Wage shocks and North American labor-market integration

Transcription

1 Texas A&M University From the SelectedWorks of Raymond Robertson September, 2000 Wage shocks and North American labor-market integration Raymond Robertson, Macalester College Available at:

2 Wage Shocks and North American Labor-Market Integration By RAYMOND ROBERTSON* This study uses household-level data from the United States and Mexico to examine labor-market integration. I consider how the effects of shocks and rates of convergence to an equilibrium differential are affected by borders, geography, and demographics. Ifind that even though a large wage differential exists between them, the labor markets of the United States and Mexico are closely integrated. Mexico's border region is more integrated with the United States than is the Mexican interior. Evidence of integration precedes the North American Free Trade Agreement (NAFTA) and may be largely the result of migration. (JEL F15, F20, J61) Negotiations for the Free Trade Agreement of the Americas (FTAA), which seeks to expand the North American Free Trade Agreement (NAFTA) to the rest of Latin America, opened without U.S. approval for fast-track authority. Much of the successful opposition to fast track was based on fears of labor-market integration. Recent increases in migration of less skilled workers and rising wage inequality in the United States fed opposition to measures that would integrate U.S. labor markets with the rest of Latin America. Evidence in this paper suggests that U.S. and Mexican labor-market integration was a reality before NAFTA. Given the attention in the popular press, it is surprising that more attention has not been given to the role of borders, demographics, and geography in international labor-market integration. The role of borders and geography in other markets has received recent attention. John McCallum (1995) examines the pattern of goods flows in the United States and Canada and finds that Canadian provinces trade more between each other than they do with the United States, even when U.S. destinations are closer. Charles Engel and John H. * Department of Economics, Macalester College, 1600 Grand Avenue, St. Paul, MN The author thanks Theresa Greaney, Daniel Hamermesh, Gordon Hanson, Douglas Holtz-Eakin, Chihwa Kao, J. David Richardson, Daniel Slesnick, Sarah West, and two anonymous referees for helpful comments and discussion. I also thank the Population Resource Center at the University of Texas at Austin and the staff of INEGI in Aguascalientes, Mexico, for their assistance with the data. Any remaining errors are mine. 742 Rogers (1996) find that borders matter for goods prices by using consumer-price indices from the United States and Canada. They find significant price differences across borders that are not explained by distance. For product markets, integration increases as price differences, net of transport costs, diminish because forces within segmented markets independently determine prices. Since labor markets in different countries are subject to considerable measures designed to maintain observed differentials, it may be misleading to rely on price convergence as a criterion of labormarket integration. If factors can adjust to restore the differential when actual wages deviate, the markets may still be considered integrated. Thus, Engel and Roger's criterion that borders matter if prices are different may not necessarily suggest labor-market segmentation. To study labor-market integration, I consider two alternative criteria: (i) how changes in wages in one country affect wages in neighboring countries and (ii) how quickly wages return to a given differential. Since the U.S. economy dwarfs its neighbors,1 one would expect that the most identifiable effects of wage shocks would be those going from the United States into its smaller neighbors. I consider the United States and Mexico because their labor markets are among the most different of any pair that has attempted free trade. Furthermore, questions of wage con- ' In 1997, the U.S. GDP reached US$7,100 billion; Canada, US$574 billion; Mexico, US$305 billion.

3 VOL. 90 NO. 4 ROBERTSON: WAGE SHOCKS AND LABOR-MARKET INTEGRATION 743 vergence, migration, and capital flows between Mexico and the United States continue to arise in policy discussions. Both Mexico's relative size and recent trade liberalization make the U.S.-Mexican case excellent for studying labor-market integration. Changes in wages may be transmitted across borders through capital movements, trade, and migration.2 Recent unilateral and bilateral moves toward integration have brought attention to the U.S.-Mexican relationship and especially to the U.S.-Mexican border region [Gordon H. Hanson (1996) and Hanson (1998)]. Mexico liberalized its foreign investment laws in Capital flows, mainly from the United States, dramatically increased (Robert Feenstra and Hanson, 1997). Foreign trade between Mexico and the United States increased greatly when Mexico joined the GATT in Immigration, measured by both legal admissions and border apprehensions, has also been increasing since 1987 (Hanson and Antonio Spilimbergo, 1999). Each of these phenomena suggests that labor markets were integrating before NAFTA. Correlated shocks to wages are evidence that factors that integrate markets affect labor-market outcomes across borders.3 To provide structure for the empirical work, I develop a simple three-region model characterized by migration and transport costs. The three-region framework provides the structure necessary to test various hypotheses about the geographic aspects of integration such as whether distance, borders, and demographic characteristics affect the transmission of shocks across countries. I apply aggregate labor-supply and -demand equations from the model to a time series of cross-section household-level data sets in a manner described by Angus Deaton (1985). The estimated reduced-form equations capture both the effects of U.S. wage changes on Mexican wages and the rate at 2 See John M. Abowd and Richard B. Freeman (1991) for excellent discussions of each of these topics. 3 Most economists have focused on the transmission of shocks through capital markets. Until recently, labor markets have received much less attention. To my knowledge, no study explores the transmission of labor-market shocks across countries, nor has addressed the effect that changes in U.S. wages have on countries bordering the United States. which wages converge to the equilibrium international differential. The empirical work suggests that even before the North American Free Trade Agreement, U.S. and Mexican labor markets were integrated. Specifically, three main conclusions emerge. First, the effects of U.S. wage shocks are stronger in Mexico's border cities than in the interior of Mexico. Second, following a wage shock, wages in Mexican border cities converge to U.S. wages more quickly than wages in the interior of Mexico. Third, 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. Although circumstantial, differences across Mexican border cities seem to suggest that of forces that could integrate labor markets--goods flows, capital movements, and migrationmigration may be the dominant mechanism. The Mexican border is more integrated with the United States than is the Mexican interior. These results are robust across differences in gender, age, and education. The results also withstand differences in the definition of the "border." Since almost 80 percent of all Mexican migrants to the United States choose Texas or California as their destination, I also examine the effects of changes in wages in these states on Mexican wages. As expected, the effects are similar, although somewhat smaller in magnitude, to those involving the entire United States. Furthermore, I find some evidence that Tijuana, a main embarkation point for migrants, may be more integrated with the United States than with Mexico's interior, and that other border cities that are less integrated with the United States are more integrated with the Mexican interior. The time period covered in this study-1987 to 1997-encompasses both the implementation of the North American Free Trade Agreement and periods of large exchange rate movements. Controlling for periods of large exchange-rate movements, results suggest that the rate of convergence between U.S. and Mexican wages slowed during the NAFTA period. This effect was larger for the interior than for the border. During the NAFTA debate, propo-

4 744 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 2000 nents argued that NAFTA would help create opportunities for workers in Mexico and thus relieve migration pressures on the United States. To the extent that migration integrates labor markets, increasing opportunities for Mexicans within Mexico may help explain why the rate of wage convergence seems to slow after NAFTA is implemented. However, sensitivity to exchange-rate movements suggests that it may be too early to fully evaluate the effect that NAFTA has had on labor-market integration. Nonetheless, although the border matters for the absolute wage differential, U.S. and Mexican labor markets are integrated and there is a significant geographic component to this integration. The paper proceeds as follows. Section I develops the model used to provide structure for the empirical work and describes various hypotheses that can be tested with equations derived from the model. Section II provides information about migration and demographic characteristics and includes tests of various hypotheses about labor-market integration. I also evaluate the robustness of the results to demographic characteristics, exchange-rate movements, and the NAFTA period in Section II. Section III offers conclusions, caveats, and suggestions for future research. I. Theoretical Considerations International capital flows, immigration, and trade are each sufficient to integrate labor markets. However, none of these reacts instantly to changes in market conditions. Capital, in the form of new plants and equipment, does not move instantly. Immigrants incur transportation costs that may vary by education and gender. Transportin goods is also expensive. Transportation costs and delayed adjustment suggest that the effect of foreign wage shocks would vary by geographic region and demographic group. These characteristics and the dynamic nature of the labor market can be captured most simply with aggregate labor demand and supply equations. In this section, I develop a simple threeregion model of international labor markets. Equations derived from the three-region model provide the foundation for the empirical work in Section II. A. The Border Region To begin the analysis, consider two contiguous regions separated by a border (the third region, the interior, will be added shortly). For simplicity, label the regions "United States" and "Mexico." Next assume that, from the firm's point of view, labor in the United States and labor in Mexico are p substitutes (Daniel S. Hamermesh, 1993). That is, firms that face an exogenous wage increase in the United States will increase their demand for Mexican workers.4 As long as capital flows are not instantaneous, the lagged U.S. wage will affect labor demand in Mexico.5 Labor demand in the Mexican region is affected by an exogenous shifter (80) and the lagged U.S. wage. The goal is to capture the possibility of gradual adjustment by using a function of current and lagged Mexican wages to express movements along the labordemand curve. A general form capturing these assumptions is (1) L] =o0 + 6Iwjts- 62[w -2bwjtb11? a31 + in which L is labor, w"' is the log(u.s. wage), and wb is the log(mexican wage), and the other symbols are parameters of the demand function. The subscript j represents an individual labor 6"group" (1 J) and the subscript t(1..t) represents the time period. The parameter y captures the responsiveness of demand to lagged wages. The final term represents a group-specific (and time-invariant) effect on labor demand. The opportunity to migrate to the United States partially characterizes the Mexican laborsupply decision. If the U.S. wage rises, workers leave Mexico and enter the United States. Again allowing for dynamic adjustment, I model movements along the labor-supply curve as a function of current and lagged Mexican wages. I assume that labor can migrate instantly between regions in response to innovations in the 4 Although a reasonable assumption for this study, there is evidence that U.S. and Mexican workers may be complements. For example, see George J. Borjas (1983). 5 Alternatively, one could assume that firms would base their location decision on the expected U.S. wage, whose best predictor is the lagged U.S. wage. The resulting functional form would be the same.

