Why Do Migrant Workers from Poorer Countries Work for Less? A Purchasing Power-Based Explanation

Why Do Migrant Workers from Poorer Countries Work for Less? A Purchasing Power-Based Explanation Hyejin Ku Department of Economics Florida State University October 12, 2009 Abstract A well-known empirical regularity in economics is that when all countries price levels are translated to dollars at the prevailing nominal exchange rates, rich countries tend to have higher price levels than poor countries (the Balassa-Samuelson e ect). This paper incorporates this phenomenon into a model of international migration in which remittances allow cross-border optimization of consumption. It argues that to the extent that the real price level in the source country is lower than that in the host country and that the migrants spend part of their income in the source country, the observed labor market outcomes of migrants in the host country may look worse than expected based on real wage di erentials alone. Empirical evidence based on the labor market experiences of Mexican workers in the U.S. lends support to this theory. Key words: Balassa-Samuelson, international migration, remittances, purchasing power parity, cost of living, real exchange rate, CPI JEL classi cation codes: F22, F24, O11, O15, O24 I am grateful to Kaushik Basu, Nancy Chau, Steve Coate, Stefan Klonner, Tim Salmon, Emma Stephens, and Asaf Zussman for valuable discussions and comments. Please direct any comments or suggestions to hku@fsu.edu. 1

1 Introduction We often observe immigrant workers from poor countries working in dirty, dangerous, and di cult (3D) jobs in a rich host country. The question that this study addresses is why immigrants from low income countries earn lower wages than equally quali ed natives or immigrants from higher income countries. A common explanation for this is that even if the real wages for these jobs may be low according to the standards of the host country, they may still be an improvement over what the migrants would have made by staying in the source country, at least in expectation (Harris and Todaro, 1970). In addition, if migration is viewed from a dynamic standpoint, even a negative premium in the current real wage may be rationalized as long as the expected stream of income from the host country exceeds that of the source country (Sjaastad, 1962). Implicit in these explanations is that in order to justify migration, the worker should experience some sort of premium in her real wage or standards of living in either a static or dynamic sense. This paper complements these existing explanations based on geographic real wage differentials by providing a new explanation based on real price di erences between markets. Speci cally, it argues that to the extent that the real price level in the source country is lower than that in the host country and that the migrants spend part of their income in the source country, the observed labor market outcomes of migrants in the host country may look worse than expected based on real wage di erentials alone. My argument hinges on the following stylized facts. First, when all countries price levels are translated to dollars at the prevailing nominal exchange rates, rich countries tend to have higher price levels than poor countries (Balassa, 1964; Samuelson, 1964). Figure 1 shows the positive correlation between the real price level and per capita income in 136 countries. If the purchasing power parity (PPP) holds true, all countries in this diagram should have a price ratio of unity. 1 A similar picture emerges when we compare the costs of living across the globe. As of 2008, the same basket of goods and services, including rent, costs 120.2 in London, 82.4 in Frankfurt, 55.9 in Prague, 49.1 in Mexico City, and 43.3 in Manila, when the price level of New York is normalized at 100 1 Rogo (1996), Goldberg and Knetter (1997), and Taylor and Taylor (2004) provide a comprehensive review on the issue of PPP. 2

(Union Bank of Switzerland, 2008). This implies that a U.S. dollar, if converted to pesos at the market exchange rate and spent in Mexico City, can buy twice as much as in New York City. Second, migrants can optimize their consumption mix across borders via remittances or repatriated savings. The prevalence of temporary migration and the ease of separating between the location of earnings and the location of consumption via remittances are, arguably, some of the key characteristics of today s migration phenomenon. 2 By linking the Balassa-Samuelson e ect to the possibility of cross-border price arbitrage via remittances, this paper argues that the greater the real price gap between the host and the source countries, the lower the real wage in the host country that can trigger migration is. As Figure 1 shows, since poorer countries are where the real price level is also lower, we can infer that holding all else constant (including skill levels and foregone real wages), a given real wage in the host country is more likely to meet the reservation real wage of migrants from poorer countries than that of immigrants from other countries. This paper by no means claims that the Balassa-Samuelson e ect, or the persistent real price gap between rich and poor countries, is the predominant force behind the poor labor market outcomes of migrant workers. Rather, it suggests that existing explanations in the literature might be underestimating the pressure for migration from poor to rich countries in the sense that the real price gap can in and of itself alter the incentive for migration independently from the e ects of other factors including the real wage gap. Related to this is the implication that holding all else constant, if the real price gap between the host and source countries narrowed exogenously, the partial e ect from the purchasing power channel proposed in this paper would include lowered migration pressure and fewer migrants willing to work in low-paying jobs in the destination. This paper starts by presenting a simple model of international migration with remittances. Though I use the term, remittances, for expositional simplicity, remittances here really refer to any device including repatriated savings that allows migrants to spend their 2 The dramatic growth of migrants remittances in the past few decades demonstrates that remittances have become an integral part of the migration experience for today s migrants. In 2004 alone, recorded remittances received by developing countries amounted to 160 billion dollars, which is equivalent to about 6.7 percent of developing countries imports and 7.5 percent of domestic investment (World Bank, 2006). 3

