)     (0.013)    (0.011)

Real exchange  -0.003***  -0.003***   -0.003***  -0.003***
rate             (0.000)    (0.000)     (0.000)    (0.000)

Constant          -0.056   0.366***     0.305**   0.460***
                 (0.180)    (0.062)     (0.123)    (0.081)

Implicit           0.507      0.419       0.179      0.268

No. of            117-31     171-31      117-22     171-22

AB(2)              0.612     -0.830      -0.064     -0.588

p Value          [0.541]    [0.406]     [0.949]    [0.556]

Hansen            12.798     18.361      12.496     17.002

p Value          [0.969]    [0.785]     [0.328]    [0.108]

Panel B. With productivity

Lagged real     0.857***   0.849***    0.838***   0.892***
wages            (0.050)    (0.026)     (0.054)    (0.027)

Log of per      0.178***    0.148**    0.168***      0.015
capita output    (0.044)    (0.075)     (0.059)    (0.040)

FD1 to GDP       0.177**      0.244      -0.143      0.740
ratio            (0.076)    (0.166)     (0.235)    (0.678)

International   0.072***     0.045*      0.044*    0.015**
migration        (0.026)    (0.026)     (0.026)    (0.006)

Domestic       0.051 ***      0.033      0.037*     -0.010
migration        (0.017)    (0.029)     (0.019)    (0.015)

Real exchange  -0.003***  -0.003***   -0.003***  -0.003***
rate             (0.000)    (0.000)     (0.000)    (0.000)

Constant       -0.99S***     -0.668     -0.798*      0.483
                 (0.316)    (0.656)     (0.461)    (0.305)

Implicit           0.154      0.164       0.177      0.114

No. of            117-31     171-31      117-22     171-22

AB(2)              1.235     -0.140       1.186     -0.479

p Value          [0.217]    [0.888]     [0.236]    [0.632]

Hansen            12.569     18.475      12.681     17.898

p Value          [0.973]    [0.779]     [0.627]    [0.268]

               Exogenous             Endogenous

Variables       Border-     South-       Border-    South-
               North (5)    Center     North (7)    Center
                             (6)                    (8)

Panel A. With education

Lagged real     0.606***   0.660***    0.838***   0.756***
wages            (0.062)    (0.064)     (0.065)    (0.065)

Log of years    0.797***   0.511***     0.250**    0.329**
of schooling     (0.122)    (0.127)     (0.097)    (0.128)


International      0.023     0.030*      -0.005    0.020**
migration        (0.034)    (0.017)     (0.015)    (0.008)

Domestic          -0.003     -0.015       0.002      0.012
migration        (0.034)    (0.023)     (0.018)    (0.012)

Real exchange  -0.003***  -0.003***   -0.003***  -0.003***
rate             (0.000)    (0.000)     (0.000)    (0.000)

Constant          -0.053   0.369***     0.305**   0.437***
                 (0.179)    (0.064)     (0.141)    (0.090)

Implicit           0.501      0.416       0.177      0.280

No. of            117-30     171-30      117-18     171-18

AB(2)              0.589     -0.855      -0.036     -0.590

p Value          [0.5561    [0.392]     [0.971]    [0.555]

Hansen            12.931     18.457      12.500     17.499

p Value          [0.967]    [0.780]     [0.130]    [0.025]

Panel B. With productivity

Lagged real     0.853***   0.854***    0.837***   0.881***
wages            (0.058)    (0.027)     (0.065)    (0.034)

Log of per      0.184***     0.143*     0.159**      0.054
capita output    (0.047)    (0.076)     (0.070)    (0.038)

FD1 to GDP

International     0.070*    0.041 *      0.040*   0.022***
migration        (0.041)    (0.022)     (0.024)    (0.007)

Domestic           0.045      0.030       0.044     -0.000
migration        (0.032)    (0.035)     (0.028)    (0.015)

Real exchange  -0.003***  -0.003***   -0.003***  -0.003***
rate             (0.000)    (0.000)     (0.000)    (0.000)

Constant       -1 028***     -0.627      -0.697      0.159
                 (0.347)    (0.655)     (0.527)    (0.268)

Implicit           0.159      0.158       0.178      0.127

No. of            117-30     171-30      117-18     171-18

AB(2)              1.395     -0.151       1.194     -0.262

p Value          [0.163]    [0.880]     [0.232]    [0.793]

Hansen            12.821     18.595      12.665     18.517

p Value          [0.969]    [0.773]     [0.394]    [0.101]

Notes: Heteroskedasticity robust standard errors are shown in
parentheses. The Hansen test reports that under the null the
overidentified restrictions are valid. AB(2) corresponds to the
Arellano-Bond test for serial correlation, under the null of
no autocorrelation. The p values for the Hansen test and the
Arellano-Bond test of second order autocorrelation (AB(2))
are shown in square brackets.

