1 Introduction

Trade openness facilitates rapid economic growth in large developing countries such as Brazil, China, India, and Mexico. For example, Rani and Kumar (2019) find evidence that trade openness causes economic growth in China, India, and Brazil. Studies like Majid (2004) find that trade openness enhances wage rates in developing countries, but only through economic growth. However, it mentions that trade openness might cause short-run declines in real wage rates and increase dispersions due to demand-driven labor market adjustments. These promises of growth and higher wages led these countries to lower tariffs and other trade barriers, reducing barriers to foreign direct investment (FDI) and entering into complex trade agreements (Robertson 2007; Harrison 2007). Others argue that globalization has accentuated wage inequality (Anzaldo Gómez et al. 2008), leaving some workers and locations behind. This paper asks whether trade openness, through NAFTA-related tariff changes,Footnote 1 increased or decreased wage inequality in Mexico.

In recent years, a growing body of literature has emerged that investigates the impact of trade liberalization on local economies, particularly focusing on the effects of tariff reductions and regional disparities (Behrens et al. 2007; Krugman 1991; Hanson 1998). These studies provide valuable insights into how changes in trade policy affect different locations based on their proximity to trading partners and their pre-existing economic conditions. As part of this broader research landscape, it is important to consider how reductions in US tariffs would benefit municipalities close to the US–Mexico border. These locations are generally characterized by well-established cross-border supply chains and enjoy lower transportation costs due to their proximity to the US market. As a result, they experienced a more significant boost in their local economies, with industries such as manufacturing taking advantage of the reduced trade barriers. However, the benefits of these tariff reductions are not equally distributed across all Mexican municipalities. Municipalities located farther from the US–Mexico border and with a less-developed economic base have benefited less from the tariff reductions. The main reason is that these municipalities often exhibit higher transportation costs and need more infrastructure to support cross-border trade. Additionally, they do not have the same access to resources and knowledge networks as their counterparts closer to the border, which hindered their ability to capitalize from NAFTA (Aroca et al. 2005; Baylis et al. 2012; Rodríguez-Pose and Sánchez-Reaza 2005).

Standard trade theory would predict that trade benefits are not equally distributed. A simple Heckscher–Ohlin model would suggest that if inputs are not completely mobile across sectors and regions, we expect factors employed in the export-oriented sectors to benefit more than those in import-competing industries. Further, we might expect those regions with lower transport costs to export markets to benefit more. Additionally, the economic effects of trade may increase the concentration of economic activity in certain regions more than others, resulting in increased wage disparities (Krugman 1991). If trade increases wages in some industries and locations, it might also induce internal migration among a subset of mobile workers. This migration can, in turn, either mitigate or exacerbate these wage differentials.

While increased trade may have benefited the Mexican economy overall, initial evidence shows that NAFTA may have worsened wage differentials in Mexico (e.g., Baylis et al. 2012; Nicita 2009).Footnote 2 Nicita (2009) shows that trade benefits have primarily gone to more skilled workers, especially in Mexican states near the US border.Footnote 3 Similarly, Hanson (2007) and Baylis et al. (2012) find that Northern states, which face lower transportation costs to the US market than the Southern states, benefited more from trade by obtaining higher prices, translating into higher wages. One disadvantage of these papers is that they do not consider that households may respond to changes in labor demand by altering the type of labor they sell or relocating. The distribution of benefits from NAFTA will presumably accrue to those already working in export industries or living in regions close to the US border and those who can more easily migrate into those regions and sectors. Conversely, the economic structural shift brought about by trade may penalize those facing higher migration barriers.

It is important to notice that if Mexico´s labor markets were fully integrated, internal migration would operate to equalize wages across labor markets, avoiding trade openness differentials in different sectors and regions. However, Mexican labor is sticky across regions. Only those who migrate benefit from better salaries. Thus, not considering how internal migration influences wages due to trade openness will miss an important part of the process since workers with low wages, especially from Mexico's rural south, migrate to the north for better salaries. This migration benefits workers from the rural south, who migrated to the north, by allowing them to get higher wages, and as a result, diminishes the regional wage gap. Thus, internal migration acts as a lubricant to diminish the wage gap, which could be an even higher regional wage disparity if internal migration were not considered.