foreign income in their countries of origin. The model is in line with the new economics of migration (Stark and Bloom, 1985; Stark, 1991), in that migration is viewed as the outcome of the collective decisions made by families or household members to overcome a variety of market failures, apart from the geographic disparity in the return on labor. 3 The main feature of my model is that I allow for the possibility of real price di erences between the host and source countries and explicitly model the role of remittances as a device for crosscountry optimization of consumption mix. 4 A testable implication of my model is that to the extent that the real price level in the source country is lower than that in the host country, the real wage in the host country which justi es migration becomes lower, and this is true for the migrants who expend a positive sum of their foreign income in their home country, but not for others. 5 Using the migration experiences of Mexican workers in the United States as an example, I present empirical evidence that is consistent with the theoretical argument of this paper. The data come from the Mexican Migration Project (MMP), a survey used in many studies such as Massey et al. (1994), Durand et al. (1996), Munshi (2003), Orrenius and Zavodny (2005), and McKenzie and Rapoport (2007). By treating di erent U.S. cities as separate host countries and exploiting the di erential change in the consumer price index (CPI) over time in those cities, I obtain variation in real price di erences between the host and source countries. Then, I compare the real wage received by migrants who send remittances or bring savings to Mexico with that received by migrants who engage in neither activity. The main empirical nding corroborates the theory. Speci cally, it shows that as the real price level in the host country relative to that in the source country rises, the change in the real wage of migrants who engage in both remittances and repatriated savings is 9:03 to 15:84 percentage points less than that of migrants who engage in neither activity. Several theoretical studies have previously addressed the issue of geographic real price 3 A prominent example is the risk diversi cation through migration among family members (Rosenzweig and Stark, 1989). 4 Hoddinot (1994) also considers a model in which utility maximization through remittances is embedded in the migration decision. However, the focus of that study was not on the issue of price asymmetry between regions, while the present study s issue is the cross-border price arbitrage via remittances. 5 In the case of a migrant who consumes 100 percent of her income in the host country, whether the currency of the host country has extra purchasing power when spent in the source country or not, this will be irrelevant. 4

di erences and the decision to migrate (Djajic, 1989) or return-migrate (Dustmann, 1995, 1997, and 2003; Stark and Fan, 2007; Stark et al., 1997). On the empirical front, Massey and Espinosa (1997) and Yang (2006) investigate the role of price di erentials between the host and the source countries in the decision to migrate or return-migrate. Other empirical studies also discuss the issue of real price gap while analyzing other aspects of migration. For instance, in studying the illegal immigration of Mexican workers using macro-level data, Hanson and Spilimbergo (1999) nd that a U.S. wage, expressed in its purchasing power in Mexico, is positively associated with the level of apprehension, a proxy for the illegal attempts at migration. My major point of departure from these studies is that I examine the e ects of real price di erences (i.e., real exchange rates) between countries on the labor market outcomes of migrants in the host country, in particular, real wages. In an earlier study, Fox and Stark (1987) established a positive correlation between the hours worked by Mexican workers in the United States and the depreciation of the peso against the dollar. However, since the authors directly employed the U.S.-Mexico exchange rate in their empirical analysis, it is di cult, based on that study, to disentangle the e ects of real price di erences from those of other macro-variables shifting simultaneously. In a separate study, Ku (2008) looks at the di erential responses to uctuations in the U.S.-Mexico real exchange rate for Mexican migrants who are high-remitting types (the treatment group) and those who are lowremitting types (the control group) and nds that the former respond more strongly to a given real exchange rate shock in terms of their labor market outcomes in the United States. While consistent with the theory of price arbitrage, the nding in Ku (2008) is still subject to the criticism above to the extent that there might be omitted covariates of real exchange rates. The approach of the present paper eliminates such problems by using the CPI of di erent U.S. cities (as separate destination countries ) to obtain variation in the real exchange rate between the host and source countries. In my analysis, all the macrolevel shocks a ecting the overall economies of the United States and Mexico are controlled for throughout by the inclusion of city-year xed e ects. In addition, the main outcome of interest in the present study is the real wage that migrants receive as opposed to the labor supply as in Fox and Stark (1987). 5

The rest of the paper is organized as follows. Section 2 presents a model of migration to illustrate environments in which geographic real price di erences matter. Section 3 describes the data. Section 4 presents the empirical results. Section 5 o ers some concluding remarks. 2 Conceptual Framework Consider a model in which there is a source country (henceforth South) and a host country (henceforth North). The price level and nominal wage o ered in the South are p and W respectively, while the North o ers p and W. The real wage in each country is then W=p w and W =p w. These countries have their own currencies, say, peso in the South and dollar in the North. The exchange rate between the currencies is denoted by e where one dollar exchanges with e pesos. The real exchange rate then is ep =p Q, which may or may not be equal to unity. Incorporating the empirically observed discrepancy between the market exchange rate and PPP is my key point of departure from the literature, where price parity between regions is typically assumed in most models. If PPP holds, Q = 1. If the currency of the North has extra purchasing power in the South, then Q > 1. If the costs of living are lower in the North, then Q < 1. Each household consists of a worker or potential migrant (1) and a dependent who is sedentary (2). The household is assumed to maximize the family utility function, U(C 1 ; C 2 ), in the sense of the unitary model (Becker, 1981). The unitary model of the household encompasses several di erent models of family structure in which a family in aggregate behaves as if it is maximizing a family utility function. 6 Alternatively, we may think of U(C 1 ; C 2 ) as the individual utility function of the worker or potential migrant who cares about her dependent s consumption for altruistic reasons. 7 Speci cally, U = C1 1 C2 ; where C 1 and C 2 refer to the consumption levels of the worker and the dependent, respec- 6 See Bergstrom (1997) for a survey of theories of the family, including the unitary approach. 7 Rapoport and Docquier (2006) provide an extensive review of the motivations for intra-household transfers in the context of migrants remittances. 6