* Significant at 10%: * significant at 5%: *** significant at 1%

In Panel A of Table 5 wage persistence seems to be in the 0.60 to 0.66 range in the first two columns and 0.77 to 0.84 in columns 3 and 4. We also find that returns to education are positive as expected. Contrary to the values of FEM in Table 4, the coefficients are always less than one in the dynamic panels. The treatment of reverse causation has also an impact on the magnitude of the coefficients. For example, in the first two columns the coefficients of years of schooling vary from 0.805 in Border-North to 0.516 in South-Center under exogenous regressors. Allowing for wage effects on migration and foreign capital flows the coefficients of years of schooling vary from 0.255 in Border-North to 0.297 in South-Center. It follows that returns to education are more valued in the South-Center than the Border-North, which is consistent with a less educated population in the former (8 years of education in Border-North vs. 7.35 years in South-Center according to Table 1). An increase of 1% in the years of education leads to a nearly 0.3% increase in wages in South-Center; a similar response in the Border-North is 0.255%. Removing the FDUGDP ratio in the right section of the table suggests somewhat wider responses of 0.25% and 0.33%, respectively. In contrast to Table 4, in the region with less education levels, increases in investment in education yield proportionally more rewards.

Perhaps surprisingly, in Table 5 we do not find statistically significant coefficients for FDI/GDP ratios. Cabral, Mollick, and Faria (2010) identify a positive result of FDI on wages under a dynamic specification in which no reverse causation is assumed. Under endogenous regressors in Table 5, the migration coefficients are statistically significant in column 4 for international migration for the panel of South-Center states. As more outflows to international areas increase, wages in the region increase by a very small amount (0.020) in Panel A, yet statistically significant at 1%, all else constant. Domestic migration, on the other hand, has no clear effects on local wages. As before, an assumed exogenous real depreciation of the peso implies lower real wages, now at around--0.003.

How sensitive are these results to the treatment of education as the basic fundamental of wages? Panel B looks at this question using GDP per capita as the main fundamental of wages, while still allowing labor market shift forces to take place both exogenously and endogenously. It follows that wage persistence is higher in this model from a low of 0.838 in column 3 to a high of 0.892 in column 4. We also find that productivity effects are positive but vary across regions: in response to a 1% increase in labor productivity wages increase by 0.168% in Border-North but do not respond in the South-Center. For exogenous regressors in columns 1 and 2, the responses are larger: 0.178% in Border-North and 0.148% the South-Center. Allowing for wage effects on migration and foreign capital flows we still do not find statistically significant coefficients for FDI/GDP ratios.

There is an important difference with the education model. Under endogenous regressors, the migration coefficients of Panel B in Table 5 are statistically significant in columns 3 and 4 for international migration, with a larger effect for the panel of Border-North states. As more outflows to international areas increase, wages increase by 0.044 (statistically significant at 10%) in Border-North and by 0.015 (statistically significant at 10%) in South-Center. And domestic migration has now a positive effect on wages in the Border-North region by 0.037 (statistically significant at 10%). Migration flows have positive, yet small, effects on real wages.

Looking at convergence, on the wage regressions of Table 5 the implied X's tend to be higher when all regressors are assumed exogenous than when all regressors are assumed endogenous." Our most likely explanation is that ignoring reverse causation mechanisms from wages to migration leads to an overestimated speed of adjustment coefficient. Since it is very likely that labor and capital respond to wage and output fluctuations, we concentrate on the estimated X's when FDI/GDP and migration flows are endogenous to wages. In columns 1 and 2 of Panel A we observe that the Border-North region presents a rate of wage convergence of 51% (obtained by--In 10.6021), above the 42% (obtained by--In [0.658]) observed across South-Center states. If true, this would imply for the Border-North region that wages converge in only 2 years! Recall that Table 3 contains absolute convergence slightly below 7% in column 4. This increase in the speed of convergence is achieved by the role of education in determining wages. In columns 3 and 4 of Panel A we observe that the Border-North region presents a rate of wage convergence of 18%, below the 27% observed across South-Center states. However, in columns 3 and 4 of Panel B we observe that the Border-North region still presents a rate of wage convergence of 18%, now above the 11% observed across South-Center states. The difference is due to the productivity variable which is higher in the Border-North: $16,185 versus $13,148 on average as discussed in Table 1. In Table 5, Panel B productivity is only statistically significant in the Border-North region when regressors are assumed endogenous. This is what makes the speed of adjustment considerably higher in the Border-North region: wages there respond to movements in labor productivity, which does not happen in the South-Center.