5 VOL. 90 NO. 4 ROBERTSON: WAGE SHOCKS AND LABOR-MARKET INTEGRATION 745 U.S. wage because labor is more mobile than factors that shift demand (such as capital).6 A general form capturing these assumptions is + (2 + 62) wt (2) Lj1= 0-( Iwjt + 51 ( )wus + it --1] + U3i- Analogous to y, the parameter (p captures the responsiveness of supply to lagged wages. The subscripts are as previously defined and the final term on the right-hand side represents a groupspecific (and time-invariant) effect on labor supply. The coefficients a, and 61 embody the migration costs and transport costs to suppliers and demanders, respectively. These costs are exogenous but are high enough to prevent U.S. and Mexican wages from equalizing.7 In the presence of such costs, an equilibrium differential will separate regional wages. When shocked, wages may move away from that differential temporarily but will return to it eventually because factors are imperfectly mobile across regions. Equilibrium can be represented as (3) O -I Wjt + -2[ pw ti1i + =60 + 6l Wjts1 I - 62b bt 1 + OT3j that is, labor supply equals labor demand. By solving equation (3) for wb j~t the current Mexican wage can be expressed as a function of the lagged Mexican wage, the current U.S. wage, and the lagged U.S. wage:? (4) Wt = - + 3i - (ut2? 62) (ut2 + 2 (62Y + 0>2P) b (0f2 + 2) Wit I 6This is true if firms require time to build, which I assume. Migrants, on the other hand, can make their decisions and react more quickly (discussions with border patrol agents suggest this is true). 7 Migration costs, such as transportation, bureaucratic requirements, and avoiding detection could account for the failure of complete factor price equalization. These are the types of policies designed to prevent wages from equalizing. 3j) 63j Rewiite equation (4) more simply as (5) t + aoj + a Wj) WJ---ao jaoao1a1> Iwbt- I + el wts 4- e2wjut _l. Following David F. Hendry and Neil R. Ericsson (1991), the long-run homogeneity between wb and wu5 implies that a,, el, and e2 sum to 18 JUse this restriction to express equation (5) as (6) ()AWbt Awjao+1A = + alawjus t + a2(wb - w )jt + Pjt When differenced, the time-invariant groupspecific terms drop out. Equation (6) can be used to test two key hypotheses about integrated labor markets. First, if wages in the United States and in Mexico respond to the same shocks, a If a shock that increases wages in the United States also causes Mexican wages to increase, a, > 0.9 Second, the third term on the right-hand side represents the adjustment to the long-run equilibrium. If Mexican and U.S. wages converge to the equilibrium differential, a2 K 0. George R. Boyer and Timothy J. Hatton (1994) suggest that the speed of convergence can be estimated 8 The sum (standard error) of a,, el, and e2 for Mexico City is (0.015), Tijuana (0.018), Ciudad Juarez (0.019), Matamoros (0.020), and Nuevo Laredo (0.019). Thus, with the data used in this paper, a1, e1, and e2 generally sum close to 1. 9 This result would seem to follow from a model with two closed countries subject to correlated shocks. However, there is no reason to expect a gradual convergence to a long-run differential if the economies were closed. Also, there would be no reason for the shock to have a larger effect on the border than on the interior if the results were driven completely by correlated shocks and not market linkages.

6 746 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 2000 as (1 - a2)1a2, which is a function of migration and transport costs embodied in a, and 81. This measure will serve as a metric of labor-market integration. The intuition of these hypotheses follows a simple marginal product of labor analysis. In equilibrium, the wage in Mexico will be equal to the wage in the United States, less the cost of crossing the border. If the marginal product of labor differs from the equilibrium differential (as the result of a shock, for example) factors gradually move until the differential is reestablished. The impact of shocks and speed at which the differential is restored (the a) may vary by demographic group. I incorporate this possibility into the empirical work. I address how these relationships may vary across regions within the same country in the next section. B. The Interior The purpose of the third region is to analyze geographic aspects of labor-market integration. In the three-region model, I distinguish between the border and interior of Mexico and allow for migration within Mexico. Robert H. Topel (1986) finds that the geographic aspect of U.S. labor markets is important because some demographic groups are less mobile than others. Geography plays a role in Mexican labor markets as well. Carol A. Zabin and Salle Hughes (1995) find that more than half of the Mexican migrants in their study first migrated to the border from the interior before entering the United States. While in the border cities, migrants acquire skills and information that prepare them for the trip to the "other side." In addition, like labor, foreign investment has the option to locate in the interior as well as the border, and investors must weigh differences in wage costs against transportation costs. Demand and supply equations capturing the movement of workers from the interior to the border, and from the border to the United States, are (7) LJit =o + 4) wstsi - -2[Wit-Wit-1] + 03tj (8) Li = So- S1Wb + S2[W t- Wjt 1] + S3tj where r1 and ; are analogous to y and (p in equations (1) and (2). The lagged U.S. wage appears in the demand equation because U.S. firms make the decision to locate in the border or the interior based on changes in the U.S. wage and the time that capital takes to become productive. The current border wage appears in the supply equation because workers in the interior first move to, and spend time in, the border region. Equilibrium in the interior can be expressed as (9) (S2 + P2)Wit= (00 - S0) + (03j - S3j) + I W>t-1 + (402 r) + S + 21 wtt To simplify equation (9), define (10) ((OU2 + 82) - (82Y + U2cp)A) = ( - ka) (A is the lag operator) so that 1 and k are constants made up of coefficients. Use the equilibrium condition for the border region [equation (3)] to express the wages in the interior as a function of the lagged interior wage and U.S. wages: (11) wit= (S2 + P2) {[MO - S0) + (43j - + si(60 - o-)(l - k)] + sia,lwsjt/ + (41 + s11k -sllk -si6ika)w`s_i + ( S20Wit-lb S3j) As in the two-region model, a similar relationship between wages in the United States and wages in the interior (w1) can be expressed as (12) Aw,t= 30o + f3aw1aw and + 02 (wi - wus)j-i + Et,