tively. Households di er in one dimension: the weights assigned to the consumption of the worker and the dependent. Without loss of generality, let indicate each household s preference weight on C 2. I normalize the utility function by multiplying 1= (1 ) 1 to it. This makes the indirect utility of di erent households comparable while maintaining the heterogeneity in the intrahousehold allocation of resources between households. To focus on the main research question, it is assumed that the migrant always works and the spouse always makes zero earnings. For the same reasons, costs involved in migration are suppressed. 8 In the South, a household maximizes U subject to the budget constraint C 1 + C 2 w. The solution to this problem is C 1 = (1 )w and C 2 = w. Hence, the maximized value of utility in the South, which I denote by U S, equals w. If the worker migrates to the North, the household maximizes U with location-speci c price indices in the budget constraints since C 2, the consumption for the dependent, must be purchased in the South. Therefore, the appropriate constraints become: C 1 + R w ; C 2 QR; where R is the remittances sent back to the South. The migrant spends part of her wage, w, on her own consumption in the North and sends the rest to her dependent in the South. The real exchange rate, Q; determines how R is translated into the level of consumption in the South. The maximized value of utility in the event of migration is U N = Q w. 9 The condition that makes the worker s migration to the North worthwhile is then Q w w: For a migrant who is indi erent between migration and no-migration, Q w = w or ln w = ln Q + ln w: (1) 8 For discussions on moving costs, see e.g. Carrington et al.(1996) and Chau (1997). 9 The solution vector is (C 1 ; C 2 ; R) = ((1 )w ; w Q; w ). 7

Equation (1) above highlights the key di erence between my model and standard models of migration. In standard models in which purchasing power parity (PPP) between regions is implicitly assumed, the indi erence condition looks similar to the following: ln w = ln w; (2) with some accounting for the costs of migration, of course. Notice that (2) is a special case of (1) in which PPP happens to hold or the real exchange rate Q is set to unity. This makes it clear that the purchasing power-based argument proposed in this paper is not contradictory to the existing explanations based on real wage di erentials and rather, it points out an additional factor that may in uence w, the real wage in the host country that makes a migrant indi erent between migration and no migration. Going back to equation (1), the condition suggests a negative relationship between the real wage of the host country, w, and the real price level of the host country relative to that in the source country, Q, holding all else constant. Notice that this relationship holds true only if the individual allocates a positive sum of her foreign income in her home country (i.e. > 0). For an individual who spends 100 percent of her foreign income in the host country (i.e. = 0), the extra purchasing power that the Northern currency has in the South does not have any impact on her labor market decisions in the North. This is the key prediction of the model that I will rely on for my empirical analysis. For a migrant s reservation real wage to have an impact on her actual labor market outcome, we require some frictions in the labor market for migrant workers. Otherwise, workers would be sorted and compensated according to their traits related to their productivity only, leaving no room for the role of reservation wages. However, evidence from the U.S. seems to point to the existence of frictions in the labor market for migrant workers. For instance, Munshi (2003) and Beaman (2007) demonstrate the role of social networks in channeling migrants into high- vs. low-paying jobs in the U.S. If the labor market for migrant workers were frictionless, the size of social networks wouldn t have had any power in explaining an individual s occupational outcome once her observable characteristics are controlled for. In the presence of frictions in the labor market, it then follows from standard search models that migrants with a low reservation real wage may indeed be receiving a lower real wage on 8

average than what is received by those with a high reservation wage since a given real wage is more likely to meet the reservation real wage of the former than the latter. 3 Data My empirical analysis is based on the migration experiences of Mexican workers during the period 1967-2003. The data come from the Mexican Migration Project (MMP), a survey commonly used in the literature. 10 The MMP started in 1982, and each year during the winter months (when seasonal migrants are home) around 200 randomly selected households are surveyed in each community chosen for the survey. Each community is surveyed only once but the heads of household are asked their entire life retrospective migration histories. For my analysis, I use only information that relies on a recall length of 15 years or fewer, assuming that people can recall events that occurred up to 15 years before with a reasonable degree of accuracy. For instance, if the head of a household was interviewed in 1995, his information pertaining to the periods between 1980 and 1995 would be used for the analysis, while information for the years prior to 1980 would be discarded. 11 The MMP is an on-going project and I use the MMP107 database, which contains information on 107 communities. My sample is restricted to the working age (16-64) individuals and information on their last migration trip to the U.S. because information on remittances and savings behavior as well as U.S. earnings is only provided for these migrants. Also, I restrict my analysis to those who migrated to a U.S. place (as de ned by the MMP) or city, that had at least 25 migrants to avoid undue in uences from outliers. After dropping observations that lack CPI data for the destination U.S. city or any relevant information such as age, level of educational attainment, wage, and remittances and savings, I am left with 1120 observations. Appendix A shows the 12 U.S. cities in which these migrants are located. Summary statistics for the migrant workers are provided in Table 1. According to the rst column, the typical Mexican migrant in my sample is 33 years of age, nished about 6 years of 10 The MMP is a collaborative research project based in Princeton University and the University of Guadalajara, and the dataset is publicly available (http://mmp.opr.princeton.edu/). 11 Later, I test whether the results are robust to changes in the length of the recall period. 9