B. Robustness Checks

As a sensitivity analysis to the basic reverse causation mechanism assumed in the dynamic panels, we further allow for wages having an impact on education, such that the returns to education of Mincerian equations operate in reverse: when wages increase, individuals decide to study more. This modification, however, does not change in any way the qualitative results above and we prefer to keep the Mincerian assumption that education is exogenous to wages with only shocks to demand and supply of labor responding to wage fluctuations.

We also replicate the estimations of Table 5 for the two alternative wages. For the maquiladora wages, the results are less conclusive than those for IMSS since only under exogenous regressors the education (or productivity) coefficient has positive and statistically significant values in Border-North. Other coefficients lose significance under endogenous regressors, except for a negative effect for international migration rate in the South-Center region: -0.157 (statistically significant at 10%) for the model with years of schooling and -0.250 (statistically significant at 5%) for the model with productivity.(12) Owing to space constraints, we make these sets of results for maquiladora wages (and for minimum wages) available upon request.

Following the suggestion of an anonymous referee, we pool the data and introduce a dummy variable (1 if the Mexican state borders U.S. states; and 0 otherwise). These estimates are available upon request but we summarize them briefly here. The sample size increases to 288 and we now have a pooled sample of all Mexican states. In the model with education, with exogenous regressors the coefficient on lagged wages is about 0.58 and the coefficient on education is estimated around 0.74. With endogenous regressors, the coefficient on lagged wages is about 0.71 or 0.74 and the coefficient on education is about 0.42 or 0.49, with no other statistically significant coefficients than the real exchange rate. The border dummy is found to be sometimes negative and significant in the model with education but is never statistically significant in the model with productivity. In all cases, however, we find serious problems with instrument validity (Hansen's p value is close to zero) under endogenous regressors. With exogenous regressors, the specification problems are less severe but borderline: p values vary between 0.105 and 0.107. Given these facts, combined with our research design through Mexican regions, we do not report this modification to models (2) and (3).


Examining the recent Mexican post-NAFTA experience, our conditional convergence model allows for key determinants of wages (either years of schooling or labor productivity) and forcing variables. In this article, we go beyond the focus on the rate of population growth and examine the net outflows of people from states together with state-bound FDI inflows. The floating real exchange rate adopted in the mid-1990s seems to perform well in wage equations handling time-varying effects and the international competitiveness of the peso.

Taken together, the results suggest that the Mexican labor market--captured by the excellent wage data based on social security payrolls as documented by Castellanos, Garcia-Verdti, and Kaplan (2004)--moves at faster rates in the South-Center to the steady-state in the model with education. The results change a little when labor productivity is used, although the speed of convergence in Border-North is unchanged at about 18% per year. With respect to the conjecture in Robertson (2000), we find support for some (yet small) effects of migration as the main force behind the adjustment of wages. Our estimates for migration are, however, much lower than those reported by Borjas (2003) and Bor-jas (2006) for the United States, Mishra (2007) for Mexican emigrants, and perhaps Schmidt and Jensen (2013) for foreign labor inflow on wages of Denmark. In contrast, since FDI flows do not have any impact on regional wages, the findings of Cabral, Mollick, and Faria (2010) appear to be sensitive to the reverse causation channel of wages affecting flows of (foreign) capital and labor across states of Mexico. In this reexamination we do not find FDI effects on wages, other than when output per capita is used under exogenous regressors for the panel of Border-North states. Allowing for a human capital channel (in which average years of schooling is 8.03 in Border-North states versus 7.35 in South-Center) reduces labor demand effects attributed to capital inflows as reported previously by Cabral, Mol-lick, and Faria (2010).

Our time span extends until 2006 and captures the period right after NAFTA taking effect in 1994 and the currency crisis of 1995 that may have contributed to higher labor migration toward higher wages and better business opportunities across states. After the U.S. financial crisis of 2008-2009, a period not included in our sample, international migration from Mexico has substantially reduced. It is also possible that domestic migration in Mexico has also reduced accordingly in more recent years. It is therefore quite likely that the responses reported in this study may change with availability of data for these more recent years. In any case, the period chosen to end right before the financial crisis is the one most likely to incur a greater extent of labor mobility, upon which it would be possible to confront the theoretical effects calibrated by Rappaport (2005) for convergence between the small and the large economy varying with the degree of labor mobility.