7 VOL. 90 NO. 4 ROBERTSON: WAGE SHOCKS AND LABOR-MARKET INTEGRATION 747 where 0, measures the effect of the U.S. shock on the interior and /2 measures the rate of convergence. The equations that provide the basis for the empirical work are (6) and (12). These equations are conductive to testing various hypothesis tests about the geographic aspects of labor-market integration. These hypotheses can best be described with an example. Suppose that wages in the United States increase. Migrants leave the border and enter the United States, which raises wages in the border region. Rising wages in the border attract migrants to the border from the interior. Migration from the interior raises wages in the interior and mitigates the effect of the U.S. shock on the border. The U.S. shock first affects the border and then the interior. As a result of delays in adjustment, the border converges to the equilibrium differential more rapidly than the wage in the interior. In the presence of transportation costs that increase with distance, the border region acts as a static and dynamic "buffer zone" between the United States and the Mexican interior. Thus, several testable hypotheses join those from the previous section. The first is that wages in the United States and in the Mexican interior region will respond to the same shocks to some degree (I31# 0), but that responsiveness is less in the interior than that between the United States and the border region (I31 < a1). The second hypothesis is that the interior wage and the wage in the United States will converge to the equilibrium differential at a slower rate than the rate of convergence between wages in the United States and the border region. A test of this hypothesis is a test that 1321 < a2!. All of these hypotheses have clear alternatives: that the border that maintains the observed wage differential between the United States and Mexico effectively segments the labor market and/or there is no geographic component to labor market integration. II. Empirical Analysis In this section I employ two groups of data. In subsection A, I draw on data from both the Mexican and U.S population censuses, the U.S. Immigration and Naturalization Service (INS), and from the Mexican National Institute for Statistics, Geography, and Information (INEGI) to describe demographic characteristics of the U.S.-Mexican border population. Specifically, I use these data to describe patterns of immigration and capital flows that may affect the transmission of shocks between Mexico and the United States. In subsection B, I describe the data used in the regression analysis. These data come from the quarterly Mexican National Urban Employment Surveys (ENEU) and are available from the first quarter of 1987 to the last quarter of The surveys are conducted in municipalities throughout Mexico. For the Mexican border region, I focus on four cities: Tijuana, Ciudad Juarez, Matamoros, and Nuevo Laredo. For the Mexican interior, I focus on the two largest Mexican urban areas: Mexico City (including municipalities surrounding Mexico Cityl0) and Guadalajara (including municipalities surrounding Guadalajara"1). The cities are chosen because they are available for the entire sample period ( ). The availability of the data determined the sample period. A. Distance, Demographics, and Migration In this section I employ data from six major Mexican metropolitan areas [shown in the map of Mexico (Figure 1)]. Mexico City and municipalities representing the State of Mexico comprise about 17 percent of the 1990 Mexican population. Guadalajara arid surrounding areas in the state of Jalisco represent another 4 percent. Together, these areas represent the "interior." Tijuana, Ciudad Juarez, Nuevo Laredo, and Matamoros represent the "border." Differences across border cities may provide insights into the mechanisms that integrate labor markets because these four cities are quite different. Ciudad Juarez and Tijuana are either adjacent to (El Paso) or within 15 miles of (San Diego) U.S. 10 I refer to municipalities in the area surrounding Mexico City as "Mexico State." They include Atizapan de Zaragoza, Coacalco, Cuautitlan, Chimalhuacan, Ecatepec, Huixquilucan, Naucalpan, Nezahualcoyotl, La Paz, Tlalnepantla, Tultitlan, Cuautitlan, and Izcalli. " I refer to Guadalajar and the municipalities surrounding Guadalajara with their common state name "Jalisco." Municipalities surrounding Guadalajara represented here are Tlaquepaque, Tonala, and Zapopan.

8 748 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 2000 San Diego #+w- Tijuana t\k,,,,,,,,,,. n gel Paso > w - >.Ciudad g W3r Juarez SanAntonio 2 Nuevo Laredo \ w}.,... \ C-- " Matamoros JAvLISCO, r MEXICO CITY ( MEXICO E; - / K K?- FIGURE 1. MEXicO cities with more than 500,000 people.12 The nearest major city to Matamoros and Nuevo Laredo is San Antonio (1990 population of 935,933), which is 145 miles away (center to center) from Laredo and 251 miles from Matamoros. Thus, if proximity to cities matters for labor-market shocks, we would expect wages in Ciudad Juarez and Tijuana to converge more quickly to the U.S.-border wage differential than wages in Matamoros and Nuevo Laredo. Table 1 contains demographic and economic information about each city. Tijuana and Ciudad Juarez are over twice as large as Matamoros and nearly three times larger than Nuevo Laredo. The border cities have a slightly larger proportion of males than that of the interior. Male labor-force participation is higher in Ciudad Juarez and Tijuana than that in the other two cities, and higher in all border cities than that in Mexico City. Manufacturing has received special attention in Mexico because of the maquiladora industry (e.g., Feenstra and Hanson, 1997). The maquiladora industry, also known as the in-bond industry, is a product of a program that allows companies in Mexico to import inputs duty free and export finished products, while paying duties only on the value added in Mexico. Females comprise a large share of maquiladora employment. Table 1 shows that both males and females are more involved in manufacturing in the border cities, but this distinction is much higher for females. The rate of female participation in manufacturing in the border cities is more than twice that of females in Mexico City. Although only 30 percent of females are economically active, more than half of all production workers in maquiladoras are female. Tijuana and Ciudad Juarez have many more maquiladora plants than the other two border cities, and Ciudad Juarez has the highest maquiladora employment. Changes in exchange rates affect the flow of capital, goods, and workers across borders. Between 1987 and 1997 Mexico experienced two periods of sharp peso devaluations: 1987 and 1995,13 Movements of the peso are depicted in Figure 2. Figure 2 shows that for the majority of the sample period, the peso was 12 In 1990, the population of San Diego city was 1,110,549 and the population of El Paso city was 515, The 1995 peso devaluation began in the last weeks of December 1994.

9 VOL. 90 NO. 4 ROBERTSON: WAGE SHOCKS AND LABOR-MARKET INTEGRATION 749 TABLE 1-DEMOGRAPHIC DATA A Mexican population census Interior Border State of Ciudad Nuevo Mexico City Mexicoa Jaliscob Tijuana Juarez Laredo Matamoros Population 8,235,744 5,561,979 1,650, , , , ,293 Population older than 12 years 6,217,435 3,956,663 1,185, , , , ,964 Male of population older than 12 years (percent) All males who are economically active (percent) All females who are economically active (percent) Economically active males who are in manufacturing (percent) Economically active females who are in manufacturing (percent) B. Maquiladora industry datac Ciudad Nuevo Tijuana Juarez Laredo Matamoros Maquiladora plants (1990) Maquiladora employment 59, ,231 16,036 38,360 Production workers who are male (percent) a Municipalities included in the State of Mexico are found in the area surrounding Mexico City and include Atizapan de Zaragoza, Coacalco, Cuautitlan, Chimalhuacan, Ecatepec, Huixquilucan, Naucalpan, Nezahualcoyotl, La Paz, Tlalnepantla, Tultitlan, Cuautitlan, and Izcalli. No other municipalities are available in the ENEU for the State of Mexico for the entire sample period. b Municipalities included in the state of Jalisco include Guadalajara (the second-largest city in Mexico after Mexico City), as well as Tlaquepaque, Tonala, and Zapopan. No other municipalities are available in the ENEU for the State of Jalisco for the entire sample period. cinstituto Nacional de Estadistica, Geografia e Informatica, Aguascalientes, In this publication there is no information on maquiladoras in Mexico City because, relative to the border, the number of maquiladoras in the interior areas is small. basically stable but experienced a controlled depreciation. Bruce A. Blonigen (1997) finds that when the U.S. dollar value falls, foreign investment in the United States increases. However, large shocks may have had the opposite effect in Mexico (Francisco Gil-Diaz, 1998). For the purposes of this study, maquiladora investment is the most relevant form of foreign investment. Figure 3 shows the monthly evolution of maquila plants and employment in Tijuana and Ciudad Juarez. Consistent with Blonigen's findings for the United States, peso devaluations precede increases in maquila plants and employment. In the sections that follow, I explore the robustness of this paper's results to the sharp changes in the Mexican exchange rate and to the degree that the Mexican exchange rate was controlled. Although greater maquiladora employment suggests maquiladora activity is a dominant characteristic of Ciudad Juarez, migration is a dominant characteristic of Tijuana. Table 2 presents information on migration from the 1990 Mexican and U.S. population census and from INS apprehension statistics. The 1990 Mexican population census shows that over 60 percent of the population of every