Migration within Mexico is substantial. Five percent of working-age men migrated from one state to another between 1995 and 2000 (Vega 2005; INEGI 2008).Footnote 4 A growing body of literature on the effect of migration on wages in Mexico focuses on the international labor movement. Hanson and McIntosh (2010) find that the increase in the labor supply in Mexico places downward pressure on wages, making external migration more attractive. External migration increases Mexican wages (Mishra 2007) and can promote local development through remittances (Unger 2005). Aroca and Maloney (2005) find that trade and FDI slow migration because increased linkages to global markets decrease the incentive to emigrate. However, if trade affects different regions within a country in different ways, it might induce internal migration, making benefits from trade available primarily to those households who can move (Arends-Kuenning et al. 2019). We disentangle the impact of NAFTA on the wage differential due to trade effects and distinguish this from the effect of the subsequent internal migration on the wage differential. In the short run, trade changes cause changes in prices and wages. In the medium-to-long term, workers respond to these changes through labor supply and internal migration. Both processes will affect wage inequality within a region. Suppose one ignores the labor adjustment to regional wage differentials. In that case, one might attribute changes in the wage distribution within a location to trade openness, which might also reflect a locational sorting of workers.

This paper offers the following contributions: First, to our knowledge, this is one of the first studies to consider the effect of trade on wages while explicitly controlling for migration.Footnote 5 Second, we address the simultaneity of migration and local urban wages by instrumenting for migration using agricultural shocks in the sending locations. Third, we explore which workers gained and lost from trade. Fourth, industrial changes brought about by trade are likely to emerge only fully over time. We include the 2010 population census to observe long-run wage changes after NAFTA and ask whether wage differentials persist as the economy adapts to trade. This research can help identify those barriers facing individuals and regions that limit their ability to benefit from trade and identify the areas of social investment and infrastructure investmentFootnote 6 that may help smooth wage inequality. Further, this information can target compensation by identifying those regions and individuals who have benefited and lost from trade. Finally, regional governments can anticipate migration and wages in their region using this estimation approach and adjust local development plans accordingly.

2 Literature review

Developing countries, such as Brazil, China, India, and Mexico, have experienced rapid economic growth. They have made significant policy adjustments to foster globalization, including lowering tariffs and other trade barriers, reducing barriers to foreign direct investment (FDI) and entering into complex trade agreements. The main motivation for these changes was the promise of growth, higher wages, and lower wage differentials (Robertson 2007; Harrison 2007). While increased trade may have benefited the Mexican economy, some initial evidence shows that NAFTA may have worsened wage differentials in Mexico (Baylis et al. 2012; Nicita 2009).

Trade can affect wage differential across skills, sectors, and regions. The Heckscher–Ohlin model of trade states that countries should benefit overall from trade, and in particular, low-skilled labor should reap higher wages in developing countries where such labor is abundant. If inputs are not completely mobile across sectors and regions, we expect factors employed in the export-oriented sectors to benefit more than those in import-competing industries. Further, we might expect those regions with lower transport costs to export markets to benefit more, which, if labor is not freely mobile, may either improve or exacerbate wage inequality depending on whether those same regions had relatively high or low wages before trade.

A few papers provide evidence of an increase in wage inequality in Mexico after NAFTA.Footnote 7 For example, Nicita (2004) finds that the effect of trade liberalization has been almost exclusively transferred to skilled workers and has increased the gap between the remuneration of skilled and unskilled jobs.Footnote 8 As noted above, Hanson (2007) and Nicita (2009) also show that trade primarily benefited certain skills and regions in Mexico.

The New Economic Geography (NEG) models propose that different factors (such as trade liberalization and labor migration) may lead to disparities across countries. More recently, studies such as Krugman and Livas-Elizondo (1996), Monfort and Nicolini (2000), and Paluzie (2001) have drawn attention to focusing on regional disparities within a country. These models focus either on how trade liberalization potentially affects regional disparities since producers get closer to the border (as is the case of Mexico with NAFTA) to obtain market access or to the country center to benefit from large markets (which is what happened to Mexico before NAFTA, where most of the producers concentrated in Mexico City, where most of the economic activity happened, like a black whole). Based on this, Hanson (1998) has already found that trade openness benefited mainly border regions by increasing economic activity and wages more than in other Mexican regions. Other empirical evidence that shows how proximity to the US–Mexico border has a positive impact on regional wages and employment growth can be found in Hanson (1997), Jordaan and Sanchez-Reaza (2006), and Faber (2007).

New Economic Geography also generates predictions about which regions might reap the gains from trade. The economic effects of trade may increase the concentration of economic activity in certain regions more than in others (Krugman 1991). This concentration generates increased labor demand in these regions and their sectors, which results in increasing wages in these markets. Other effects of trade, such as skill-biased technological change, modifications in industry-specific wage premiums, foreign investment, quality upgrading, skill scarcity, exchange rate, and demographic changes, have all been suggested as being more accurate explanations for the increase in wage inequality (Robertson 2007; Ranjan 2008).