With a focus on the evolution of wages, this article does not address employment effects as did, for example, Aterido, Hallward-Driemeier, and Pages (2011) when studying Mexico's Seguro Popular (SP) introduced in 2002. It is possible that income effects exist, thus implying lower labor force participation in the Mexican labor market. Recent research by Behrens and Sato (2011) investigates the effects of migration on the wage gap between skilled and unskilled labor. Our focus herein is on the regional aspects of migration, but we leave a more detailed investigation into different levels of human capital (available in an unbalanced manner for the maquiladora wage data) for future research.

This article examines Mexico's (real) wage movements across its 32 subnational entities for post-North American Free Trade Agreement years. Employing dynamic panel data methods, we obtain the following results. First, education (or labor productivity) has slightly higher wage effects in the Border-North region. Second, allowing for foreign capital and labor to respond to wages, returns to education have higher effects in South-Center Mexico, the region with (average) lower education levels. Third, convergence rates become lower with endogenous foreign capital and migration flows: wages move faster in the South-Center region than in Border-North. Overall, migration flows have greater effects on wages than foreign direct investment inflows. (JEL F15, F21. F22, F43, 047)


AB: Arellano-Bond

DGMM: Dynamic Generalized Method of the Moments

FDI: Foreign Direct Investment

FEM: Fixed-Effects Models

GDP: Gross Domestic Product

GMM: Generalized Method of the Moments

IMSS: Instituto Mexicano de Seguro Social

INEGI: Instituto Nacional de Estadistica y Geograffa

NAFTA: North American Free Trade Agreement

OLS: Ordinary Least Squares

SGMM: System Generalized Method of the Moments


Aterido, R., M. Hallward-Driemeier, and C. Pages. "Does Expanding Health Insurance Beyond Formal-Sector Workers Encourage Informality? Measuring the Impact of Mexico's Seguro Popular." IZA Discussion Paper No. # 5996, September, 2011.

Barro, R. J., N. G. Mankiw, and X. Sala-I-Martin. "Capital Mobility in Neoclassical Models of Growth." American Economic Review, 85(1), 1995. 103-15.

Behrens. K.. and Y. Sato. "Migration. Skill Formation, and the Wage Structure." Journal of Regional Science, 51(1), 2011.5-30.

Blundell, R.. and S. Bond. "Initial Conditions and Moment Restrictions in Dynamic Panel Data Methods." Journal of Econometrics. 87(1). 1998, 115-43.

Borjas, G. J. "Native Internal Migration and the Labor Market Impact of Immigration." Journal of Human Resources, 41(2), 2006, 221-58.

--. "The Labor Demand Curve Is Downward Sloping: Reexamining the Impact of Immigration on the Labor Market." Quarterly Journal of Economics, 118(4), 2003, 1335-74.

Bowsher, C. G. "On Testing Overidentifying Restrictions in Dynamic Panel Data Models." Economics Letters. 77(2). 2002. 211-20.

Cabral, R. and A. V. Mollick. "Mexico's Regional Output Convergence after NAFTA: A Dynamic Panel Data Analysis." Annals of Regional Science, 48(3). (2012), 877-95.

Cabral. R., A. V. Mollick, and J. R. Faria. "Capital and Labor Mobility and Their Impacts on Mexico's Regional Labor Markets." Journal of Development Studies, 46(9), 2010. 1523-42.

Card, D. "Immigrant Inflows, Native Outflows. and the Local Labor Market Impacts of Higher Immigration." Journal of Labor Economics. 19(1), 2001. 22-64.

Caselli. F., G. Esquivel. and F. Lefort. "Reopening the Convergence Debate: A New Look at Cross-Country Growth Empirics." Journal of Economic Growth, 1(3), 1996. 363-89.

Castellanos, S. G., R. Garcfa-Verd6, and D. S. Kaplan. "Nominal Wage Rigidities in Mexico: Evidence from Social Security Records." Journal of Development Economics, 75(2). 2004, 507-33.

Chiquiar, D. "Globalization. Regional Wage Differentials and the Stolper-Samuelson Theorem: Evidence from Mexico." Journal of International Economics, 74(1). 2008. 70-93.

--. "Why Mexico's Regional Income Convergence Broke Down." Journal of Development Economics, 77(1), 2005, 257-75.