schooling, is not uent in English, and is undocumented. Also, the average migrant is likely to be married with two children under 18. The next three columns provide information on migrants according to their remittances and savings status: migrants who send remittances, migrants who return with savings, and migrants who engage in both activities. No large di erence is discernible in the characteristics of migrants across the categories. Table 2 provides more details on the remittances and savings situation of the migrants. Panel A shows that 77 percent of them send remittances, 53 percent of them return with savings, and 42 percent of them engage in both activities. The average amount of remittances and savings seems to be quite similar, but the stated purposes for each activity di er. In the case of remittances, health expenses in combination with food and maintenance, account for over 75 percent of the responses. This is consistent with the fact that remittances, by de nition, are meant to support the dependents left behind in the home country. In contrast, the purposes of bringing savings home are more varied and they seem to be tied less to supporting dependents and more to bene ting the migrants themselves upon return. The U.S. Bureau of Labor Statistics provides information on the consumer price index (CPI) of nine U.S. metropolitan statistical areas (MSAs). The CPI is comparable over time for an individual MSA but not comparable across MSAs at a given point in time. This does not undermine my analysis, however, since my empirical strategy relies on the di erential change in CPI over time in separate cities. Figure 2 shows the trends in CPI in di erent MSAs in the U.S. According to the gure, the trend in CPI is not parallel across di erent cities, which provides the source of variation in prices for my analysis. In some of the speci cations, I use additional macroeconomic controls. The variables include the state-level GDP and minimum wage data. The GDP data come from the Bureau of Economic Analysis in the U.S. Department of Commerce and the minimum wage data are obtained from the U.S. Department of Labor. Appendix A links each city in the MMP sample to the corresponding MSA and state. 10

4 Empirical Framework The testable implication of the theoretical model is that the greater the purchasing power of the host country currency when spent in the source country (i.e. the higher the real price level in the host country is relative to the source country), the lower the real wage that a migrant receives conditional upon being in the host country. In addition, this response should be observed among migrants who expend part of their foreign income in the source country, but not among those who spend all of their income in the host country. My empirical strategy is closely tied to these predictions and it proceeds by treating Mexico as a single source country and di erent U.S. cities as separate host countries. Consider the following regression: y ijt = c + jt + D i + D i P jt + (D i Z jt ) + X ijt + " ijt ; (3) where y ijt is the real wage (in logarithm) of a Mexican migrant i residing in U.S. city j in year t, c is constant, jt is city-speci c year e ects, and D i is a variable indicating whether the migrant sent remittances to Mexico or brought savings upon return. 12 P jt denotes the CPI (in logarithm or in level depending on the speci cation) of U.S. city j in year t. Z jt is a vector of time-varying economic variables relevant for the city, including the state-level real GDP (in logarithm) and real minimum wage. 13 The vector X ijt indicates an extensive list of the characteristics of individual migrants that may a ect their labor market outcomes either in Mexico or the U.S. Speci cally, it includes age, age squared, education, education squared, English pro ciency, legal status for U.S. migration (documented vs. undocumented), gender, marital status, number of children under 18, ownership of land, property and business, and the community of origin xed e ects. I cluster the standard errors by destination city*community of origin level to account for potential persistence in the wages of migrants from the same communities in Mexico due, for example, to the bene cial roles of social networks (Munshi, 2003). Since all migrants source country is the same, Mexico, the changes in CPI of individual 12 The real wage, y ijt, is constructed by dividing the nominal wage (in dollars) by the price level of the U.S. city in the relevant year. 13 Ideally, I would have liked to use the GDP data at the MSA level but for the time period covered in the present study, the state is the lowest level for which GDP data are available. 11

U.S. cities over time provide the source of variation in the real exchange rate (Q in the theoretical model) between the host and source countries. The city-year xed e ects, jt ; will absorb all the e ects of economic uctuations or policy changes in the destination city level, including CPI. Note that the e ects of year-to-year changes in the overall economies of Mexico and the U.S. are subsumed in jt. Therefore, it will be redundant to also include year xed e ects in the regression in addition to the city-year xed e ects. 14 The main coe cient of interest is and its sign is expected to be negative. That is, to the extent that the real price level in the host country is higher than that in the source country, a migrant who spends at least part of her foreign income in the source country tends to be associated with a lower real wage abroad than one who doesn t. To make sure that is not driven by covariates of CPI in the destination city, the regression also includes the interaction between D i and Z jt, a vector of other economic variables of city j in year t. Note that the identi cation of does not rely on the comparison of migrants between di erent U.S. cities: jt will account for the baseline di erences between migrants who are located in di erent cities. In addition, will account for the baseline di erences between migrants who send remittances or bring savings home, and those who engage in neither activity. As should be clear from equation (3), my empirical strategy does not rely on the actual real exchange rate between the U.S. and Mexico at the macro-level. The real exchange rate is correlated with many covariates potentially a ecting Mexico-U.S. migration and would complicate inference. Instead, individual U.S. cities serve as di erent destination countries here. The main advantage of this approach is that macro-level uctuations that a ect the overall economy of Mexico or the U.S. are controlled for in the analysis. There are, however, two caveats to this approach. First, this approach requires an assumption about a migrant s behavior: that whether to remit anything or none to the source country is connected to the migration decision and is decided prior to seeing one s foreign wage precisely. In the context of Mexico-U.S. migration, such assumption seems reasonable when the question is whether 14 Denote the real exchange rate between a U.S. city j and Mexico in year t by Q jt. Then Q jt = e t p jt =p t, where e t is the nominal exchange rate between the U.S. and Mexico, p jt is the price level in U.S. city j, and p t is the price level in Mexico. Essentially, I obtain variation in Q jt by varying p jt while holding e t and p t constant. The communities of origin may, of course, be di erent for migrants who are in the same U.S. city. However, unless the CPI of that city is tied long term to particular communities in Mexico, treating Mexico as a single source country will not unduly in uence the outcome. 12