Mexico’s trade liberalization via NAFTA has caused important changes in regional economic growth, exacerbating the disparities between the North and South of Mexico which have existed since industrialization began in the 1930s (López Malo 1960; Hanson 2007; Baylis et al. 2012). The regional distribution of poverty is illustrated in Fig. 1. Here, we observe the poverty headcount, which is the share of people living on less than $2.00 USD per person per day (Walton and López-Acevedo 2004). The darker colors denote states with a higher share of people living on less than $2 dollars per person per day. States in the South, in dark red,Footnote 9 have 76% of their people living on less than two dollars per person per day, whereas Northern states, in light gray,Footnote 10 have only 28% of their population in this situation.

Fig. 1
figure 1

Poverty Headcount 2002

Geography may also play a role in determining the distribution of the benefits of trade. In the case of Mexico, one might anticipate that, due to lower transportation costs, regions closest to the U.S. border, which also tend to be wealthier, might stand to gain from trade. Similarly, those regions with pre-existing export industries, such as the Northern manufacturing centers, would likely benefit the most from trade (Rostow 1960). Further, the urban labor market will benefit more than workers in rural regions because of their higher reliance on skilled wages. In contrast, rural laborers tend to work more in agriculture and often consume most of what they produce (Nicita 2009). Thus, we may expect increasing inter-regional wage disparities, which may induce migration.

There is a growing literature on the effect of migration on wages in Mexico, primarily focused on the effect of the international labor movement. Mishra (2007) finds that “emigration has a strong and positive effect on Mexican wages due to changes in local labor supply” (pg. 180). Unger (2005) also finds a positive link between migration and local development, working through remittances. Aroca and Maloney (2005) find that trade and FDI slow migration in the sense that increased linkages to global markets decrease the incentive to emigrate. However, if trade affects different regions within a country differently, it might induce internal migration, making benefits from trade available primarily to those households who can move (Arends-Kuenning et al. 2019).

In low-skilled labor-abundant developing countries, trade liberalization should increase the income of low-skilled workers, decreasing income disparity. However, anecdotal evidence indicates that NAFTA increased the gap between rich and poor in Mexico, and empirical evidence is mixed (Hirte et al., 2020; Brülhart, 2011; Bosch & Manacorda, 2010; Chiquiar 2008 & 2005; Nicita 2009; Gonzalez-Rivas 2007; Hanson 2007). Because trade may affect wages differently across regions within the country, accurate measures of wage effects must incorporate intra-national migration. We specifically consider rural-to-urban migration and find that working-aged men get a knock in their wages from trade openness. However, trade openness affects those with higher incomes more than those with low incomes, reducing wage disparities. We also find that workers far from the US–Mexico border earn significantly lower wages than their counterparts on the border. As a result, we find that in urban areas, trade liberalization has reduced wage disparities among working-age men.

3 Methods

We estimate the wage equation using two-stage least squares (2SLS). The two equations are the following:

$$P\left({M}_{it}=1| {\Delta {\tau }_{t};\;\text{GVA}}_{it-1};\;{\text{dist}F}_{i};\;\Delta {\tau }_{t}*{\text{GVA}}_{it-1};\;{\text{dist}F}_{i}*\Delta {\tau }_{t}*{\text{GVA}}_{it-1};\;{I}_{it};\;{H}_{it};\;{S}_{rt-5}; \;{\text{NCY}}_{rt}\right)$$
(1)

where \({M}_{it}\) 1 if the individual i migrated within Mexico within the last five years; 0 otherwise, \(\Delta {\tau }_{t}\) % change in Tariffs from t − 1 to t, \({\text{GVA}}_{it-1}\) Total Gross Value Added (GVA) in real 2003 Mexican pesos for municipality-of-destination, \({\text{dist}F}_{i}\) Road distance (in thousands of kilometers) from the capital of municipality-of-destination i to the closest US border crossing point, \({I}_{it}\) Vector of individual characteristics (i.e., education, age, indigenous status, # of working hours, and whether the individual owns a business), \({H}_{it}\) Vector of household characteristics in time t (i.e., electricity, # of people, water, and drainage), \({S}_{rt-5}\) Vector of state-of-originFootnote 11 characteristics r,Footnote 12 in time t − 1, \({\text{NCY}}_{rt}\) Sum of the number of negative changes in corn yields in the last five years in the state-of-originFootnote 13r, in time t

$$\text{ln}\left({\upomega }_{it}\right)=f({\Delta {\tau }_{t};\; \text{GVA}}_{it-1};\;{\text{dist}F}_{i};\;\Delta {\tau }_{t}*{\text{GVA}}_{it-1};\;{\text{dist}F}_{i}*\Delta {\tau }_{t}*{\text{GVA}}_{it-1};\;{I}_{it};\;{H}_{it};\;{S}_{rt-5};\;\widehat{{P(M}_{it})})$$
(2)

where \({\upomega }_{it}\) Observed wage of individual i in year t, \({\widehat{M}}_{it}\) instrumented probability to migrate.