Coates, D., and T. Gindling. "Is Hispanic Population Dispersion into Rural Counties Contributing to Local Economic Growth?" Contemporary Economic Policy, 31(4). 2013, 649-68.

DiCecio. R.. and C. S. Gascon. "Income Convergence in the United States: A Tale of Migration and Urbanization." Annals of Regional Science, 45(2), 2010. 365-77.

Esquivel, G. "Convergencia Regional en Mexico, 19401995." El lrimestre Economic[degrees], 66(4). 1999, 725-61.

Felbermayr, G.. S. Hiller, and D. Sala. "Does Immigration Boost Per Capita Income?" Economics Letters, 107(2), 2010, 177-79.

Freeman, R. "People Flows in Globalization." Journal of Economic Perspectives, 20(2). 2006. 145-70.

Hanson, G. H. "What Has Happened to Wages in Mexico since NAFTA'?" NBER Working Paper No. # 9563, 2003. Islam, N. "Growth Empirics: A Panel Data Approach." Quarterly Journal of Economics, 110(4), 1995, 1127-70.

Kaufman, M., P. Swagel, and S. Dunaway. "Regional Convergence and the Role of Federal Transfers in Canada." IMF Working Paper. May, 2003.

Longhi. S., P. Nijkamp, and J. Poot. "A Meta-analytic Assessment of the Effect of Immigration on Wages." Journal of Economic Surveys, 19(3), 2005,451-77.

Mayoral. F. M.. and C. Garcimartin. "The Impact of Population on the Reduction of Steady-State Disparities across Spanish Regions." Annals of Regional Science, 50(1), 2013,49-69.

Milanovic, B. "Economic Integration and Income Convergence: Not such a Strong Link?" Review of Economics and Statistics. 88(4). 2006, 659-70.

Mishra, P. "Emigration and Wages in Source Countries: Evidence from Mexico." Journal of Development Econotnic's. 82(1), 2007. 180-99.

Mollick. A. V., and R. Cabral. "Productivity Effects on Mexican Manufacturing Employment." North American Journal of Econotnics and Finance, 20(1), 2009, 66-81.

Ostbye, S., and 0. Westerlund. Is Migration Important for Regional Convergence? Comparative Evidence for Norwegian and Swedish Counties, 1980-2000." Regional Studies, 41(7), 2007, 901-15.

Ozgen, C.. P. Nijkamp, and J. Poot. "The Effect of Migration on Income Growth and Convergence: Meta-Analytic Evidence.- Papers in Regional Science, 89(3), 2010, 537-61.

Rappaport. J. "How Does Labor Mobility Affect Income Convergence?" Journal of Economic Dynatnics & Control, 29(3), 2005,567-81.

Robertson, R. "Exchange Rates and Relative Wages: Evidence from Mexico." North American Journal of Economics and Finance, 14(1). 2003, 25-48.

--. "Wage Shocks and North American Labor-Market Integration." American Economic Review, 90(4), 2000, 742-64.

Rodriguez-Oreggia, E. "Regional Disparities and Determinants of Growth in Mexico." Annals of Regional Science. 39(2), 2005, 207-20.

Roodman, D. -A Note on the Theme of Too Many Instruments." Oxford Bulletin of Economics and Statistics, 71(1), 2009, 135-58.

Schmidt, T. D., and P. S. Jensen. "Foreign Labor and Regional Labor Markets: Aggregate and Disaggregate Impact on Growth and Wages in Danish Regions." Annals of Regional Science, 50(3), 2013, 809-40.

Verhoogen, E. "Trade, Quality Upgrading and Wage Inequality in the Mexican Manufacturing Sector." Quarterly Journal of Economics, 123(2), 2008, 489-530.

Williamson, J. G. "Globalization, Convergence. and History." Journal of Economic History, 56(2). 1996. 277-306.

(1). While it would be interesting to explore capital and labor integration by contrasting pre-NAFTA to post-NAFTA subsamples. there are no annual data available--at the state level--for the pre-NAFTA years. Chiquiar (2005) and Rodriguez-Oreggia (2005) employ data with 5-year frequency (1970, 1975, 1980, and 1985) from a dataset that Instituto Nacional de Estadistica y Geografia (INEGI) used to report based on census data but discontinued in 1985. Apparently they both completed the subsequent years 1988, 1994. and 2000 with Census data. We are using a totally different dataset also compiled and reported by INEGI starting in 1993 with annual frequency. This dataset is not compatible with the data employed by Rodriguez-Oreggia (2005) and Chiquiar (2005) because it is not based on Census data.