10 750 THE AMERICAN ECONOMIC REVIEW SEPTEMBER a Year 19** FIGURE 2. NOMINAL EXCHANGE RATE Note: The years on the horizontal axis mark January of the reference year. area, except Tijuana and Mexico State,14 were born in the same city. In Tijuana this figure is just less than 42 percent. The percentage of people living outside the state in 1985 is nearly twice as high in Tijuana as that in the next highest border city, and over five times that percentage in Mexico City. A similar story emerges for the percentage of people living out of the country for both males and females. U.S. INS apprehensions, both in absolute number and relative to city size, are nearly double in the sector across from Tijuana as those in the sector across from the next highest site, Ciudad Juarez. The 1990 U.S. population census statistics further corroborate Tijuana's importance as a migration center. Over 57 percent of Mexican-born respondents to the U.S. census are found in Californiamore than double the 22 percent living in Texas. The picture is even starker for the Mexican-born who entered between 1985 and 1990 (roughly the first half of the time period covered in this analysis). Thus, if migration is a domi- 14 Note that Mexico State has migration statistics simila to the border cities. As a center for economic activity, Mexico City still draws many migrants from throughout the country. In contrast to the border region, hardly any of the migrants to Mexico State were outside the country in nant mechanism for transmitting labor-market shocks and restoring the equilibrium wage differential, we expect the results of the model to be stronger in Tijuana than in the other border cities. Table 2 also shows that Mexican-born migrants are, on average, less educated than U.S. citizens. In 1990, over 78 percent of all U.S. workers had graduated from high school. In contrast, less than 31 percent of all Mexican-born U.S. residents have a highschool education. Migrants in California and Texas generally have less education than those born in Mexico but living in other states. Mexican-born respondents to the U.S. census are more often male. Whereas females are also very mobile, these data seem to suggest that less-educated males are the more mobile and thus may have smaller migration costs. These findings are similar to those documented by Stephen J. Trejo (1997) in his analysis of the economic performance of Mexicans in the United States. Overall, Tables 1 and 2 suggest Tijuana and Ciudad Juarez differ from the other two border cities in several ways. They both have more migration, more maquiladoras, and are closer to major U.S. cities. Migration is more pronounced in Tijuana and Ciudad Juarez has a

11 VOL. 90 NO. 4 ROBERTSON: WAGE SHOCKS AND LABOR-MARKET INTEGRATION Tijuana Maquila Establishments C. Juarez Maquila Establishments Year 19** FIGuRE 3A. MAQUILADORA ESTABLISHMENTS IN TIJUANA AND CIUDAD JUARE Z Source: INEGI. + Tijuana Maquila Employment C. Juarez Maquila Employment v Year 19** FIGURE 3B. MAQUILADORA EMPLOYMENT IN TIJUANA AND CIUDAD JUAREZ Source: INEGI. Employment is in number of workers. larger share of maquiladora employment. These differences are used in interpreting the results in subsection C below. B. Survey Data If worker heterogeneity matters for labor markets, as Topel (1986) suggests, data that can distinguish wages by geographic region and also by demographic group would be crucial for a successful examination of how shocks are transmitted across regions. The data should also be comparable across both countries in these dimensions. The survey data used in this study are well suited for both demands. The U.S. data are from the Current Population Surveys (CPS).

12 752 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 2000 TABLE 2-MIGRATION STATISTICS A Mexican population census Interior Border Mexico State of Ciudad Nuevo City Mexico Jalisco Tijuana Juarez Laredo Matamoros Population born in city (percent) Males out of state in 1985 (percent) Females out of state in 1985 (percent) Males out of country in 1985 (percent) Females out of country in 1985 (percent) INS apprehensions' n.a. n.a. n.a. 485, ,212 86,169 98,737 B U.S. population census Rest United 1 Percent PUMS States California Texas Population born in Mexico (percent) High-school grad U.S. bomb High-school grad Mexican bomb U.S.-born female (percent) Mexican-born female (percent) Mexican-born living in (percent) Mexican-born, arrived , living in- (percent) asource: See Hanson et al. (1999). Apprehensions are cumulative and probably include multiple attempts by some individuals. b These figures reflect the percentage of the population that has at least a high-school education. The National Urban Employment Surveys (ENEU) are the source of the Mexican data. The characteristics of the CPS are well known. The Mexican data are similar to the CPS and are used by the Mexican government to calculate unemployment rates in urban areas. Whereas the CPS are conducted monthly, the ENEU are conducted quarterly. To make the data sets comparable, I aggregate the CPS data into quarters. Mexican wages are initially expressed in dollars using the dollar/peso exchange rate from each quarter. When perfectly flexible, the exchange rate adjusts for the differences in inflation between the two countries. I explore the robustness of the results that follow to the dramatic movements in the peso during 1987 and 1995 and to the degree to which the nominal exchange rate was controlled. The Mexican data contain monthly earnings, whereas the CPS report weekly earnings. For comparability, I calculate monthly U.S. earnings as the usual weekly wage times 4.2 (weeks per month). I excluded workers who worked less than 20 hours in the reference week or had no hours recorded in the survey. Males and females younger than 16 and older than 70 years were also excluded from the sample. The data span 1987:1-1997:4, a period of increased capital, labor, and goods flows between Mexico and the United States. Table 3 contains summary statistics for the U.S. and the Mexican data. Five variables are presented. Strong regional differences are apparent. The average number of years of education in the United States is just over 13, and the average age of workers is slightly above 38 years. The fraction of males and females is nearly equal. On the other side of the border, wages are lower. U.S. workers earn over six times the wage of their Mexican counterparts. Mexican workers are almost 6 years younger

13 VOL. 90 NO. 4 ROBERTSON: WAGE SHOCKS AND LABOR-MARKET INTEGRATION 753 TABLE 3-SUMMARY STATISTICS FOR U.S. AND MEXICAN DATA, FIRST QUARTER 1990 Mean Mean monthly Coefficient of Percent Mean age education earnings variation of male (years) level (years) (dollars) log(earnings) United States Mexican border Tijuana Ciudad Juarez Matamoros Nuevo Laredo Mexican interior Mexico City Mexico State Jalisco Source: CPS, ENEU. on average and they have about five fewer years of education. The levels of education for Mexican workers in the survey data are very similar to those of Mexican-born migrants in the U.S. census, albeit slightly lower in the United States. That is, Mexican migrants in the United States have slightly less education than Mexican nationals in Mexico. This finding is consistent with other studies of U.S. immigrants (Borjas 1994; Trejo 1997). The differences between the United States and Mexico are not surprising. What is less well known is the pattern of differences between workers in the border and workers in the interior of Mexico. Table 3 shows that workers on the border are less educated, younger, and have higher wages. Workers in Mexico City are about three years older and average one additional year of schooling. One possible explanation for the wage differential on the border is the higher fraction of males in the border sample, but the difference in number of males cannot account for the entire wage differential.'5 The presence of transportation costs and differences in regional labor demand (as suggested by the model) could also explain the difference. It is also worth mentioning that wage inequality [as measured by the coefficient of variation of log(wages)] is higher in Mexico than in the United States. 15 Regressions that control for gender suggest that wages are 41 percent higher in Tijuana, 21 percent higher in El Paso and Matamoros, and 2 percent higher in Nuevo Laredo. Industry controls were not included. The region with the highest degree of wage dispersion is the interior, but the border region is still much less equitable than the United States. Ciudad Juarez and Tijuana, both centers of maquiladora activity, have a more unequal distribution of wages than the other two border cities. Since we also expect heterogeneity across groups of workers, I follow Deaton (1985) and Martin Browning et al. (1985) and combine a time series of cross-sectional data sets into a pseudopanel. The data are quarterly average wages for 80 different age-education categories (eight education groups and ten age groups). The age groups are advanced through time and span five years each. The education groups are: no education, education up to 4th grade, 5th or 6th grade, 7th or 8th grade, 9th grade, 10th grade, 11th or 12th grade, and more than high school. Average wages for each age-education cell are arranged into a panel data set that I use in the estimation. Using this technique, we can follow the movements of individual cohorts across time to see how they are affected by wage shocks. Figure 4 illustrates the path of wages averaged over all groups in the United States, the Mexican border region (averaged across all four cities), and the Mexican interior (averaged across all three regions). The wages used in estimation for the interior are the average across Mexico City, Mexico State, and Jalisco. Average wages for the border cities are calculated individually for each border city.