3.1 First stage

In the first stage, we predict the probability that a person migrates as a function of trade openness with the USA, distance to the U.S.–Mexico border, and individual/household/origin-region characteristics. The complete migration function is Eq. (1). To predict migration, Sahota (1968) uses the geographical distance from the region k's capital to the region j's capital. We instead use the distance from the capital of each municipality to the closest US border-crossing point (distFi) from the municipality where the person lives because economic opportunities provided by NAFTA will be greater closer to the US border due to the accessibility to markets (Hanson 1996).

To capture trade openness, we include the change in tariffs (∆τt) and the Gross Value Added (GVA) in location i for period t − 1 (GVAit−1), from the municipality where the person lives to observe whether and how economic growth influences migration and wages. We include an interaction of change in tariffs (∆τt) and the GVAit−1 to capture the potential growth or contraction in GVA associated with a reduction in tariffs in the municipality (∆τt* GVAit−1). Finally, the model includes a triple interaction (distFi*∆τt* GVArt−1), which includes distance to the US border, explained above. This triple interaction shows that the potential growth or contraction of economic activity (GVA) due to trade openness (reduction in tariffs) did not affect all municipalities evenly. We expect that municipalities near the U.S.–Mexico border benefited more from NAFTA than those further away (Baylis et al. 2012).

Because the structure of the local economy and distance to the border may also affect wages, to address the potential endogeneity between regional wages and migration, we include a factor that affects labor supply but not wages other than through changes in local labor supply.

In recent decades, empirical research has emerged studying climate-induced migration through its effect on crop yields. Kaczan and Orgill-Meyer (2020) provide an overview of recent empirical studies and find that the relationship between climate change and migration is complex and context-dependent. Negative climatic conditions, such as droughts, increase migration rates. However, the impact of climate shocks on migration is often mediated by various factors, such as individual and household characteristics, socioeconomic factors, and public policies.

Beine and Parsons (2015) find that climatic factors significantly influence migration patterns, with temperature and precipitation levels having a considerable effect on migration decisions. Moreover, the impact of climate change on migration is expected to be most pronounced for countries with a large agricultural sector, as their populations are more vulnerable to climate shocks. Kleemans (2015) examines the effects of risk and liquidity constraints on migration decisions in rural Indonesia. She finds that the impact of climate shocks on migration differs based on the number of climate shocks experienced. In the case of a single climate shock, migration rates decline as households face liquidity constraints and cannot finance migration. However, when households experience preceding positive shocks in the presence of liquidity constraints, they may relax those constraints and increase out-migration, as they view migration as a risk-management strategy.

In summary, the literature on climate-induced migration suggests that the relationship between climate change and migration is complex and context-dependent. The impact of climate shocks on migration may vary depending on the number and intensity of shocks experienced and various individual, household, and socioeconomic factors. Kleemans' (2015) findings emphasize that the effects of climate shocks on migration are not linear and depend on factors such as risk, liquidity constraints, and the frequency of shocks.

To do this, we include negative (weather-driven) crop yield shock at the state-of-origin level. Following studies on climate-induced migration, we use negative changes in corn crop yields (NCYrt) as an instrumental variable to predict people's migration responses. We calculate negative changes in corn yield (negative shocks) as yields below one standard deviation from the state's mean. These negative changes in corn yields (at the state level) are a good instrument because they influence migration outflows (Feng et al. 2010) without correlating with non-agricultural wages in urban areas.Footnote 14 We assume that shocks at the state level do not affect food prices because corn is an internationally traded commodity.Footnote 15 Specifically, we use corn yield shocks, which have been shown to influence migration (Hunter et al. 2013; Nawrotzki et al. 2013; Feng and Oppenheimer 2012; Feng et al. 2010) yet are unlikely to affect wages in manufacturing retail, or service sectors in urban areas except through labor supply.

We also control for household characteristics and characteristics of the source and destination municipalities (i.e., # of people, households with electricity, water, and drainage). We created a pooled cross section of individuals in all municipalities over three years (1990, 2000, and 2010).

3.2 Second stage

In the second stage, following Nicita (2009), we estimate a wage function based on individual data as a function of trade-related, demographic, and household characteristics and the instrumented probability of migration for individual i. The complete wage function is Eq. (2). Like Nicita (2009), we include control variables such as age, years of education, gender of the worker, and whether s/he is a business owner. We ran the regression for separate segments of the wage distribution to analyze trade openness's effect on regional wage differentials. We define regional wage differentials by analyzing the lower and higher segments of the wage distribution. We define the segment of low-wage earners by separating those earning less than one standard deviation below the mean wage each year. In the same way, the high-wage earners' segment is defined as those earning more than one standard deviation above the mean wage for each year.