to send any remittances at all or not while the amount sent (conditional on remittances) may depend on the realized wage in the U.S. Second, the estimate of will be subject to bias due to potential interaction between D i and P jt. For instance, a migrant who is determined to send remittances may choose a city-year in which P jt is relatively high. If such interaction between D i and P jt is idiosyncratic, this would not pose a threat to the current approach. In contrast, if the interaction is systematic (i.e. correlated with worker characteristics that may in uence their wages but are not included in the regression), then the estimate of will be biased due to omitted variables. I address this issue by including in the regression a rather extensive list of human capital, demographic, and wealth status of migrants that may directly in uence their wages. 4.1 Baseline Results Table 3 presents the estimates of equation (3). Column (1) shows that the coe cient of the interaction between remittances or savings status and the price level, is indeed negative and statistically signi cant. This is in contrast to the direct e ect of sending any remittances or savings,, which, if anything, is positive. Columns (2) through (4) show that the estimate of remains statistically signi cant and its magnitude becomes slightly larger as more control variables are included in the regression. Column (5) also includes the controls for the migrants communities of origin in Mexico. According to column (5), as the CPI in the destination goes up one standard deviation (32:32) above the mean (122:4), the change in the real wage for a migrant who sends remittances or bring savings back is 17 percentage points below that for those who engage in neither activities. 15 This implies that to a given change in the relative purchasing power of the currency of the host country, the real wage response of the migrants who spend at least part of their foreign income in the home country is over and above that of others. This is a nding consistent with the main theoretical prediction discussed earlier. Next, I test whether or not the results in Table 3 are robust to withstand the inclusion of some covariates of CPI. The set of control variables includes the real GDP and the real 15 ( 0:649) (32:32=122:4) = 0:17 13

minimum wage. 16 The results are presented in Table 4. As columns (1) through (4) show, the estimate of goes down slightly when the real GDP and the real minimum wage are included. However, column (4) shows that the estimate of is still negative and statistically signi cant at one percent level. The estimate in column (4) shows that as the CPI in the destination goes up one standard deviation (32:32) above the mean (122:4), a migrant who sends remittances or bring savings back is associated with a change in the real wage that is 13 percentage points below what is experienced by those who engage in neither activity. 17. My next question is on the channels through which the observed wage-compromise occurs: do the migrants in the non-zero remittances or savings group receive a lower real wage because they are sorted into lower paying jobs and if so, is there a wage-compromise e ect even within the same occupations? To answer this question, I repeat the regressions in columns (1) through (4) in Table 4 but this time with occupation- xed e ects. 18 The estimates are reported in columns (5) through (8) of Table 4. As the estimates in column (8) show, once the occupation is controlled, the coe cient is no longer statistically signi cant. This implies that the wage-compromise e ect that we observed earlier in column (4) occurs mostly through sorting into low vs. high paying jobs, and the real wage received by workers in the non-zero remittances or savings group is no less than others once their occupations are controlled for. Next, I conduct several robustness checks, which are similar to those discussed in Munshi (2003). The results are provided in Table 5. Column (1) is based on all migrants in the sample. In columns (2) and (3), I experiment with di erent sample periods discarding the survey year and the years just prior to that year. This is to address the concern that some migrants who were particularly successful or unsuccessful in the U.S. labor market might not have returned to Mexico at the time when the MMP survey was conducted. As we see in columns (2) and (3), the coe cient virtually remains the same with these sampling restrictions. Next, I turn to the issue of recall bias. Due to the retrospective nature of the MMP data, bias due to a measurement error is a potential cause for concern. As a 16 I would have liked to include unemployment rates as well. Unfortunately, the data on unemployment rate at the MSA-level were not available for most of the time period covered in my sample. 17 ( 0:486) (32:32=122:4) = 0:13 18 I use the classi cation of occupations as provided by the MMP. 14

robustness check, I experiment with a longer length of recall. The rationale for this is that we would expect any reporting errors to be more frequent the further back we go into past. Column (4) shows that in the extended twenty-year sample, the estimate of goes down in magnitude but it is still statistically signi cant at the conventional level. Column (5) shows that the main results are not driven by the small number of female workers in the sample. In column (6), I restrict the sample to undocumented migrants only. The coe cient is smaller in magnitude than in column (1) and it is no longer statistically signi cant. This suggests that it is the documented migrants, if any, who respond more strongly to the opportunities price arbitrage. This seems to have to do with the fact that undocumented migrants are sorted into the lowest paying jobs anyway, and thus there is little room for further compromisation in wages. Column (7) shows that the main results remain the same when the sample is restricted to workers less than 45 years of age. In column (8), I restrict the sample to individuals with less than ten years of education. The coe cient is no longer statistically signi cant in this subsample. As for the case of undocumented migrants, this may be attributed to the fact that less educated workers tend to be sorted into low-paying jobs regardless of the real price level. Lastly, I turn to the issue of functional forms. In column (9), I report the estimates of equation (3) in which Log(CP I) is replaced with CP I in level. The estimate is similar to that in column (1) when evaluated at the mean value of CPI. 19 4.2 Di erential Responses by Remittances and Savings Status So far I have assumed that the e ects of purchasing power di erences on the real wage are common for migrants who send remittances and those who bring savings upon return. From an analytical standpoint, remittances and savings are equivalent in that they both entail spending a fraction of a migrant s foreign income in the source country. However, it is an empirical question whether the two activities would actually have the same impact on inducing migrants to internalize the opportunities for geographic price arbitrage: as 19 Denote the coe cient on D Log(CP I) by and that on D CP I by e. Then, e = =CP I. According to column (1), = 0:486 and the mean of CPI is 122:4. Then, =CP I = ( 0:486)=122:4 = 0:00397 when evaluated at the mean. The estimate of e in column (9) is 0:00425, which is not widely di erent from 0:00397. 15

we saw earlier in Table 2, the stated purposes of sending remittances and savings look rather di erent, from which we can infer that their usages are also likely to di er. To examine if remitters and savers would behave di erently, I classify the migrants into four groups depending on their remittances and savings status: neither send remittances nor bring savings upon return (group 0); send remittances but do not bring savings (group 1, denoted by D 1 ); bring savings but do not send remittances (group 2, D 2 ); and send remittances and bring savings (group 3, D 3 ). Then, using group 0 as a benchmark, I test if groups 1, 2, and 3 have di erent labor market outcomes than the base group. In a regression framework, this corresponds to the following y ijt = c + jt + X k k D k i + X k k D k i P jt + X k (D k i Z jt ) k + X ijt + " ijt : (4) which is identical to (3) except that D i is now replaced with Di 1, Di 2, and Di 3. The coe cients of interest are then 1, 2, and 3. The estimates of (4) are presented in Table 6. According to column (1), the baseline e ects of belonging to groups 1 through 3 are positive, if anything. In contrast, the coe cients for the interaction between Di k and P jt are negative for all three groups. However, as columns (2) through (4) show, the estimates remain statistically signi cant only for groups 1 and 3 when the interaction between group status and other macroeconomic variables is controlled for. According to column (4), as the real price level in the destination country rises one standard deviation (32:32) above the mean (122:4), migrants who send remittances but do not bring savings are associated with a change in the real wage that is 10:4 percentage points what is experienced by the baseline group. The corresponding di erence for migrants who engage in both remittances and repatriated savings is 15:8 percentage points. However, for those who engage in savings but not remittances, their real wage is no more (or less) responsive than that for the baseline to a change in the relative real prices. Next, I examine sorting into occupations as a potential channel through which these results are reached. To this end, I repeat the regressions in columns (1) through (4) while including occupation xed e ects. The estimates are reported in columns (5) through (8). In column (8), we see that none of 1, 2, and 3 are statistically signi cant this time. This suggests that the negative e ects in column (4) have mostly to do with the 16

sorting of migrants into low- vs. high-paying jobs depending on types of migrants. Table 7 presents the results of several robustness checks similar to Table 5. Depending on sample restrictions, the estimates of 1, 2, and 3 seem to vary. However, the estimate of 3 is quite robust across sampling restrictions. This is intuitive in that the migrants who engage in both remittances and savings activities are the ones mostly likely to assess their foreign wage in the basket of the source country. Therefore, when the opportunity for price arbitrage arises, these migrants would be most willing to work in low-paying jobs, in comparison to those in the baseline group. The estimates of 3 range from 0:34 to 0:60 in the full sample and the estimate is up to 0:67 in the sample of individuals less than 45 years of age. The results from the full sample can be interpreted as follows: as the real price level in the host country goes up one standard deviation (32:32) above the mean (122:4), a migrant who engages in both remittances and repatriated savings is associated with a change in the real wage that is 9:03 to 15:84 percentage points below the corresponding change for the migrants who engage in neither activities. 4.3 Discussion To put the estimates based on the U.S. city-level analysis in perspective, consider extrapolating the results to the case of the overall U.S. and Mexico as the host and source countries, respectively. Over the period considered in this study, 1967-2003, the real exchange rate between the U.S. and Mexico uctuated around the mean of 1:73 with the standard deviation of 0:37. 20 When put in the context of Mexico-U.S. migration, the estimates in Table 7 can be interpreted as follows: As the purchasing power of the dollar in Mexico goes up one standard deviation above the mean, the change in the real wage of Mexican migrants who engage in both remittances and repatriated savings is 12:83 percentage points below that of migrants who engage in neither activities. 21 This is certainly an e ect that is economically signi cant and surprising in that a variable that is not at all intrinsic to the migrant s ability to perform in the job, namely real price di erences between the host and source countries, can in uence 20 The minimum was 1:26 and the maximum was 2:72. This means that the U.S.-Mexico real exchange rate was never at PPP during the period. 21 ( 0:6) (0:37=1:73) = 0:1283 17

her labor market outcome. 22 To the extent that the migrants who neither remit nor bring savings serve as a good benchmark, this paper suggests that the motivation for cross-border price arbitrage may in uence the labor market outcomes of migrants in a way that standard models of migration with PPP would not have predicted. 5 Conclusion Real price or cost-of-living di erences between regions, especially between rich and poor regions, are a well-known phenomenon. This paper investigates cross-border optimization of consumption mix through remittances or repatriated savings as a strategy that migrants from poor countries may adopt in order to take advantage of the persistent real price gap between regions. This is a phenomenon that is becoming increasingly relevant with the rise of temporary migration and the decline in transportation costs, including the costs of sending remittances. An important consequence of the migration strategy discussed in this paper is that the real wage that a migrant is willing to accept in the host country goes down when the real price level in the source country is lower than that in the host country. This in turn suggests that when we hold other factors constant, the observed labor market outcome in the host country may look worse for a migrant originating from a poorer country in which real price level tends to be lower. Using data on the labor market outcomes of Mexican migrant workers in the U.S., this paper showed some evidence that is in line with the main theoretical argument. Speci cally, it nds a negative association between the real price level in the host country (relative to that in the source country) and the real wage received by migrant workers. This paper contributes to the literature by incorporating a macro-level phenomenon, namely, real price di erences between rich and poor countries, into a migration model and providing an explanation for why migrant workers from poorer countries may be working for less in a rich host country. Though the empirical analysis in this paper considered the case of Mexico-U.S. migration, the analytical framework should be applicable to other pairs of countries or regions such as east-west migration within Europe 22 Note that the underlying di erences between migrants in the remittances and no-remittances groups are accounted for. In addition, di erences in the human capital and demographic characteristics of all migrants are controlled for throughout. 18

or rural-urban migration in China. References [1] Balassa, Bela (1964). The Purchasing Power Parity Doctrine: A Reappraisal, Journal of Political Economy, 72(6):584-96. [2] Beaman, Lori (2007). Social Networks and the Dynamics of Labor Market Outcomes: Evidence from Refugees Resettled in the U.S., working paper, Northwestern University. [3] Becker, Gary, 1981. Treatise on the Family, Cambridge: Harvard University Press. [4] Bergstrom, Theodore C. (1997). A Survey of Theories of the Family, in Handbook of Population Economics, edited by Mark R. Rosenzweig and Oded Stark, North-Holland. [5] Carrington, William J., Enrica Detragiache, and Tara Vishwanath (1996). Migration with Endogenous Moving Costs, American Economic Review, 86(4): 909-930. [6] Chau, Nancy H. (1997). The Pattern of Migration with Variable Migration Cost, Journal of Regional Science, 37(1): 35-54. [7] Djajić, Slobodan (1989). Migrants in a guest-worker system, Journal of Development Economics, vol. 31. [8] Durand, Jorge, William Kandel, Emilio A. Parrado, and Douglas S. Massey (1996). International Migration and Development in Mexican Communities, Demography, 33(2): 249-64. [9] Dustmann, Christian (1995). Savings Behavior of Return Migrants: A Life-Cycle Analysis, Zeitschrift fur Wirtschafts- und Sozialwissenschften, 115: 511-533. [10] Dustmann, Christian (1997). Return Migration, Uncertainty and Precautionary Savings, Journal of Development Economics, 52: 295-316. [11] Dustmann, Christian (2003). Return Migration, Wage Di erentials, and the Optimal Migration Duration, European Economic Review, 47: 353-369. 19

[12] Fox, Marc and Oded Stark (1987). Remittances, Exchange Rates, and the Labor Supply of Mexican Migrants in the U.S., Harvard University Migration and Development Program, discussion paper no. 33. [13] Goldberg, Penelopi K., and Michael M. Knetter (1997). Goods Prices and Exchange Rates: What Have We Learned? Journal of Economic Literature, 35(3): 1243-1272. [14] Hanson, Gordon H., and Antonio Spilimbergo (1999). Illegal Immigration, Border Enforcement, and Relative Wages: Evidence from Apprehensions at the U.S.-Mexico Border, American Economic Review, 89(5), 1337-1357. [15] Harris, John R., and Michael P. Todaro (1970). Migration, Unemployment and Development: A Two-Sector Analysis, American Economic Review, 60: 126 142. [16] Hoddinot, John (1994). A Model of Migration and Remittances Applied to Western Kenya, Oxford Economic Papers, 46(3): 459-476. [17] Ku, Hyejin (2008). Labor Migration as Price Arbitrage: Evidence from Mexican Workers in the U.S., working paper, Florida State University. [18] Massey, Douglas S., L. Goldring, and Jorge Durand (1994). Continuities in Transnational Migration: An Analysis of Nineteen Mexican Communities, American Journal of Sociology, 99(6): 1492-1533. [19] Massey, Douglas S., and Kristin E. Espinosa (1997). What s Driving Mexico-U.S. Migration? A Theoretical, Empirical, and Policy Analysis, American Journal of Sociology, 102(4): 939-999. [20] McKenzie, David, and Hillel Rapoport (2007). Network E ects and the Dynamics of Migration and Inequality: Theory and Evidence from Mexico, Journal of Development Economics, 84: 1-24. [21] Munshi, Kaivan (2003). Networks in the Modern Economy: Mexican Migrants in the U.S. Labor Market, Quarterly Journal of Economics, 118(2):549-99. 20

[22] Orrenius, Pia M., and Madeline Zavodny (2005). Self-Selection among Undocumented Immigrants from Mexico, Journal of Development Economics, 78: 215-240. [23] Rapoport, Hillel, and Frederic Docquier (2006). The Economics of Migrants Remittances, in Handbook on the Economics of Reciprocity, Giving and Altruism, Volume 2: Applications, edited by Serge-Christophe Kolm and Jean Mercier Ythier, Elsevier-North Holland. [24] Rogo, Kenneth (1996). The Purchasing Power Parity Puzzle, Journal of Economic Literature, 34(2): 647-668. [25] Rosenzweig, Mark R. and Oded Stark (1989). Consumption Smoothing, Migration, and Marriage: Evidence from Rural India, Journal of Political Economy, 97(4): 905-926. [26] Samuelson, Paul A. (1964). Theoretical Notes on Trade Problems, Review of Economics and Statistics, 46(2):145-54. [27] Sjaastad, Larry A. (1962). The Costs and Returns of Human Migration, Journal of Political Economy, 70(5): 80-93. [28] Stark, Oded (1991). The Migration of Labor, Oxford, UK; Cambridge, MA: B. Blackwell. [29] Stark, Oded, and David E. Bloom (1985). The New Economics of Labor Migration, American Economic Review Papers and Proceedings, 1985: 173 178. [30] Stark, Oded, and Simon C. Fan (2007). The Analytics of Seasonal Migration, Economics Letters, 94(2): 304-312. [31] Stark, Oded, Christian Helmenstein, and Yury Yegorov (1997). Migrants Savings, Purchasing Power Parity, and the Optimal Duration of Migration, International Tax and Public Finance, 4: 307-324. [32] Taylor, Alan and Mark Taylor (2004) The Purchasing Power Parity Debate, Journal of Economic Perspectives, 18(4): 135-158. 21

[33] Union Bank of Switzerland (2008). Prices and Earnings: A Comparison of Purchasing Power around the Globe, UBS. [34] World Bank (2006). Global Economic Prospects 2006: Economic Implications of Remittances and Migration, The World Bank. [35] Yang, Dean (2006). Why Do Migrants Return to Poor Countries? Evidence from Philippine Migrants Response to Exchange Rate Shocks, Review of Economics and Statistics, 88(4): 715-735. 22

Appendix A: Geographic Areas by Data Sources City MSA State State Data on migrants CPI GDP Min wage (MMP) (U.S. Bureau of Labor Statistics) (U.S. Dept of Commerce) (U.S. Dept of Labor) Atlanta Atlanta GA GA Chicago Chicago-Gary-Kenosha IL IL Dallas Dallas-Fort Worth TX TX Houston Houston-Galveston-Brazoria TX TX Los Angeles-Long Beach Los Angeles-Riverside-Orange County CA CA New York New York-Northern New Jersey-Long Island NY NY Orange County Los Angeles-Riverside-Orange County CA CA Philadelphia Philadelphia-Wilmington-Atlantic City PA PA Riverside-San Bernardino Los Angeles-Riverside-Orange County CA CA San Diego San Diego CA CA San Francisco San Francisco-Oakland-San Jose CA CA San Jose San Francisco-Oakland-San Jose CA CA Note: Each city includes at least 25 migrants in the MMP sample. 23

Table 1: Characteristics of Mexican Migrants in the U.S. 1967-2003 Variables All Remit Return with savings Remit and save Panel A: Prices, wages, and migrant characteristics CPI (base year = 1982-1984) 122.40 123.61 124.22 125.32 (32.32) (32.19) (32.82) (32.86) Hourly Wage (current USD) 6.54 5.69 7.57 6.06 (27.47) (6.57) (36.84) (3.52) Real wage (1982-1984 USD) 0.05 0.05 0.06 0.05 (0.16) (0.05) (0.21) (0.03) Hours worked per week 45.71 46.38 45.99 46.45 (14.01) (13.74) (13.60) (13.20) Year of migration 1988.40 1988.70 1988.85 1989.11 (6.73) (6.71) (6.87) (6.89) Age at time of migration 33.12 33.74 33.29 33.69 (10.46) (10.27) (10.87) (10.60) Education (years) 6.12 5.74 6.38 6.05 (4.00) (3.77) (4.08) (3.80) English pro ciency (0-4) 1.18 1.19 1.30 1.29 (1.29) (1.26) (1.32) (1.28) Undocumented 0.73 0.76 0.72 0.74 (0.44) (0.43) (0.45) (0.44) Female 0.03 0.02 0.03 0.01 (0.17) (0.15) (0.16) (0.12) Married 0.74 0.77 0.71 0.74 (0.44) (0.42) (0.45) (0.44) Minors (number of children < 18) 2.40 2.64 2.27 2.46 (2.26) (2.34) (2.28) (2.34) Number of Land parcels owned 0.15 0.17 0.18 0.19 (0.43) (0.45) (0.45) (0.47) Number of Property owned 0.58 0.60 0.60 0.60 (0.58) (0.58) (0.58) (0.56) Number of Business owned 0.14 0.13 0.13 0.12 (0.37) (0.36) (0.35) (0.34) Panel B: Occupational categories Agriculture, husbandry, forestry/ sheries 0.15 0.16 0.15 0.16 Skilled Manufacturing 0.20 0.20 0.18 0.18 Unskilled Manufacturing 0.33 0.33 0.33 0.33 Service 0.22 0.23 0.24 0.25 Other 0.10 0.09 0.10 0.08 Observation 1120 861 597 472 Note: Standard deviations are in parentheses. 24