14 754 THE AMERICAN ECONOMIC REVIEW SEPTEMBER Mexico Interior United States o Mexico Border I I I II I I I IT T Year 19** FIGURE 4. LOG QUARTERLY WAGES (DOLLARS) Notes: Wages are weighted averages of all age-education groups using cell sizes as weights. The border wage series is the weighted average of all four border cities. The interior wage series is the weighted average of all three interior regions. Mexican wages are converted to dollars using the peso/dollar exchange rate. C. Estimation As noted earlier, the base estimation equations are (6) and (12). To ensure that these equations are identified, I invoke the assumption that Mexico is too small to affect aggregate wages in the United States.16 To test the hypotheses, I follow a dummy variable approach similar to tests for structural change. I pool the data from the interior and border of Mexico. I then estimate one equation that combines (6) and (12) by including a dummy variable for each border city and interact the dummy variables with the other regressors. If f31 < a1 and 1021 < ca2i then the interaction terms should be significantly different from zero (and of the expected sign). The possibility that a and,b may vary by education group is examined by divid- 16 This assumption seems supported by the recent literature on wage inequality, international trade, and immigration [e.g., Abowd and Freeman (1991)]. Although there is some controversy over methodology, most studies find a small role for immigration (Borjas, 1994). David Card (1990), in particular, finds that a large and sudden influx of immigrants into Miami did not affect wages. ing the sample into different groups. The estimation equation is (13) Awg,= zod + [z1d]awj,s + [z2(wg -w "%,-I]D + -jt, where j indexes the age-education cohort and g indexes the region; D is a vector with a 1 followed by dummy variables representing each border city; and the z terms represent vectors of coefficients. The coefficients on the interaction terms represent the differential effect of shocks to U.S. wages and rates of adjustment for the border cities. The error term sj, is the typical idiosyncratic error term found in panel data estimation, less the groupspecific effect. If wages in the Mexican interior converge to U.S. wages, then the noninteracted effect in Z2 should be negative. If the interior responds to U.S. shocks then the noninteracted effect in z, should be greater than zero and significant. The second set of hypotheses question whether shocks affecting the United States have a stron-

15 VOL. 90 NO. 4 ROBERTSON: WAGE SHOCKS AND LABOR-MARKET INTEGRATION 755 ger effect in the border region of Mexico than in the interior of Mexico, and whether the rate at which wages return to their equilibrium differential is greater on the border. These hypotheses are tested using the significance of the coefficients on the interaction terms both as a set and individually. If the border cities are more responsive to U.S. shocks, then the interaction terms in z1 should be positive and significant. If the border cities converge more rapidly to the equilibrium differential between the Mexican and U.S. wage, then the interaction terms in Z2 should be negative and significant.17 There are three estimation issues. First, since the data are differenced, we expect no individualspecific fixed effects to remain. Breusch-Pagan Lagrangian multiplier tests confirm this expectation. Second, the variance of the disturbance term may vary across age-education groups. In the results I present robust standard errors, which allow for clustering around the ageeducation groups and account for this potential heterogeneity. Third, following Deaton (1985), I weight observations using the age-education group cell sizes. This technique increases efficiency because it uses information about the accuracy of the cell means (as suggested by the population size used to calculate each cell mean). Table 4 shows the estimation results for ageeducation group wages averaged over males and females. Both main effects are significant at the 1 percent level and have the expected sign. The core hypothesis thus receives support: wages in the United States affect wages in Mexico. The positive sign on the interaction terms suggests 17 When taken as a whole, these hypotheses may seem to suggest that, if true, wages in the United States and Mexico are cointegrated. There are two necessary conditions for cointegration. The first is that the series have a unit root. The second condition is that the difference between the series be stationary. Only if the first condition is satisfied is the second considered. To examine the hypothesis of a unit root in these panel data sets, I applied both the test of Andrew Levin and Chien-Fu Lin (1993) and the test of Kyung S. Im et al. (1995). Using both tests, the data fail to reject the null of stationarity. That is, neither wage series exhibits a unit root. This may be surprising, but the time period covered is short ( ) and is considered a period without wage growth in both countries. Since the first condition of nonstationarity is not satisfied, the series cannot be cointegrated (even though the difference between the two series is obviously stationary). TABLE 4-UNITED STATES WAGE SHOCKS (MALES AND FEMALES TOGETHER BY EDUCATION GROUP) Regression equation (13): Awg = z0d + [z1d]awjut + [z2(wg - WUS)1_]D A. Wage shock + t All At least 9 Less than education years of 9 years of groups education education Interior (0.018) (0.041) (0.020) Tijuana (0.026) (0.052) (0.028) Ciudad Juarez (0.023) (0.073) (0.022) Matamoros (0.029) (0.078) (0.029) Nuevo Laredo (0.027) (0.065) (0.029) F-test p-value B. Convergence Interior (0.020) (0.023) (0.022) Tijuana (0.012) (0.015) (0.014) Ciudad Juarez (0.013) (0.017) (0.014) Matamoros (0.011) (0.016) (0.014) Nuevo Laredo (0.012) (0.017) (0.015) F-test p-value Intercept R N 15,090 7,125 7,965 Notes: Heteroskedasticity-consistent standard errors are in parentheses. The main city effects and year dummy variables are included in each regression but are not shown. The interior represents wages averaged over the three interior regions described in the previous tables. The interior is the omitted variable and thus represents the main effect. The border city coefficients are the interaction terms. Thus, the total effect for a border city is that city's coefficient plus the interior coefficient. The F-test is the joint significance test of the border interaction coefficients. Cell sizes were used as weights in estimation. that, relative to the interior, U.S. shocks have a stronger effect on each border city. The difference is statistically significant for every city except Nuevo Laredo. The F test suggests that the group of interaction terms is significantly

16 756 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 2000 different from zero. This supports the prediction that U.S. wage shocks have a stronger effect on the border than on the interior of Mexico. The second group of interaction terms is significant at the 1-percent level and all suggest that border wages converge to the U.S.-border wage differential at a much faster rate than wages in the interior. Thus, the second hypothesis also receives support: there is an important geographic component to labor-market integration. Boyer and Hatton (1994) suggest that the speed of convergence can be estimated as (1 - 'y2)/y2. As suggested earlier, this provides one metric for labor-market integration: the speed of convergence to an equilibrium differential. Calculating this metric with the estimates in Table 4 suggests that wages in Tijuana would converge to the equilibrium differential in just over one quarter, whereas it would take more than two quarters for interior wages to converge. As a group, the interaction terms are significant at the 1-percent level as well. Note that the point estimates also suggest that wages in Tijuana and Ciudad Juarez converge at a faster rate than that of wages in the other two border cities, as would be expected if foreign investment and migration were important forces integrating labor markets. The next step is to explore the robustness of the results. Workers with varying levels of education may respond differently or have wages that converge at different rates. For example, Borjas (1994) finds that the education level of migrants has been falling, which may suggest that less-educated workers are more likely to migrate. To investigate this possibility of differences across education groups, I divided the sample into those with less than 9 years of education and those with at least 9 years of education and performed the same regression. The results are found in the second and third columns of Table 4. Again, the main effects are highly significant for both education groups, and larger in magnitude for more educated workers. This suggests more-educated workers in Mexico City may be more sensitive to changes in U.S. labor markets. One explanation for this may be that they may be less constrained by transportation costs from the interior. That is, less-skilled workers may first pass through the border region to migrate, whereas more-skilled workers in the interior may fly directly. For the more educated, the interaction terms on the effect of U.S. shocks are not significant for three border cities. In Nuevo Laredo, the coefficient is negative, significant, and nearly half the magnitude of the main effect. This is somewhat surprising, but perhaps less so when one recalls that Nuevo Laredo has less migration, fewer maquiladora plants and jobs, and less recent interstate migration than the other three border cities. Thus it does not seem surprising that the city for which the rate of wage convergence for the more educated is most similar to the interior is also Nuevo Laredo. For those with less education, three of four shock interaction terms are statistically significant. All have the expected sign. As a group and individually, all of the convergence interaction terms are significant and have the expected sign. Convergence is much higher in Tijuana and Ciudad Juarez. Since these cities all have more maquiladoras, more migration, and are closest to major U.S. cities-all factors that may help integrate labor markets-these results seem consistent with the model. Taken as a whole, the hypotheses are supported by both education groups. The main effects and interaction terms are significant and have the expected sign in all regressions except for the more-educated workers. These results seem consistent with findings that migrants have less education and are consistent with the hypothesis that migration is an important factor integrating labor markets.18 The results in Table 4 were generated from data that were averaged over males and females. If males and females have different labormarket experiences-as suggested by the summary statistics-males and females should be considered separately. Table 5 contains results from estimating equation (13) with data averaged over males (the first three columns) and again with data averaged over females (the second three columns). The fit of the regression, as measured by the R2, is roughly equal for males and females. Again, both main effects are significant and have the expected sign. For all males, the effects of the border city shock terms have the expected sign and are statistically sig- 18 These patterns of sign and significance also emerge when Mexico City, Mexico State, and Jalisco are used independently as representatives of the interior, although magnitudes vary somewhat across different interior regions.

17 VOL 90 NO. 4 ROBERTSON: WAGE SHOCKS AND LABOR-MARKET INTEGRATION 757 TABLE 5-U.S. WAGE SHOCKS (MALES AND FEMALES SEPARATELY BY EDUCATION GROUP) Males Females A. Wage shock (i) (ii) (iii) (iv) (v) (vi) All At least 9 Less than All At least 9 Less than 9 education years of 9 years of education years of years of groups education education groups education education Interior (0.014) (0.023) (0.017) (0.024) (0.040) (0.023) Tijuana (0.020) (0.054) (0.022) (0.032) (0.076) (0.034) Ciudad Juarez (0.023) (0.085) (0.024) (0.041) (0.064) (0.046) Matamoros (0.035) (0.063) (0.039) (0.035) (0 042) (0.038) Nuevo Laredo (0.028) (0.050) (0.032) (0.039) (0.055) (0.045) F-test p-value B. Convergence Interior (0.019) (0.026) (0.017) (0.020) (0.020) (0.022) Tijuana (0.013) (0.019) (0.013) (0.016) (0.019) (0.021) Ciudad Juarez (0.012) (0.020) (0.014) (0.020) (0.018) (0.026) Matamoros (0.013) (0.018) (0.016) (0.021) (0.025) (0.022) Nuevo Laredo (0.011) (0.016) (0.014) (0.018) (0.019) (0.021) F-test p-value Intercept R N 14,674 6,978 7,696 11,606 5,503 6,103 Notes: This table contains the results from six regressions (2 genders X 3 education groupings). Heteroskedasticity-consistent standard errors are in parentheses. The main city effects and year dummy variables are included in each regression but are not shown. The interior represents wages averaged over the three interior regions described in the previous tables. The interior is the omitted variable and thus represents the main effect. The border city coefficients are the interaction terms. Thus, the total effect for a border city is that city's coefficient plus the interior coefficient. The F test is the joint significance test of the border interaction coefficients. Cell sizes were used as weights in estimation. nificant in three. Comparing point estimates between Tables 4 and 5 suggests that wages of males converge slightly more rapidly than in the pooled sample. This may be explained by the higher propensity of males to migrate, as suggested by the higher share of male migrants in the United States. Males also adjust more rapidly in the border cities. The F-test shows that the interaction terms for both the shock and convergence terms are jointly significant. A similar story emerges when males are divided by education level. In each case, the sets of interaction terms are jointly significant at the 5-percent level. Both more- and less-educated males experience faster wage convergence in Tijuana and Ciudad Juarez. For both education groups, the rank order of the border convergence coefficients follows that of apprehensions and the number of maquiladora plants (but not maquiladora employment).

18 758 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 2000 The results for females are strong with the exception of more-educated females. When all education groups are pooled, all main and interaction coefficients are significant at the 2- percent level. The effect of U.S. wage shocks and the rate of convergence is much higher for females on the border than in the interior. Again, Tijuana and Ciudad Juarez have the largest coefficients for the convergence terms. The sharp contrast between the border and interior for females is like the sharp contrast in female participation in manufacturing (Table 1), which, unlike the same statistic for males, sharply rises when one moves from Mexico City to a border city. As with males, labor-market integration differs by education group for females. The effect of shocks and rates of convergence are higher for less-educated females in every border city. The point estimates for both the effect of shocks and rate of convergence are especially higher for less-educated females in Ciudad Juarez and Tijuana. Given that Ciudad Juarez is a major emigration point (presumably into Texas), it is interesting to note that the percentage of female Mexican-born people in Texas is higher than that in California. Again, the results from Tables 4 and 5 suggest that differentiating males and females generate results that strongly support the hypotheses about labor-market integration. Since nearly 80 percent of all Mexican migrants end up in either California or Texas, one may wonder how shocks in California and Texas affect Mexican wages. Table 6 shows the results of estimating equation (13) using the average wage in Texas and California in place of the U.S. wage. There are three interesting results. First, the effect of the wage in U.S. border states on Mexico City is smaller than the effect of the U.S. average wage (comparing point estimates in Table 6 with those in Table 4), perhaps because some migrants choose other U.S. states as final destinations. Second, although the magnitudes are smaller, the effects of shocks in the U.S. border states are similar. Both the shock and convergence terms have the expected sign, and the effects on Ciudad Juarez and Tijuana are stronger than those in the other two border cities. Third, dividing the sample by education groups also generates similar results. The hypothesis that the interaction terms are TABLE 6-TEXAS AND CALIFORNIA WAGE SHOCKS (MALES AND FEMALES TOGETHER BY EDUCATION GROUP) A. Wage shock All At least 9 Less than education years of 9 years of groups education education Interior (0.014) (0.024) (0.016) Tijuana (0.018) (0.024) (0.023) Ciudad Juarez (0.015) (0.021) (0.018) Matamoros (0.020) (0.046) (0.018) Nuevo Laredo (0.019) (0.038) (0.019) F-test p-value B. Convergence Interior (0.015) (0.020) (0.019) Tijuana (0.010) (0.013) (0.014) Ciudad Juarez (0.011) (0.016) (0.014) Matamoros (0.010) (0.012) (0.012) Nuevo Laredo (0.010) (0.016) (0.011) F-test p-value Intercept R N 14,047 6,942 7,105 Notes: This table contains the results from three regressions (one for each education grouping). Heteroskedasticity-consistent standard errors are in parentheses. The main city effects and year dummy variables are included in each regression but are not shown. The interior represents wages averaged over the three interior regions described in the previous tables. The interior is the omitted variable and thus represents the main effect. The border city coefficients are the interaction terms. Thus, the total effect for a border city is that city's coefficient plus the interior coefficient. The F-test is the joint significance test of the border interaction coefficients. Cell sizes were used as weights in estimation. jointly equal to zero (the F-test) is rejected at the 1-percent level for both the shock and convergence terms and all education groups. These results are robust to the same division between males and females and by education groups (results available on request from the author). Another question is how integrated the Mex-

19 VOL. 90 NO. 4 ROBERTSON: WAGE SHOCKS AND LABOR-MARKET INTEGRATION 759 TABLE 7-EFFECTS OF INTERIOR WAGE SHOCKS ON MEXICAN BORDER CITIES BY EDUCATION GROUP, MALES AND FEMALES A. All education groups Ciudad Nuevo Tijuana Juarez Matamoros Laredo Shock (0.041) (0.037) (0.035) (0.039) Convergence (0.033) (0.033) (0.021) (0.039) N 3,026 2,972 2,964 2,971 R B. At least 9 years of education Shock (0.060) (0.050) (0.055) (0.058) Convergence (0.042) (0.042) (0.033) (0.056) N 1,408 1,360 1,363 1,375 R C. Less than 9 years of education Shock (0.053) (0.055) (0.042) (0.048) Convergence (0.036) (0.047) (0.025) (0.048) N 1,618 1,612 1,601 1,596 R Notes: This table contains the results from 12 regressions (4 cities X 3 education groupings). Heteroskedasticity-consistent standard errors are in parentheses. Cell sizes were used as weights in estimation. ican border region is with the Mexican interior. Table 7 contains results from estimating equation (6) with the change in the border wage regressed on the change in the interior wage and the lagged log difference between the two wages. Equations for each border city were estimated separately. If cross-border migration or transport costs are higher than those for travel within Mexico, we would expect both the effects of shocks and the rate of convergence to be higher in these regressions. However, if the border cities are more integrated with the United States than with the Mexican interior, then the estimated coefficients in Table 7 should be smaller in absolute value than those found in Tables 4-6. All of the coefficients in Table 7 have the expected sign and are highly significant (both statistically and economically). The point esti- mates suggest several interesting results. Overall, the coefficient estimates are generally larger than those in the previous tables. This is consistent with the hypothesis that the border matters, as other authors have found for other markets (McCallum, 1995; Engel and Rogers, 1996). Furthermore, workers in cities that are more integrated with the United States are less integrated with the Mexican interior. Shocks in the interior have a larger effect on Nuevo Laredo and Matamoros and wages in these cities converge more rapidly to interior wages than wages in Tijuana and Ciudad Juarez. This "reverse" pattern is also apparent when education groups are distinguished. More-educated workers in the Mexican border are more closely tied to the interior than are less-educated workers. Overall these results are consistent with the hypothesis that within-country migration costs are lower than between-country migration costs and that foreign investment (maquiladoras) and international migration are important factors integrating labor markets. Given the large wage differential between the two countries, it is remarkable that the effect of foreign and domestic shocks and rates of convergence are as similar as they are in Tijuana and Ciudad Juarez. These results are robust to the division according to gender and education group. D. Exchange-Rate Movements and NAFTA The evidence presented in Tables 1-7 suggest that U.S. and Mexican labor markets are integrated and that the pattern of integration differs across region and demographic characteristics in ways that are consistent with the hypothesis that capital flows and migration are important factors integrating labor markets. On January 1, 1994, the North American Free Trade Agreement (NAFTA) went into effect, which-at least according to popular rhetoricwas expected to have large effects on migration, capital flows, and good flows. Thus, the period examined in this study lends itself to the possibility of considering the "natural experiment" presented by NAFTA. As Figure 2 shows, however, Mexico experienced sharp devaluations of the peso in November 1987 and December Since the Mexican government alternated between floating and crawling-peg exchange-rate regimes, it

20 760 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 2000 is importanto first explore the robustness of the results in the previous section to exchange-rate movements. Dramatic movements in the peso could bias the effects in two ways. First, if changes in the U.S. wage and changes in the peso were correlated, then the regression equation (13) would measure this correlation rather than the relationship between the U.S. and Mexican wages. To evaluate this possibility, I regressed changes in the peso on changes in the U.S. production-worker average wage series from the Bureau of Labor Statistics.19 I found no significant correlation. This suggests that this first source of bias is not important. The other source of bias involves measurement error. Excessive volatility or government intervention in the exchange-rate market would mask true changes in Mexican wages because the exchange rate may not perfectly offset differences in inflation across countries. This would bias the shock coefficient toward zero and generally bias the convergence coefficient downward because the exchange-rate movements would magnify the measured responsiveness to the lagged difference in wages. I evaluate the possible effects of this bias in two ways. First, I regress equation (13), excluding the periods of dramatic exchange rate movements. The first column of Table 8 contains these results. The results suggest that the bias does not affect the main conclusions in the paper. All coefficients have the expected signs and the difference between the interior and the border region remains statistically significant. This result also emerges when the United States is represented by California and Texas rather than the entire United States and when I distinguish workers by gender and education. Interestingly, excluding 1987 and 1995 does have one interesting effect-it reduces the measured degree of integration of the interior region (a move in the opposite direction of the expected bias). However, both the effect of U.S. shocks on the border region and the rate of convergence in the border region remain about the same. The marginal effects of the border region adjust with the change in the interior coefficient to keep the total effect (the sum of the main and interaction 19 The data used in this regression are monthly data from January 1983 to June TABLE 8-EXCHANGE-RATE A. Wage shock ROBUSTNESS CHECKS (MALES AND FEMALES IN ALL EDUCATION GROUPS) Excluding CPI- CPI and adjusted adjusted 1995 wages wages Interior (0.018) (0.017) (0.016) Tijuana (0.029) (0.030) (0.030) Ciudad Juarez (0.026) (0.026) (0.026) Matamoros (0.031) (0.031) (0.030) Nuevo Laredo (0.030) (0.030) (0.030) F-test p-value B. Convergence Interior (0.019) (0.017) (0.018) Tijuana (0.014) (0.020) (0.020) Ciudad Juarez (0.013) (0.025) (0.025) Matamoros (0.011) (0.019) (0.019) Nuevo Laredo (0.013) (0.023) (0.024) F-test p-value A Exchange rate (0.015) Intercept R N 12,648 15,090 15,090 Notes: This table contains the results from three regressions. Heteroskedasticity-consistent standard errors are in parentheses. Cell sizes were used as weights in estimation. coefficients) about equal. This may suggest that large exchange-rate movements alter the behavior of agents in the interior but not the border, but exploration of this hypothesis is beyond the scope of this paper. The second way I evaluate the effects of exchange rates is by generalizing equations (6) and (12). First, I separate the log(mexican wages) into their (nominal) peso value and the log(exchange rate). Next, I subtract the differenced log(exchange rate) from both sides to include the change in the log(exchange rate) as

21 VOL. 90 NO. 4 ROBERTSON: WAGE SHOCKS AND LABOR-MARKET INTEGRATION 761 a regressor.20 The implied coefficient on the differenced exchange rate is 1. This specification lends itself to a test of how well the exchange rate adjusts over the sample period to reflect differences in price levels across countries because the wages are now expressed in nominal units. When this generalized form is estimated, the resulting coefficient (standard error) on the differenced exchange rate is (0.021), which is significantly different from 1. This result suggests that the exchange rate only partially reflects differences in inflation across countries (i.e., purchasing power parity does not hold). The dramatic growth in the Mexican current account balance over the sample period reflects this result. It is worth noting that the other parameters in the estimation-the shock and convergence coefficients-are statistically significant and have the expected sign. Furthermore, the border's shock and convergence terms are significantly different from those in the interior, although the magnitude of the difference is smaller than that in Table 4. The theoretical model in Section I, however, is not based on nominal wage shocks but rather is based on a marginal product analysis. Since the exchange rate may only partially adjust for differences in price levels across countries, I regress a generalized version of equation (13) with real wages, in which the nominal wages are adjusted by each country's consumer price index (CPl).21 The second and third columns of Table 8 contain this generalized version of (13). In the second column, I exclude the exchange rate and regress (13) using wages adjusted for inflation in each country. In the third column, I include the differenced log(exchange rate). Several patterns emerge from Table 8. First, the qualitative results are robust to both methods of controlling for the exchange rate. All of the coefficients have the expected sign and nearly all are significantly different from zero. The Mexican border cities still appear to be much more integrated with the United States. Including the exchange rate does not affect the 20 The lagged value of the log(exchange rate) also appears in the generalized equation. 21 Adjusting wages by the domestic CPI removes the upward slopes seen in Figure 4, but otherwise generates a very similar pattern of changes across time. overall pattern of the results. The coefficient (standard error) on the exchange rate is (0.015), suggesting that changes in the exchange rate do not affect real wages.22 The only apparent effect of including the exchange rate as a regressor is that the difference between the border and the interior becomes starker. The rates of convergence for Tijuana and Ciudad Juarez remain much higher than those for the other two cities and the rate is still greatest in Tijuana. Furthermore, the magnitudes of the shock and convergence terms are very similar in Tables 4 and 8. For example, the shock (convergence) terms for Tijuana are (-0.486) in Table 4, (-0.493) in the first column of Table 8, and (-0.487) in the third column of Table 8. Comparisons for the other regions yield si-milaresults. The same pattern emerges when workers are divided by education group. Thus, the results from the earlier tables seem quite robust to controlling for exchange rate movements. Since exchange-rate movements do not affect the results in this paper, I next examine the hypothesis that NAFTA changed the datagenerating process, affecting my measures of labor-market integration. To allow for the possibility of a NAFTA-induced structural change, I perform the same regression as in Table 4 but with an additional dummy variable equal to 1 for the period. I interact the NAFTA-period dummy variable with the main and interaction effects and test the individual and joint significance of the resulting coefficients. Table 9 contains results from three regressions: without exchange-rate controls, excluding 1987 and 1995, and using the generalized equation with CPI--adjusted wages. All three regressions shown in Table 9 use data pooled over males and fernales. Two main results emerge from regressions without exchange rate controls. First, the size of the main effects (the response to shocks in the interior) diminishes sharply when the extra set of interaction terms is inicluded. Second, the joint significance tests suggest that the effects of shocks are not significantly different between 22 This result reflects quarterly changes over the whole sample period and does not necessarily imply that the December 1994 depreciation did not affect real wages.

22 762 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 2000 TABLE 9-MARGINAL EFFECTS OF NAFTA PERIOD (MALES AND FEMALES IN ALL EDUCATION GROUPS) A. Wage shock No exchange-rate controls Excluding 1987 and 1995 CPI-adjusted wages Main Main Main Interior (0.021) (0.039) (0.017) (0.027) (0.017) (0.031) Tijuana (0.040) (0.058) (0.036) (0.056) (0.038) (0.053) Ciudad Juarez (0.038) (0.071) (0.036) (0.070) (0.036) (0.065) Matamoros (0.043) (0.068) (0.041) (0.078) (0.038) (0.066) Nuevo Laredo (0.037) (0.060) (0.039) (0.059) (0.036) (0.058) F-test p-value B. Convergence Interior (0.006) (0.007) (0.009) (0.009) (0.016) (0.014) Tijuana (0.014) (0.004) (0.014) (0.006) (0.019) (0.005) Ciudad Juarez (0.014) (0.002) (0.014) (0.003) (0.027) (0.002) Matamoros (0.013) (0.002) (0.012) (0.003) (0.018) (0.002) Nuevo Laredo (0.013) (0.002) (0.013) (0.003) (0.023) (0.002) F-test p-value A Exchange rate (0.009) Intercept R N 15,090 12,648 15,090 Notes: This table contains the results from three regressions. Heteroskedasticity-consistent standard errors are in parentheses. Cell sizes were used as weights in estimation. the border and the interior in the NAFTA years, but the rates of convergence are significantly larger. In general, the signs of the interaction terms suggest that labor markets were more integrated in the NAFTA years. The results from performing individual regressions for each city, using Texas and California rather than the entire United States, and dividing the sample by gender and education groups are similar. Overall, the results from these regressions (available on request from the author) also suggest that the shock terms are not significantly different but the rates of convergence increased during the NAFTA period. Although the main results of this paper are not altered by excluding periods of large exchange-rate movements, the effects observed during the NAFTA period are potentially more sensitive because the NAFTA period coincided with the dramatic peso devaluation of The second and third regressions in Table 9 include exchange-rate controls. Here the expected bias becomes more apparent. When the volatile years are excluded, the main shock and convergence coefficients return to magnitudes more similar to those found in earlier tables and are once again statistically significant. The NAFTA-period interacted shock terms are gen-

23 VOL. 90 NO. 4 ROBERTSON: WAGE SHOCKS AND LABOR-MARKET INTEGRATION 763 erally not significant but are quite large (this is especially true for more-educated workers). What may be more interesting, however, is that once the volatile years are excluded, the rateof-convergence interaction terms that represent the NAFTA period switch sign and become generally positive. This result is also robust to all variations explored in Tables 4-6 and is consistent with the prediction that NAFTA would reduce migration by generating opportunities for Mexican workers in Mexico. The results from the generalized equation are very similar to those generated by excluding 1987 and However, there are at least three reasons why these results may not be directly the result of NAFTA. First, NAFTA was anticipated in Mexico from the time that negotiations were announced in the early 1990' s. Economic actors may have altered behavior at any time before, during, or after the implementation of NAFTA. Second, the full provisions of NAFTA will take 15 years to enter into force and the earliest provisions were the least controversial. Third, the peso collapse in late December 1994 triggered a serious recession, which lasted beyond 1995 and may mask the responsiveness of Mexican wages to changes in U.S. wages. Each of these reasons suggests that it may be too early to evaluate the complete effects of NAFTA on labor-market integration. III. Conclusion Using a simple model of aggregate labor supply and demand and household-level data, this paper finds that even before NAFTA, U.S. and Mexican labor markets were integrated. Although the border matters for the absolute wage differential, U.S. and Mexican labor markets are integrated. U.S. wages affect Mexican wages and, when shocked, wages converge back to an equilibrium differential. Wages in the border region exhibit stronger responses and tend to converge faster to the equilibrium U.S.- Mexican wage differential. This result is robust across education and gender groups and across periods of large currency movements. Since changes in wages may be transmitted across borders through capital movements, trade, and migration, determining the contribution of each factor to the transmission of shocks is increasingly relevant for academics and important for policymakers. Flows of labor, goods, and capital have all increased between the United States and Mexico and continue to rise. In this study, Tijuana, the border city with the highest migration flows, consistently exhibits the most rapid convergence and strongest effect of U.S. wage shocks. This is consistent with the idea that migration is the most important mechanism integrating U.S. and Mexican labor markets. Much care must be taken with this conclusion, however, because data restrictions precluded direct tests of the migration hypothesis. Discerning the exact mechanisms of labor market integration, and the relative contribution of each, is a fertile area for future research. Finally, this paper takes a preliminary step toward evaluating the effect that NAFTA may have on labor-market integration. Controlling for periods of large exchange-rate movements, results presented here suggest that the rate of convergence between U.S. and Mexican wages slowed during the NAFTA period. This effect was larger for the interior than for the border. However, it is too early to fully evaluate the effect of NAFTA on labor-market integration. Future research in this area would be extremely fruitful as trade agreements in the Western Hemisphere and other regions proliferate. REFERENCES Abowd, John M. and Freeman, Richard B. Immigration, trade, and the labor market, NBER Project Report. London: University of Chicago Press, Blonigen, Bruce A. "Firm-Specific Assets and the Link Between Exchange Rates and Foreign Direct Investment." American Economic Review, June 1997, 87(3), pp Borjas, George J. "The Substitutability of Black, Hispanic, and White Labor." Economic Inquiry, January 1983, 21(1), pp "The Economics of Immigration." Journal of Economic Literature, December 1994, 32(4), pp Boyer, George R. and Hatton, Timothy J. "Regional Labour Market Integration in England and Wales, ," in George Grantham and Mary MacKinnon, eds., Labor market evolution: The economic history of market integration, wage flexibility and the

24 764 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 2000 employment relation. London: Routledge, 1994, pp Browning, Martin; Deaton, Angus and Irish, Margaret. "A Profitable Approach to Labor Supply and Commodity Demands over the Life- Cycle." Econometrica, May 1985, 53(3), pp Card, David. "The Impact of the Mariel Boatlift on the Miami Labor Market." Industrial and Labor Relations Review, May 1990, 43(2), pp Deaton, Angus. "Panel Data from Time Series of Cross-Sections." Journal of Econometrics, October-November 1985, 30(1-2), pp Engel, Charles and Rogers, John H. "How Wide Is the Border?" American Economic Review, December 1996, 86(5), pp Feenstra, Robert and Hanson, Gordon H. "Foreign Direct Investment and Relative Wages: Evidence from Mexico's Maquiladoras." Journal of International Economics, May (3-4), pp Gil-Diaz, Francisco. "The Origin of Mexico's 1994 Financial Crisis." Cato Journal, Winter 1998, 17(3), pp Hamermesh, Daniel S. Labor demand. Princeton, NJ: Princeton University Press, Hanson, Gordon H. "Economic Integration, Intraindustry Trade, and Frontier Regions." European Economic Review, April 1996, 40(3-5), pp "Regional Adjustment to Trade Liberalization." Regional Science and Urban Economics, July 1998, 28(4), pp Hanson, Gordon H.; Robertson, Raymond and Spilimbergo, Antonio. "Does Border Enforcement Protect U.S. Workers From Illegal Immigration?" Working Paper No. 7054, NBER, Hanson, Gordon H. and Spilimbergo, Antonio. "Illegal Immigration, Border Enforcement, and Relative Wages: Evidence from the U.S.- Mexico Border" American Economic Review, December 1999, 89(5), pp Hendry, David F. and Ericsson, Neil R. "An Econometric Analysis of U.K. Money Demand in Monetary Trends in the United States and the United Kingdom by Milton Friedman and Anna J. Schwartz." American Economic Review, March 1991, 81(1), pp Im, Kyung S.; Pesaran, M. Hashem and Shin, Yongcheol. "Testing for Unit Roots in Heterogeneous Panels." Working Paper, Amalgamated Series: 9526, University of Cambridge, Instituto Nacional de Estadistica, Geografia e Informatica. Estadistica de la Industria Maquiladora de Exportacion Aguascalientes, Mexico, Levin, Andrew and Lin, Chien-Fu. "Unit Root Tests in Panel Data: New Results." Discussion Paper No , University of California, San Diego, McCallum, John. "National Borders Matter: Canada-U.S. Regional Trade Patterns" American Economic Review, June 1995, 85(3), pp Topel, Robert H. "Local Labor Markets." Journal of Political Economy, June 1986, pt. 2, 94(3), pp. S11J-43. Trejo, Stephen J. "Why Do Mexican Americans Earn Low Wages?" Jourmal of Political Economy, December 1997, 105(6), pp Zabin, Carol A. and Hughes, Salle. "Economic Integration and Labor Flows: Stage Migration in Farm Labor Markets in Mexico and the United States." Intermational Migration Review, Summer 1995, 29(2), pp