To determine regional wage differentials throughout Mexico, we use individual-level wages, individual and household characteristics, and regional-level data regarding economic growth, education, migration, and other characteristics. Finally, the estimated probability of migrating caused by trade openness and other variables, P(\(\widehat{M}\) it), is included in the regression.

This paper asks whether NAFTA increased wage differentials between high and low wages and regions once internal migration is considered. Standard Heckscher–Ohlin trade theory predicts that trade would decrease the premium to high-skilled labor in Mexico; because Mexico is abundant in low-wage earners, it will export those goods that use low-skilled labor intensively while importing goods that intensively use high-skilled labor, driving up the relative wages for low-skilled workers and decreasing the relative premium for high-skilled workers (Stolper and Samuelson 1941). One assumption embedded in this result is that workers are freely mobile between sectors and regions. New Economic Geography (NEG), for its part, argues that companies will move where low-wage workers are available, increasing wage differentials across space (Krugman and Venables 1995).Footnote 16 Combining the NEG and the standard trade theory, we obtain the following testable hypotheses:

  1. 1.

    Over the past decades, trade openness has caused a substantial decrease in wage differentials in Mexico, with changes in tariffs decreasing wages for high-wage workers and increasing them for low-wage earners.Footnote 17

  2. 2.

    However, trade openness has differential effects regionally. Trade-related wage increases have gone primarily to workers near the US border, increasing wage differentials across space (Nicita 2009).

  3. 3.

    Large wage differentials will exist for those locations where workers are less mobile (Morrison et al. 2007; Finan et al. 2005).

To study the effect of NAFTA on migration, we first predict the probability that an individual will migrate based on the potential growth in municipal Gross Value Added (GVA) associated with tariff reductions from NAFTA. Because migration and wage outcomes are jointly determined, likely both related to unobservable individual characteristics, we instrument for migration using local crop yield shocks, specifically corn, which have been shown to influence migration (Hunter et al. 2013; Nawrotzki et al. 2013; Feng and Oppenheimer 2012; Feng et al. 2010) yet are unlikely to affect wages in the manufacturing, retail, or service sectors in urban areas except through labor supply: A crop yield shock would influence labor migration out of agriculture into other sectors,Footnote 18 but is unlikely to directly affect demand for manufacturing, retail or services in urban areas, which supply products to a national and international market as farmers only account for 0.6% of the labor force, and comprise a small share of demand for retail and services (INEGI 2009). We test this assumption by observing the labor demand effect to be larger in smaller, more rural municipalities (which are not the main recipients of migrants) and stronger in the service sector (Arends-Kuenning et al. 2019; Baylis et al. 2012; Luckstead et al. 2012). Next, in the second stage, we estimate a wage equation as a function of trade, demographic and household characteristics, and the previously instrumented migration probability. By analyzing trade openness and distance to the border, we find that workers closer to the US–Mexico border get a higher wage than their far-off counterparts. Further, we find that all workers get a reduction in their wages from NAFTA. However, low-wage men get a lower reduction in their wages than those with high wages. Thus, trade liberalization has decreased wage differentials.

4 Data

We use the 1990, 2000, and 2010 micro-samples of the Population Census, collected by the National Institute of Statistics and Geography (INEGI), which provides household-level data on the Mexican population. These data create cross-sections across time that span the introduction of NAFTA.

Most migrants come from the Southern states of Guerrero, Oaxaca, Veracruz, Puebla, and Hidalgo (SEDESOL 2004) to recipient states in the north, such as Sinaloa, Sonora, Baja California, and Baja California Sur (see Fig. 2).

Fig. 2
figure 2

Source: CONAPO, with information from INEGI's 2000 Population Census (Vega 2005, p 17)

Net Migration by State, 1995–2000.

The regional distribution of wage differentials before and after NAFTA is shown in Figs. 3 and 4, respectively. Here, we observed the percentage of the employed population earning up to one and more than ten minimum salaries.Footnote 19 The closer this percentage is to zero, the higher the employed population that receives an average wage similar to their counterparts and a lower wage differential. The closer to one, the more the wage difference. In 1990, four years before NAFTA, there were more states with a larger wage difference, especially concentrated in the southern states. The average wage difference in 1990 was 22%, with Chiapas having the largest wage difference of 41% and Baja California with the lowest wage difference of 13%. In 2010, sixteen years after NAFTA, the average wage difference decreased significantly to 13% (similar to the percentage only two states had in 1990), with Chiapas again with the worst wage difference but only 20% and Coahuila with the lowest wage difference of 9%. As observed, the wage differential has decreased considerably since NAFTA. Also, most states with a low wage difference are on the US–Mexico border and in the Bajío region.

Fig. 3
figure 3

Source: Constructed with data from INEGI (2020)

Wage differentials before NAFTA (1990).

Fig. 4
figure 4

Source: Constructed with data from INEGI (2020)

Wage differentials after NAFTA (2010).

The variables used are described below. Summary statistics are provided in Table 1.

Migration (Mit): Migration data come from the 1990, 2000, and 2010 Population Censuses, with a question asking in what state (or municipality) the interviewee resided five years earlier. Though this approach might be standard, these data have the drawback of failing to count migrants who might have left and returned over the five years. In the sample, 6.8%, 5.6%, and 4.4% of the men reported migrating in 1990, 2000, and 2010, respectively.

GVA: We include the Gross Value Added (GVA) measurements for the destination areas in period t − 1. These data were obtained from INEGI's economic censuses in 1989, 1999, and 2009. We use this variable to capture the effect that the potential growth or contraction of economic activity (GVA) will have on migration and wages. We expect that municipalities with a large growth of economic activity will attract more migrants and provide a higher wage (Arends-Kuenning et al. 2019).

% Change in Tariffs \((\Delta {\tau }_{t})\): Here, we explicitly measure the effect of the change in US tariffs on Mexican exports to capture potential trade-driven growth (Aguayo-Téllez et al. 2010). Therefore, to identify NAFTA's effect on migration and wages, this paper uses the different tariffs available for the different sectors. These data were obtained from the United States International Trade Commission (USITC 2014). With an annual frequency, we use the data available of the US tariffs on Mexican exports at the 1-digit Standard Industrial Classification (SIC) level for the light/heavy manufactured, mining, and intermediate goods, which we matched to the manufacturing, mining, and commerce sectors, respectively. These tariffs are aggregated across different goods sectors and weighted by their respective national export trade volumes. This tariff-change measure is the same for the entire country and varies over time.

Transportation cost (distF): We consider that economic growth will be correlated with transportation cost to the US border, which we proxy with the road distance (measured in thousands of kilometers) from the region of destination to the closest US border-crossing point. To create the border distance variable, distF, we first obtain the names of the municipality capitals (INEGI 2008). Second, we calculate the road distance from each municipality capital to the US border crossing points by entering the destination and origin points on the “Traza tu Ruta” web page of the Secretaría de Comunicaciones y Transportes (2008). Finally, we chose the shortest distance for each municipality capital from the different distances provided by each border crossing point. For municipality capitals that do not appear as origin points, we calculate the distance of the nearest available city or town and add the road distance from that point to the municipal capital of interest, which we manually calculate using a map of Mexico. The average distance to the border was about 900 km (Table 1).

Infrastructure (Infrastructure): Investment in infrastructure provided by the local governments plays an important role in migration decisions and wages since people tend to migrate from places with low infrastructure levels to places with high levels of infrastructure (Arends-Kuenning et al. 2019). However, to minimize the noise caused by including all the infrastructure variables in the regression, we include a principal component indexFootnote 20 of the three infrastructure variables (the percentage of households with water, electricity, and sewage from the region where the person lived five years ago). This information was obtained from the INEGI's population censuses (see Table 1).

Population density (Pop. Density): Greenwood (1997) mentions that migration is directly related to the population size of the origin. Thus, we control for the population size from the region where the person lived five years ago because regions with larger populations will have more out-migration (Rupasingha et al. 2015). We use the population density (population per square kilometer) that municipalities and states report, including children and elderly, in every population census.

Unemployment. The level of unemployment plays an important role in the migration decision because higher levels of unemployment trigger migration to places with low levels of unemployment (Arends-Kuenning et al. 2019). Therefore, we use the unemployment rate (for men and women) from the region where the person lived five years ago. The unemployment rate includes unemployment and underemployment. Underemployment is when a worker is underused, either because s/he cannot use all their capabilities or is part-time or idle (McKee-Ryan and Harvey 2011). INEGI's economic censuses provide this information. The unemployment rate (for men and women) was 53, 54, and 51% in 1989, 1999 and 2009, respectively.

Negative crop yields in the last five years in the state-of-origin (NCYrt): Following Feng et al. (2010), we took the corn yield at the state level for all the years (1994–2009). Corn is the most important crop grown in Mexico. Of the total area harvested in Mexico in 2018 (16,118,051 hectares), 44% was dedicated to growing corn. The second-largest crop grown is beans, which are produced on only 10% of the total area harvested.Footnote 21 These data were obtained by the Sistema Estatal y Municipal de Bases de Datos (SIMBAD) from INEGI (2010). We calculate the negative changes in corn yield (negative shocks) as yields below one standard deviation from the mean for that specific state. After, we sum up those states with negative corn yields in the last five years and create a variable with the number of years a state-of-origin experienced negative corn yields (0 to 5). Experiences of negative corn shocks differ, varying from a low of only 0.05 in 2000 to 1.02 in 2010 (Table 1). We map these negative corn shocks for the three periods (1990, 2000, and 2010) to show the regional variation (see Figs. 5, 6, and 7 in the annex).

4.1 Individual characteristics

Age: For this study, we consider only males of working age (18–65 years) because we see a large increase in labor force participation of women from 1990 to 2000, whereas 78 and 80% of men of working age were participating in the labor force in 1990 and 2000, respectively. While the question of how trade affected female labor market participation is interesting, we wanted to focus on the population already likely to be in the labor force. Hanson (2007) and Nicita (2009) also work with the working-age male population due to the same problem. Hanson explains that female participation in the labor force is low and varies considerably across time. He further argues that including women creates a sample selection problem since many women report zero labor earnings but may work in family businesses or on family farms.Footnote 22 We use a quadratic function to capture a potentially nonlinear effect of age. Here, we expect that the older the person, the lower their migration probability, but the higher their wage.Footnote 23 The average age of the men in the sample ranged from 34 in 1990 to 36.5 in 2010 (Table 1).

Indigenous language: There is a large amount of literature on the internal migration of the indigenous population in Mexico, who are constantly searching for a better standard of living.Footnote 24 Therefore, we include the question “Do you speak an indigenous language?” from INEGI's population census to identify the indigenous population, which migrates more than the average population (Asad and Hwang 2019). The percentages of the sample who reported speaking an indigenous language were 4, 4.8, and 4.4% for 1990, 2000, and 2010, respectively.

Working hours: We include the number of hours worked in the week. These data were obtained from INEGI's population censuses. Working hours might reduce migration because people with good jobs will be less likely to migrate. However, working hours might indicate a higher average wage, as it might be a proxy for a more remunerative and secure employment opportunity (Morrison et al. 2007). The average working hours were 46, 49.7, and 49.6 for 1990, 2000, and 2010.

Business owner: We include the percentage of the population that owns a business. This data was obtained from INEGI's population censuses. Business ownership may increase the transaction cost of moving because people who own their business will be less likely to migrate and give up their capital and connections to local markets (Greenwood 1997). In addition, business owners are less likely to migrate because their income is higher than that of those who work for someone else. The percentage of business ownership was 3.2, 3.4, and 3.7% for 1990, 2000, and 2010, respectively.

Education is the stock of productive skills and technical knowledge embodied in labor. Men's educational levels increased in Mexico over time, completing 8.0 years on average in 1990 and 9.8 years on average in 2010 (Table 1). Mexico has a competitive advantage in unskilled labor-intensive goods. Then, the effect of the education variables will be (∂y/∂edu) > 0. That means those with more education will earn higher wages.

5 Results

5.1 1st Stage—probability of migration for working-age male population

In the first stage, we regress a variable indicating migration on drivers associated with trade. Table 2 reports the linear probability model (LPM)Footnote 25 regression results from the first stage of the probability of migration.

5.2 2nd stage

We next regress wages on the trade-induced changes in market access and predicted migration. Overall, the coefficients on the core variables are generally statistically significant and with the predicted signs (Table 3). The first column shows the result of the second-stage regressions of the working-age male population. Columns 2 and 3 show the result only for the working-age population for the low and high wages. Table 4 shows the marginal effects of changing tariffs and distance to the border. Next, we explore each of these results.

5.3 Whole working-age male population

To capture the effect of local trade-induced market access, we use the change in tariff (∆τt), the GVA of the municipality-of-destination, and the distance to the US–Mexico border (distFi), as well as their interactions. Due to the resulting coefficients for each variable and its interactions, the way to know the overall effect of these variables is by obtaining the marginal effect of each of them, which we show in Table 4.

The marginal effect of a change in tariff (∆τt) means that a one-percent decrease in tariffs decreases the percentage of wages by 0.3%, suggesting that if the US market demand for products from a location increases, workers in that location tend to receive a lower wage. This relates to Majid's (2004) finding that trade openness might cause short-run declines in real wage rates.

The marginal effect of GVA means that an increase of 1 million Mexican pesosFootnote 26 in the economic activity of the destination’s economy will generate an increase in the wages of all the working-age men's population by 0.2%. This result means that the more the municipality’s economy grows, the more the wage of all the workers benefits. This result concurs with the NEG model idea that producers get closer to the US–Mexico border to obtain market access to larger US markets (Blankespoor et al. 2017). This increase in economic activity increases labor demand, which as a result, increases wages. Hanson's (1998) findings concur with these results since it found that trade openness benefited by increasing economic activity and wages.

We also find that distance to the border (distFi) plus the interaction of distance with GVA and change in tariff (distFi*∆τt* GVAit−1) have a negative and significant effect on wages (column 1). Overall, the main marginal effect of distanceFootnote 27 is that working a thousand kilometers away from the border decreases the average wage by 19% (see Table 4). This evidence does not reject our second hypothesis that, following Nicita's (2009) findings, the benefits of NAFTA have gone primarily to workers near the US–Mexico border. The impact is larger on workers further away, but it is negative, increasing regional wage differentials.

The estimated probability of migrating, caused by trade openness and other variables, P(\(\widehat{M}\)it) is significant when considering the overall working-age population and tells us that people who migrate receive 4% less wages than those who do not.

5.4 Low vs. high wages

When we divide the data between high- and low-wage men, we find that the marginal effect of a decrease in tariffs due to NAFTA is negative for both groups, high wages and low wages (see Table 4). However, while high wage workers lose 0.9% of their wages for a 1% decrease in tariffs, low-wage workers lose only 0.5%. This result supports the first hypothesis that trade openness has decreased wage differentials because the tariff reductions associated with NAFTA have harmed less low-wage workers than high-wage workers, decreasing the wage differential.

The marginal effect of economic activity (GVA) shows that those regions that experience economic growth (by an increase in 1 million Mexican pesos on the economic activity) will harm low-wage workers since it will reduce their wages by 0.3%, while it will increase the wages by 1.1% to high-wage workers. This might relate to what Robertson (2007) found: that expanding assembly activities in Mexico increases the demand for less-skilled workers who pay lower wages.

Moving to the third hypothesis, we observe that those low-wage workers who migrate, P(\(\widehat{M}\)it), see an increase in their final wages. In contrast, high-wage workers who migrate see a reduction in their wages. Thus, we see two types of migrants: low-skilled workers migrating to occupy better-wage jobs and high-skilled migrant workers who get lower-paid jobs. The coefficients on P(\(\widehat{M}\)it) represent the Local Average Treatment Effect, which is the effect on people whose migration behavior was affected by the negative crop yields. High-wage workers who are less likely to migrate (compared to low-wage workers) due to negative crop shocks find lower-paying jobs when they migrate, and vice versa for low-wage workers. This evidence suggests that migration allows low-wage men to end up in higher-paying jobs than those who do not migrate, on average, over the whole country. This result supports the findings of Morrison et al. (2007) that poor households prevent and mitigate risk by migrating to locations with more remunerative and secure employment opportunities.

6 Conclusions

Considering internal migration, this paper explores the factors that influence Mexico's regional wage differentials and the effect of NAFTA. We use individual-level wages, individual and household characteristics, and regional-level data regarding economic growth, education, migration, and other characteristics, to determine regional wage differentials throughout each Mexican region. Thus, this study sheds light on how trade openness affects individual and wage differentials.

This research provides initial evidence of the mechanism of internal migration to reduce wage differentials, suggesting that trade liberalization has reduced wage differentials, leading to a decrease in regional polarization, partly due to internal migration. Apart from the slight reduction in wages due to trade openness, men with lower wages benefited more from NAFTA than those with higher wages, indicating a decrease in the wage differential. The potential effect of NAFTA on migration is also stronger for high-wage men than for low-wage men because low-wage workers cannot afford the minimum amount of money necessary to migrate. Also, the region's economic activity induced migration but offered a lower wage to low-income workers. This increases the wage differential because the increase in economic activity due to trade openness has only benefited high-income workers but not those low-income workers.

The effects of trade liberalization, such as regional Government investment in transportation (to reduce transportation costs), have slightly increased migration toward the US–Mexico border (Arends-Kuenning et al. 2019). This result conforms with earlier evidence by Krugman and Livas-Elizondo (1996), who find that trade leads to more migration because the US market is increasingly important.

While workers near the US market earn a higher wage, workers far from the USA receive a lower wage.Footnote 28 This spread increases regional wage differentials. However, north–south disparities are only one part of the story. Large manufacturing and service sectors are associated with lower wages.

This study's potential policy implications are that investment in economic activities different from those near the US–Mexico border (such as commerce and mining sectors) can be used to ease regional wage inequality. This evidence suggests that internal migration policies encourage economic growth and reduce wage differentials. Instead of deterring mechanisms of labor adjustment, governments should foster them to reduce regional disparities by providing credit and subsidizing migration (Kleemans 2015), for example. Concerns about preventing internal migration should be replaced by arguments that internal migration can reduce regional disparities. This finding should contribute to the current debates about trade and migration in Mexico and the USA. The concerns should focus on improving people's welfare and encouraging regional development rather than stopping internal (and external) migration. However, it is important to mention that those policies should have broad access to ensure they reach all households and regions. This will avoid increasing inequality among households and regions.