(2.) The DF, due to its location as the capital of Mexico. is an important attractor of business in Latin America. For this reason, its share of FDI/GDP of 8.57% is much higher than other states. If one excludes DF from the total of 19 South-Center states, the average of the FDI/GDP goes down from 1.14% (displayed in Table 1) to 0.725%. Alternatively. excluding DF the average state in the South-Center region would be almost three times less than the average of states in the Border-North region: 2.12% shown in Table 1.

(3.) Note that since domestic migration rates are calculated as the difference between outflows and inflows of people over total state population, the denominator for each figure is different, making the average national rates and the sum of regions different from zero.

(4.) In Rappaport (2005. 574). the theoretical mechanism is as follows: "outflow of labor drives down the marginal product of capital. which in turn drives down the shadow value of installed capital, upon which investment depends." It is thus possible that investment falls with higher labor mobility and wages fall rather than increase.

(5.) A previous version of this article examined output convergence as well. However, as an anonymous referee correctly pointed out, there are important problems with a Solow model approach to output convergence based on population growth rates and savings rates. The reason is the lack of availability of domestic capital at the state level for the Mexican case. Since savings are similar across regions, other variables are needed for regional convergence. We allow for population growth rates, education, the proportion of rural population, the number of telephone lines, and the share of manufacturing GDP on total state GDP. Yet we do not have a reliable measure of domestic capital stock at the state level that complements the amount of foreign capital coming in through FDI toward states.

(6.) Similarly, Schmidt and Jensen (2013) find no effect of foreign labor inflow on wages of Denmark hut there are negative effects (1-statistic of -1.77) when an instrumental variable for social networks is employed and when no controls are included.

(7.) According to Bowsher (2002)'s Monte Carlo study, the power of the Sargan test is maximized employing three lags.

(8.) Coates and Gindling (2013) find, for example, for their panel of 3,101 U.S. counties that the 2SLS-IV estimates are their preferred estimators due to the problems found in specification tests for DGMM and SGMM in their income growth regressions.

(9.) In a complementary table available upon request, we perform FEM allowing for lagged wages as regressors. In column 1, for example, the wage persistence is estimated at 0.677 with time and fixed effects for Border-North and at 0.593 for South-Center. The education coefficient is estimated at 0.605 for Border-North and at 0.711 for the South-Center region, while all other coefficients are not statistically significant for Border-North and migration has negative effects in the South-Center region. The R2 statistics become closer to I. Since FEM are not suitable for treating lagged dependent variables, we prefer to report in Table 4 the estimations without lagged wages in the first-pass approach.

(10.) As an alternative to the real exchange rate, we also construct a measure of total trade (exports plus imports) divided by GDP to capture the dynamism of external trade in Mexico. The specification with total trade, however, has problems with the wage equation when AB(2) statistics are always rejected At standard significance levels.

(11.) We measured the speed of adjustment to the steady-state and calculated it as--In ([gamma]) in either Equation (2) or (3).

(12.) One possible interpretation for these negative coefficients is as in the two-country model by Rappaport (2005), in which firm investment falls with higher labor mobility and wages fall rather than increase. An alternative view is that the gaps in the data for maquiladora wages remove much of the credibility in these estimates with maquiladora-based wages.

* The authors wish to thank, without implicating, Joao R. Faria. David J. Molina, and two anonymous referees for their helpful comments on previous versions of this article, as well as seminar participants of the Annual Meeting of the Allied Social Sciences Associations (ASSA) in Denver, CO, January 2011.

Mollick: Professor of Economics, Department of Economics and Finance. University of Texas--Pan American, 1201W. University Dr., Edinburg, TX 78539-2999. Phone +1956-665-2494, Fax +1-956-665-5020, E-mail

Cabral: Associate Professor. Escuela de Graduados en AdministraciOn Publica y Politica Ptiblica, Tecnologico de Monterrey, Campus Monterrey, Ave. Rufino Tamayo, Garza Garcia, NL, CP. 66269, Me'xico. Phone +52-818625-8347, Fax +52-81-8625-8385. E-mail
COPYRIGHT 2015 Western Economic Association International
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2015 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Mollick, Andre Varella; Cabral, Rene
Publication:Contemporary Economic Policy
Geographic Code:1MEX
Date:Jan 1, 2015
Previous Article:Comparing federal and private-sector wages without logs.
Next Article:Fiscal multipliers during the global financial crisis: fiscal and monetary interaction matters.

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters