MAUaC* ,qqg Frontier Issues in International Migration Oded Stark Drawing on the general assumption that information is imperfect, this article addresses three main issues: First, why do some migrants return even though the intercountry wage differential does not reverse? And who returns? Second, wby do migrants who stay tend to share their higher earnings with others at origin, even in the absence of altruism or of a need to establish an exchange relationship? And can the size of these transfers be predicted? Thir4 what explins the earnings of migrants? Why do they often dominate the earnings of equvalent native-born workers even if differences in human capital are fully controlled for? The article suggests these answers: First, when informational symmetry is reestablishe4, the low-skill workers, who are no longer pooled with the high-skill workers, return. Second, migrants' remittances are conceived as side-payments, made under asymmetric information, by bigh-skill migrant workers to low-skill workers, who, if they were to migrate, would erode the wages of the high-skill workers. And third, the edge migrants have over naive-born workers arises from the lower recog- nition costs of partners to trade whose type is unknown. T his article delineates a number of frontier issues in international labor migration. Why do migrants return even though the wage differential does not reverse? Who returns? Why do migrants who remain and enjoy higher wages share their earnings with others at home, even in the absence of altruism or of a need to establish an exchange relationship? And why do the earnings of migrants often dominate those of native-born workers even when differences in human capital are controlled for? By the very nature of these issues, the analysis is preliminary and suggestive, the link with ^.mpirics is partial and often indirect, and the policy repercussions are tentative. Should these issues prove responsive to concrete policies, answers to these questions could serve real-world concerns. The analysis delineates a number of concrete, testable implications and several distinct policies. Oded Stack is professor of development economics at the University of Oslo. The first scction of this article is based on Stark (1995b); the second and third sections are based, respectively, on chapters 4 and 5 of Stark (1995a). Proccedings of the World Bank Annsal Confernce on Developmet Economics 1994 01995 The International Bank for Reconstruction and Devdopment / THE WORID LANK 361 3 Fron;te ,zon ti ssues in Intenational Migration Migration is significant in almost every country, whether as an inflow or an out- flouv. The questions of how migrants perform in the marketplace and what explains their performance, which to a considerable extent involve efficiency and equity con- cans, are basic to research on migration. Moreover, remittances are not a pittance. In 1980 workers' remittances accounted for as much foreign exchange as exports did for Pakistan and Upper Volta, more than 60 percent of exports for Egypr, Porrugal, and Turkey, and about 40 percent for Bangladesh and Yugoslavia (World Bank 1984). Despite difficulties in assessing return migration, there are reasons to believe that the numbers are often large, even in countries ordinarily viewed as countries of des- tination such as the United States and Germany. Data for 1908-57, when the United States collected information on the arrival and departure of migrants, show that of the 15.7 million migrants admitted, 4.8 million-nearly one-third-departed (LaLonde and Topel 1993). Moreover, the foreign population in the Federal Republic of Germanv was no larger in 1988 than it was in 1983 (OECD 1990); dur- ing these six years the outflow of the foreign population was 2,378,000. Of course, this outflow includes asylum seekers whose requests were refused. The number of asylum seekers received by Germany during the period was 389,000. Abstracting from the lags and subtracting all the asylum seekers, the outflow totals 2 million, which all indications suggest consists largely of returns. Because no administrative barriers hamper migration between Puerto Rico and the United States, issues of asy- ium and work permits do not arise, making estimates of return migrants much sim- pler. In 1980 about 1 million people born in Puerto Rico were residing in the United States, but 283,000 returned to Puerto Rico from the United States, where they had resided between 1970 and 1980 (U.S. Department of Commerce 1983, 1984). A recent survey of young Irish labor market particpants provides additional help- ful evidence. Of 1,299 Irish school leavers in 1981-82, 378 migrated at least once between leaving school and 1988; of these, 117 returned by 1988. The returnees were significandy less educated than the migrants who remained. Controlling for labor market conditions at destination, return migration is more likely from European Union countries-other than the United Kingdom-in which employers are said to "fail to recognize Irish qualifications" (Reilly 1994). I have addressed the questions posed in this article before, in The Migration of Labor (Stark 1991). This artide steps beyond that work. Whereas the book highlights the importance for migration outcomes of the relationship between rmigrants and their fam- ilies at home, this artide emphasizes the relationship between migrants and their employers and among migrants themselves. Moreover, this artide assumes that infor- mation is imperfect and examines the repercussions of this assumption for migration outcomes. Agents that meet to conduct a transaction usually do not fully know the char- acteristics of fellow agents that impinge on the transactions outcome. In the context of an employment relationship this applies to the skdll level of (new) employees (the first and second sections). In the context of marketplace trades this applies to the type of partner (the third section). This article demonstrates the considerable explanatory and predictive powers of analyses based on the assumption of imperfect information. Stark 363 In the first section, Dynamics and Return, the theory of labor migration under asymmetric information is used to generate an integrated set of predictions. The key results are the prevalence of return migration and a characterization of the migrant workers who return. In particular, in the discrete case with only two skill levels, low- skill workers return while high-skill workers stay. The wage of the high-skill work- ers increases over time, but not as a result of skill enhancement. When individual skill levels of migrant workers are unknown to employers, the workers receive a wage based on the average product of the group. This averaging invites low-skill workers to move along with high-skill workers. Once informational symmetry is restored through monitoring and observation and wages are adjusted accordingly, low-skill workers return. The high-skill workers stay, and their wages rise. Now a yet higher-quality subset of workers migrates, and the process is replicated. By migrat- ing, workers constituting the first subset block the migration of workers constituting higher-uality subsets. But by subsequently exposing themselves to identification, these workers pave the way for the migration of higher-quality workers. Thus migra- tion proceeds in waves, with each wave brealkng into workers who stay and work- ers who return; within waves the returning mi-grants are the low-skill workers. The second section of the article, Remittances as Side-Payments, supplements the model developed in the first secton with a strucmre to incorporate the possibility of transfers by migrants. The key results are the identification of a new motive for remittances and the determination of their size as a function of model parameters. The idea here is that since high-skill workers benefit if low-skill workers do not migrate, they should be willing to remit to induce them to stay put. The conditions under which such transfers are made are detailed, and their magnitude is deter- mined. Migrants remit to nonmigrants not out of altruism, but out of self-interest: remittances protect the wages of high-skill workers from the contaminating pres- ence of low-skill workers in the same pool. The basic premise of the first two sections is that incentives to migrate, return, and remit arise from wage differentials conditional on informational regimes. The issue of incentives lies at the heart of migration research, and it has also recently been incorporated into the analysis of the absolute and relative performance of migrants in the marketplace at destination: if migrants differ from nonmigrants in that they face an exogenous probability of return migraton and the receipt of lower wages, they have an incentive to work harder and save more than nonmigrants (Stark 1991, chap. 27 and 28). But the key to migrants performance may be found on yet another ring- a group attribute. The third section, Recognition Costs and Performance, considers a popula- tion consistng of two groups, mgants and mdigenous people. Each group consists of agents who trade cooperatively, C, and agents who trade noncooperatively, NC. Assume that members of each group trade only with members of their own group (this assumption can be relaxed). Agents do not know the type of the agents with whom they trade (again, information is imperfect), but they can obtain such information at a cost The idea is that the cost that migrants incur in assessing whether a fellow migrant is a type C or a type NC is lower than the analogous cost incurred by nonmigrants. In 364 riontier Issues in International Migration this situation the equilibrium share of C-type agents in the migrant group is higher than that in the nonmigrant group. And since, by construction, the payoff matrices of each of the groups are the same, the per capita payoff of migrants is higher than that of nonmigrants: the migrants outperform the nonmigrants. If the cost-of-information advantage is not present, however, migrants do not fare better than nonmigrants. The key results here are thus the ability to explain the differential performance of migrants without recourse to individual human capital attributes, and a prediction that when there are different groups of migrants, all else equal, the groups with low recognition costs will outperform the groups with high recognition costs. The theoretical literature on labor migration is remarkably silent on return migra- tion. Yet the number of migrant workers in any country is as dependent on depar- tures as on arrivals and, as already demonstrated, evidence suggests that migration is often shorter than the duration of life or of working life. With conventional the- ory attributing migration to a positive wage differential, a conventional explanation of return migration is, not surprisingly, a negative wage differential (Dustmann 1993). While this explanation might be correct, considerable return migration appears to take place even in the absence of a reversal of the relative wages of the sending and receiving countries. The typical explanations are that migrants return because of failure or because of success: if realiry does not tally with expectations or if the draw from a mixture of good draws and bad draws (random shocks) is bad, migrants may return. Alterna- tively, migrants whose returns on human or financial capital are higher at home than abroad may find it optimal to return (Borjas and Bratsberg 1992). There are other reasons for return migration, however. One is that a family mem- ber may migrate to diversify the ways in which the family earns its income. If income eamed abroad and income earned at home do not covary fully and there is a post- migration pooling and sharing of income, the family's risk is lowered. Just as bear- ing one risk makes agents less willing to bear another risk, not bearing that one risk makes agents more willing to bear another risk (Pratt and Zeckhauser 1987; Kimball 1993). People at home are then free to experiment with a relatively high-risk, high- return option, for example, a high-yield seed variety. If the experiment is successful, the need for migration-provided insurance ceases. Thus the reason for return nigra- tion is not that the migrant accumulated capital with an expected high return at home but rather that his or her migration facilitated a high-return investment at home by others (Stark 1991). Return migration may also occur because of the higher purchasing power of sav- ings (generated from work abroad) at home than abroad. Researchers have incorpo- rated the possibility that consumption at home is preferable to consumption abroad into carefully argued models (Hill 1987; Djaiic and Milbourne 1988). In research in progress Stark, Helmenstein, and Yegorov (1994) attempt to identify what underlies such a consumption preference, if it exists, and to account for return migration even if that preference is absent. Specifically, we investigate the role of purchasing power of given amounts of savings in facilitating different levels of consumption at home and abroad. While earlier work considered the effect of the probability of return migration Star* 365 on migrants' optimal savings (Stark 1991, chap. 27), this work develops the opposite line of inquiry how migrants' savings determine the optimal timing of return. Among other things, this work points to a negative relationship between the optimal duration of migration and the purchasing power differential. Further, in some cases it shows a negative relationship between the optimal duration of migration and the wage abroad. Yet return migration may be caused by reasons beyond the disposition of savings or the elimination of risks. And little is known "about whether it is high- or low- skilled workers who choose to return home" (LaLonde and Topel 1993). Research on migrants' performance has essentially assumed that the key to the relation of migrants' income to that of the native-born population lies in their abil- ity, skills, and productivity. This assumption has channeled considerable research energy into measuring, estimating, and testing migrant characteristics (Borias 1985; Chiswick 1986a and 1986b; LaLonde and Topel 1993). Recent work has considered the possibility that the performance of migrants could be attributed directly to their incentives to exert effort and to save, thereby shifting the focus of analysis from the vector of characteristics to the structure of incentives (Stark 1991, chap. 27 and 28). Yet groups too, not only individuals, have characterstics, and incentives contin- gent on an exogenous probability of return migration are nullified when the proba- bility of return approaches zero. Research on remittances has tended to attribute motivations to remit to altruism- concern for the well-being of family members who stay behind-or to self-interest arising from the intention to return, aspirations to inherit, maintenance of invest- ments at home, or a need for insurance against unfavorable income realizations at des- tination, such as spells of unemployment (Stark 1991). But these considerations do not exhaust the set of motives, and in particular situations none of them may apply. Dynamics and Return This section presents an implementation of the theory of labor migration under asymmetric information wherein return migration is an integral, structural part of the migration process. The addition of an explicit intertemporal dimension to the static model of labor migration under asymmetric information (Stark 1991) ampli- fies the model and makes it possible to differentiate and characterize workers who never migrate, workers who migrate and stay at their destination, and workers who migrate and subsequently return home. Suppose, first, that in a given occupation workers fall into two groups according to their skills-low-skill and high-skilL The theory offers the following predictions: migration is ex post fully positively selective even though ex ante it is not; it breaks into workers who stay and workers who return; and the returning migrants are low- skill workers. The judgment concerning the selective nature of migration is thus sen- sitive to the time at which the judgment is made. Whereas the end result of migration is not sensitive to the information regime (symmetric or asymmetric), the migration path is: it is single phased under symmetric information, but multiphased under asymmetric information. With the introduction of some auxiliary structure, 366 Frontier Issus in Interntional Migration the theory identifies a procedure that allows the receiving country to skim off the high-quality workers without engaging in costly screening. Suppose next that workers in the profession constitute four skill levels. A plausible implementation of the theory of labor migration under asymmetric information gen- crates the following predictions: migration is sequential or phased; not all wc;kers who end up as migrants move at the same point in time. Each wave of migration breaks into workers who stay and workers who return. The century-old "law of migra- tion" that "each main current of migration produces a compensating counter-current" (avenstein 1885) turns out to be a derivative of a variant of the asymnmetric-infor- mation approach to migration. Within waves (cohorts) the returning migrants are the low-quality workers; thus mnigration is ex post positively selective within cohorts. When the migration process is fully completed, migration is mildly positively selec- tive-the average quality of migrants is superior to the average quality of workers found at origin-but not all migrants are of higher quality than al workers at origin. (Only in the case of two types of workers does migration turn out to be ex post fully positively selective.) The average quality of migrants rises with each cohort. We next present the basic model of labor migration under asymmetric informa- tion, and follow it with migration patterns arising from an example with two skill levels. We then examine a case with four skill levels and derive the resulting migra- tory patterns. Finally, we place the approach utilized in this section of the artide in the context of related research on labor migration. Labor Migration under Asymmetric Information: The Basic Model Assume a world consisting of two countries: a rich country, R, and a poor country, P. We can likewise assume a country consisting of a rich urban area and a poor rural area. In a given occupation let the net wages for a worker with skill level 0 be WR (9) in the rich country and W. (0) in the poor country, such that aw, (0) / DO > 0 and aWR (9)1 aO > 0.1 (Thus workers' productivities in the sending and receiving coun- tries are ranked identically.) To reflect the fact that R is rich and P is poor, WR (0) > Wp (0) for all e2 Also, without loss of generality, let 0 be defined upon the dosed interval [0,1] and let the density function of P workers on 9 be F(9). In addition, given that P workers are likely to have a preference for a P lifestyle because of cultural factors, social relationships, and so on, it is assumed that they apply a discount factor to R wages when comparing them to P wages. Thus when making the migration decision, they compare kWR (9) with W. (e), where 0 c k c 1. A P worker will therefore migrate from P to R if (1) kWR (9) >,Wp (0). Clearly, without further restrictions on WR (0) and Wp (9) there may be several val- ues of 9 for which kWR (9) - Wp (9) = 0. Hence there may be several distinct skill groups along the skill axis (figure 1). Thus the workers in skill intervals [O,913, [E,e'31, and [04,1] migrate, whereas those in the complementary intervals do not. The case in which there are at least Stark 367 Figure 1. Disjoint Skill Intervals of Migration under Symmetric Information _~~~~~~~~~~~ I I 0 el fit83 84 1 three distinct groups (for example, along the 0 axis, migrating, nonmigrating, and migrating)-a situation that can occur only if at least one of the Wp (0) and W. (0) functions is nonlinear in 0-will be referred to as the nonconvex case. Similarly, the case in which there are two or fewer distinct groups will be referred to as the convex case. Let us now assume that the skil of each potential migrant is known in P where he or she has been observed for a while, but is unknown in R. When markets are isolated in the sense that information does not flow across them (or does not flow costlessly and freely), employers in one market may possess information on worker productivity-for example, as a byproduct of monitoring and coordinating activities-but the information is employer- or market-specific. Also, for the moment, let us exclude the possibility that true skill is revealed in R over time. Facing a group of workers whose individual productivity is unknown (only the distribution of earnings abilities is known), the employer will offer the same wage to all, and it will be related to the average product of the group. Let us assume that the actual individual wage offered is equal to the average product of the group and that wage offers are known to all workers.3 Hence denoting by WR the wage payable in the rich country to a migrant of unknown skill level and assuming n distinct migrating groups, WR is given by 368 Aontier Issues m Intenational Migration (2) WR = ITWA (0)F(0)d0/ X JF(O)d i=ld / 1 where 0' is the lowest skill level and Oi is the highest skill level among the migrants in group i, where i is one of the continuous groups migrating, and where the skill level increases with i. (Note that 0 < 01 < Wp (i*). Also, since 6 Wp (6) so that a worker of skil level 6 also mnigrates. The implication of this result is that under asymmetric information everyone with a skill level less than or equal to E" migrates, so that all workers in the interval [0,6"] migrate. Note the contrast with the case of fIll information, as shown in figure 1, where the migration pattern could be nonconvex. Thus under asymmetric information the wage payable to all migrating workers in R is (3) WR = JWR (G)F(O)dOl | F(O)dO 0 0 where 0* is the top skill levd migrating. Thus WR can be written as WR (0*). Under asymmetric information, then, workers of skill level 0 for which (4) kWR (0) > Wp (0) will migrate from P to 4 Given this characterization of the migration pattern under asymmetric informa- tion, we can now proceed, first, to an example of a convex (two-group) case and then to an example of a nonconvex case. A Convex Case Assume there are just two types of workers: low-skill workers, whose skill level is 1, and high-skill workers, whose skill level is 02, with skill-related wage rates W, (el) and W, (02); in the poor country i = P and in the rich country i = R. Assume that low-skill workers constitute a percent and high-skill workers 1 - a percent of workers in the profession. Suppose that no costs are associated with migration except those embodied in k, and that k is such that kWR (01) < WP (01) yet kWA (p2) > WP (02). This assumption is introduced to capture the differing migration incen- Stark 369 tives of the symmetric and asymmetric information states. It implies that under symmetric information only the relatively high-skill workers will migrate. How- ever, if we assume that (5) aXkWR (01 ) + (1 -a) kWR (02.) > Wr (02) then, under asymmetric information, the 02 workers will again migrate but this time the 01 workers will migrate as well (a result that follows immediately from the above lemma). If employers in R correctly identify the skill levels of individual workers and adjust pay accordingly at the end of the first employment period, the low-skill work- ers will return to P and the high-skill workers vill stay in R. Since 9I workers are not pooled with O2 workers, 02 workers' R wage can only be higher, that is (6) kWR (02) = OEkWR (02) + (1- a) kWR (02) > akWR (01)+ (1-ot) kWR (H2)- By assumption, the right-hand side of the inequality is larger than the alternative poor country wage, W. (02). This outcome has three implications. First, considering the entire migration expe- hience, we see that migration is positively selective. Even though no selectivity is observed initially-both low-skill workers and high-skill workers leave-with the passage of time and the removal of informational asymmetry, the return of the low- skill migrants to their home country produces a feature of positive selectivity. Whereas initially migration is not selective in skills, ex post it is. Second, the judgment concerning the selective nature of migration is sensitive to the time (phase) at which the judgment is being made. (At first, migration appears not to be selective; at last, it is fully positively selective.) Empirical findings con- cerning the selective nature of migration are thus phase dependent. Third, even though the end result of migration is not path dependent, the sym- metric information single-phase path (with only workers of skil level 02 migrating) is different from the asymmetric information multiphase path (with group 02 found in R only when migratory moves halt altogether). What if the rich country wishes to have only high-skill migrant workers from the start, that is, it is unwilling to await return migration by the low-skill migrant work- ers? Suppose that screening individual migrants (would-be or actual) is cosdy or unreliable. The asymmetric information approach identifies an instrument that facil- itates skill differentiation. Return migration and this instrument are thus mutually exdusive. The rich country can announce an entry tax (visa fee) of T units. This tax must be large enough that low-skill workers find migrating not worthwhile under asym- metric information but not so large as to swamp the high-skill workers' discounted wage differential. To secure these dual requirements, it is necessary to find the min- imal tax that solves (7) k[cWR (01) + (1- a)WR (02)- < WI (Op. 370 Frontier Issues in International Migration That is, the tax T should solve (7') k[aWR(O) + (1 -a)WR()-(T0)J=WPI(1) where E > 0 is a sufficiently small constant, while maintaining (8) k[W,, (02) -if] > WI (02)- From equations 7 and 8 we obtain (9) kaxWR (01)+ k(l -a)WR (02) -WrP(01) kT < kWR (92) - Wp (02). Existence then requires that (10) WP (02) -'WP (0k) C ak[WR ((2) - WR (01)]. Existence is thus more likely the steeper the wage profile is by skill in the rich coun- try relative to that in the poor country. If the share of low-skill workers in the occu- pation under review, a, is relatively large, and if the ratc-of-location discount is not high, the entry tax that solves equation 7 will also fulfill equation 8. A numerical example illustrates the convex case. Suppose W. (01) = 7, Wp (°2) = 9, WR ,(1) = 10, WR (02) = 20; F(9) is such that a = 1 - a = 1/2; and k = 2/3. Thus under symmetric information only 02 migrate as kWR (02) = 2/3 x 20 > 9 but kWR (81) = 2/3 x 10 < 7; under asymmetric information both skdll levels migrate as kaWR (0l) + k(1 - a)WR (0Z) = 2/3 x 1/2 x 10 + 2/3 x 1z2 x 20 = 10 > Wp (01) = 7; W; (02) = 91. As for the tax scenario, equation 7 gives a tax T = 4.5 + & With this tax in place it can be seen that low-skill workers will be worse off regardless of whether they migrate with the high-skill workers or alone: in the first case, their prediscounted wage will be 10.5 - E units, which is worth less to them than the alter- native home-country wage (2/3 (10.5 - a) < 7); in the second case, their predis- counted rich-country wage will be 5.5 - E, wvhich is below their home-country wage. Not so for the high-skill workers, whose post-tax, discounted rich-country wage is still superior to the home-country wage (2/3 (15.5 - £) > 9). A Nonconvex Case Assume that there are four types of workers with skill levels Oi increasing in i (i = 1, ...,4) with corresponding wage rates of Wp (0-) in the poor country and WR (9k) in the rich country. Suppose that F(9) is given, that is, the proportion of skill type i in the profession is a,. Once again it is assumed that no costs are associated with migra- tion except those embodied in k. Suppose that even though WR (0,) > WP (0,) for all i = 1, ...,4, the skill-specific wage rates are such that kWR (02) > Wp (e) and kW, (04) Stark 371 > WP (04), whereas kWR (P l;) Wv (0) kil =1 and k (ai J aiWR (O) > WP (04) ;=3 i=3 while k fxa j lai WR (0) < WP (03) and 4 k Xa;WR (0i) Wp (02) = 8 and kWR (O4 = 1/2 x 48 > w,(e4) = * X Liy:, but kWR (O,) = 112 x 10 C WP (01) = 6 and kWR (03) = 1'2 x 40 < Wp (03) = 21), under informational asymmetry only 0, and 02 migrate. Clearly, 0. and 04 are better off staying in P since kW1 (0*) I O9=93 = 1/2 x 80/3 < WP (03) = 21 or kW-i(t*) I -=9_ = I X 128/4 cWP (04) = 211/2. Verification that 0, + 02 will migrate is also straightforward as kW(0*) 9.=e2 = 1l2 X 40/2 > [Wp (01) = 6; WP (02) = 81. Once information is revealed completely, workers of sldll level 01 return to P and workers of skill level 02 stay in R. Types 03 and 04 now migrate as kWR (0*) 1 8=O4 = 1/2 X 88/2 > [W, (03) = 21; WP (04) = 211/2]. Once again, the reve- lation of full information splits the mnigrants into retnrnees and stayers: workers of skill level 03 return to P, since kWR (03) = 1/2 x 40 < WP (03) = 21, while workers of skill level 04 stay in R because, for them, kWR (04 = 1JZ x 48 > W7p (0W) = 211/2. Since the (location discounted) earning profile of the 04 workers is 211/7, 22, and 24, advancing their migration results in an earning sequence of 142/3, 24, and 24 that under any of the restrictions postulated above cannot be sustained. Complementary Remarks This setup-all workers know what wages will await them; in response, they stay put, rigrate and stay at destination, or migrate and return; and stayers, movers, and 374 Fontier I" i Intenationl Migraion those who return are fully characterized-is new. In many professions (for example, science and engineering) in which employers know little about new workers' abilities and in which workers' abilities correlate strongly with productivity, the improvement in information as time passes rests with the employers, not with the migrant workers. The sequential, relative, and return attributes of migration as derived in this sec- tion of the article do not arise, then, from imperfect information about wage rates at destination. If they did, even if the migrants had precise information on their expected wage at destination, the reality of wage variance would induce some migrants to return and others to stay. But if we recognize that worker attributes dif- fer, the attributes must be systematically correlated with realized wage rates. It is not enough merely to argue that return migration is a decreasing function of pre- migration information (McCall and McCall 1987) or that "migration back to an original location occurs because expectations were not fulfilled" (Polachek and IHorvath 1977). Dynamics in general, and return migration in particular, could be generated by changes in information in a more subtle way. Suppose that workers have informa- tion on wages in location i where they are currently located, and on wages in a series of other locations j, k, I, and so on. Suppose furither that workers always have more information on the location they are actually in than on other locations. Finally, suppose that the value of this information relates inversely to its quantity. Suppose now that workers move from i to j. Then, not only does information on j become less valuable than it was before the move, it could also become less valuable than information on 4 i, I, and so on. Since the only way to convert information on a wage elsewhere-that is, now, on wages in 4 k, or IJ and so on-to an actual wage is to move, one move may well lay the ground for subsequent moves. Clearly, one such move is back to i. Here, too, changes in information could play a role-moti- vating migration, induding retrn migration-but the changes take place in the information the migrant workers have, not the employers, and a systematic link with workers' attributes is missing. A simple cobweb model could generate some dynamics if we assume, again, that realized wages differ from anticipated wvages. An initial wave of migrants lowers the wage at destination, an outcome not foreseen by the migrants. Consequently, some migrants return. As a result the wage at destination rises somewhat and thus pulls in some migrants. And so on. Once again, this approach assumes homogeneity of workers' attributes, that the workers drawn in and the workers pushed out are always randomly selected, and that workers are unable to assess accurately their des- tination wages. Finally, sequential migration could arise because the technology of production exhibits economies of scale to the application of skilL Consider the following exam- ple. For each skill level 0, workers in economy R are paid more than workers in economy P with the wage differential increasing in 9. Skills can be acquired, albeit at a cost, and migration from P to R can take place at a cost c. Initially, the system is in full equilibrium with no migration. Suppose that as a consequence of an exoge- Stark 375 nous shock, c falls so that now WR (01) - c > WP (O"), where 0" is the top skill level. As 0"-type workers migrate, they confer both positive and negative externalities: the productivity of skilled workers in R rises because of the enlargement of the pool of skilled workers and the operation of scale economies. WR (0") rises as a consequence. Workers in P with skill levels below 0" who previously had no incentive to invest in acquiring additional skills now find that the joint return to such investment and migration is greater than the sum of the returns arising from each of these invest- ments undertaken separately. They also witness a decline in their wage earnings due to the absence of the 0" workers. They therefore invest in acquiring skills and then nigrate, spurring a second wave of migrants. The process repeats until the cost of migration exactly offsets the increase in the wage differential induced by the (two- ended) scale economies, or until all skilled workers leave P for R. Notice that if the reason for the initial distribution of skills is ability (see Miyagiwa 1991), the quality of migrants, as measured by their ability, will decline in the order of their cohort. Remittances as Side-Payments Several reasons besides the large magnitude of remittances encourage modeling of remittance behavior. First, predictions of the response of remittances to perturba- tions in the incomes of both remitters and recipients are sensitive to the motive for remitting. Policy is likewise sensitive. Consider altruism and exchange as motives. According to the altruism hypothesis, remittances received and recipients' pretrans- fer income will be inversely relatedc altruistic migrants will remit more when recip- ients' income falls short. But if transfers are motivated by self-interest and exchange considerations, recipients' income and remittances can be positively related. For example, if transfers pay for care of cattle left behind, an increase in the recipients' market wages will drive up the price of the services they provide. If remittances are viewed as an informal mechanism to increase welfare and equalize income, govern- ments may refrain from hindering them or may, indeed, support policies favoring them. The policy choice should be contingent on the underlying motive. Second, remittances are a unique form of transfer. Although transfers between family members who coreside are difficult to measure, remittances between migrants and their famnilies who stay behind are measured more easily. And observed transfers between family members can unravel hidden underlying intrafamily relationships. Third, interest in nonmarket exchange, intrafamily transfers, and intergenera- tional linkages is growing. The study of remittances constitutes part of this inquiry and could advance it considerably. In this section we study a strategic motive for remittances that has not been explored in the literature. The basic idea is that when employers at destination lack information on the skills of individual migrants, they will pay all migrants a wage based on the average product of the group. High-skill workers should be willing, then, to make a transfer to low-sill workers to dissuade them from migrating. Thus migrants renit to nomnigrants motivated not by altruism but by pure self-interest. 376 Frontier Issues in International Migration Motivating Remittances Consider the basic model and the two-group example of the first section, but suppose now that workers can form cohesive groups and act jointly. Since the high-skill work- ers wish to protect their wages from contamination by the presence of the low-skill workers in the same pool, they should be willing to make a transfer to the low-skil workers to induce them to stay put. This will free the high-skiU workers from being pooled with the low-skill workers right from the start. Of course, the transfer (a cost) must be smaller than the associated benefit conferred by the difference between the R country wage of the high-skill workers if they were to migrate alone and their R coun- try wage if the low-skill workers were to mnigrae with them. Put differently, the trans- fer must be smaller than the high-skill workers' symmetrc information-asymmetric information wage differential. Formally, the transfer T has to fulfill the condition (11) t < WR (O2)-[WR (e) + (1 )R ](O) where (12) T =kaWR (01)+k(1-a)W (92)-Wp (81) + i where E > 0 is a sufficiently small constant. From equations 11 and 12 we obtain (13) kacWp (0)+k(1-a)WR (02) WP (01) CT < TWR (02) |[aWR (O1)+(l -a)WR (02)] = 4[WR (02)-WR (G1l- Considering the last expression in equation 13 we see the importance for existence of a steep wage profile by skill in the rich country. For T that fulfills equation 13 we thus obtain the following: by offering the low- skill workers t, the high-skill workers succeed in having the former stay put. Notice that the low-skill workers cannot extract a transfer larger than T by threatening to migrate as this threat would not be credible: if they were to migrate, these workers would receive a payment valued at kW- = koxWR (01)+ k (1 -a) WR (02). But if they stay put they receive Wp (O1) + T, which is larger than JAV by a And, of course, the high-skill workers are still better off in the wake of such a transfer because they are left with W. (0Z) - T which is worth k [WR (02) - TI to them, and this, by construc- tion, is better than a payment worth AAR1 Subject to the existence condition for TP holding, six predictions and implications emerge. First, if workers can act jointly, they will form action groups by type, and migration will be selective right from the start; only the high-skill workers will migrate. Testable implications then are that selectivity and remittances are positively related and that return migration and remittances are negatively related. Second, maigrants remit to nonmigrants to buy them off-to prevent them from migrating and contaminating wages at destination. Remittances will thus be targeted to those at home who have earning power since there would be no need to remit to Stark 377 those who would not credibly threaten to migrate. Migrants who remit to nonmi- grant members of their own households, or even to their community at large (as, for example, Turkish migrant workers in Germany are reported to do), may do so in part to enhance the welfare of the stayers, but also to improve their own well-beingY8 Quite often, migrants from a given area in P work together in the same site in P., so remitting to a well-defined target set of potential migrants effectively preserves the migrants' wage. This small-scale effect also helps prevent free riding by an individ- ual migrant who might be tempted to avoid remitting while enjoying the benefits of other migrants' contributions. Third, the role of remittances is cast in a new light. Remnittances enhance alloca- tive efficiency by countering the effect of informational asymmnetry, thus enabling all agents to locate on the utility frontier as implied by the first-best allocation rule of equation 1. Fourth, in addition to explaining why remnittances are initiated and predicting their precise magnitude, the strategic motive explains why they come to a halt. Once high- quality workers are identified, their wage is immune to erosion from migration of low- quality workers. The need to buy off the low-quality workers evaporates, and remittances cease. Fifth, group formation may involve some organizational cost. The asymmetric- information approach to labor migration predicts that the formation of groups is more likely when the rich-country wages of the high-skill workers and the low-skill work- ers differ gready and when skill levels of individuals become known only slowly. Sixth, suppose an entry tax T is in place that is large enough to dissuade low-skill workers from migrating under asymmetric information but not so large as to swamp the high-skill workers' own discounted wage differential (A formal derivation of the entry tax was provided in the first section.) Even if workers could form groups by type, low-ill workers would be unable to extract a transfer because they lack a credible threat of migration. Thus taxing migrants and transferring remittances to nonmigrants are mutually exdusive. Consistent with our third point, the entry tax enhances efficiency. Take a numerical example. Suppose Wp (01) = 7, Wp (92) = 9, WR (0) 10, WR (02) = 20, F(0) is such that a = 1- a = 1/2, and k = 2/3. If the high-skill workers offer to transfer T =3 +Fe, given by equation 12, the low-sill workers will stay put. Any threat to migrate to extract a larger transfer would not be credible: if they were to migrate these workers would receive 15 (which is worth only 10 to them). But if they stay put they receive an assured 7 + 3 + a >10. And of course, the high-skill workers are sdll better off in the wake of such a transfer because they are left with 17 - e, which is worth 111/3 - a to them, and this is better than a payment worth 10. This example helps elucidate an additional point about incentives and coordina- tion. Suppose 01 and O2 are a pair of brothers, and suppose that there are two such pairs. If all four migrate and the brothers share their wage evenly, each will receive 15, the value of which is 10. Now suppose that a high-skill brother attempts to induce his low-skill brother, z, to stay put. This scheme will not work. In the absence of migration by z, the remaining migrants wrill receive 113 (10 + 20 + 20), the value 378 Frontier Issues in International Migration of which is 111/9. This is better than 10 and therefore desirable, but there is only (11/9 - E) 3/2 = 12/3 - E to be transferred by zs brother to z, and since 7 + 12/3 - E < 10, the unwarranted migration cannot be blocked. If, however, z's brother per- suades the other migrant workers to join him in remitting to z, the scheme will work. Say each needs to remit 1 + E, which means that z will receive 10 + s, while each migrant will be left with 50/3 - 1 - £ and this is worth 2/3 (47/3 - E) = 104/9 - E, which is better than 10. Of course, a still better outcome would be secured if the two high-skill workers banded together to 'screen out" their two low-skill broth- ers; numerically though, the result would be identical to the one involving equal shares of each type of worker alluded to in the preceding paragraph. What is espe- cially interesting about this example is not only that coordination can secure a Pareto improvement that individual action cannot, but also that some migrants may remit to nonmigrant workers for whom they have no altruistic feelings. Moreover, by illustrating the dependence of strategic remittances on coordination and monitoring, the example implicitly suggests that remittances per migrant will decline as the group of migrants expands. The free-riding problem associated with buying off the low-skill workers who, from the point of view of the high-slill workers, are public "bads," is more likely to arise the larger and less cohesive is the group of high-skill workers. Complementary Remarks Earlier research sought to model and test the idea that migration is a strategy to secure remittances for the migrant's family (Stark 1991). The motives were taken to be concern for those staying behind or the need to pay them for past, ongoing, or future services. This article steps outside this framework. It argues that migrants may remnit to nonfamily members to protect their wages from contamination by fellow migrants in a pool whose members receive a wage based on the average product of the group rather than on individual product. If this motive operates, screening migrants at the point of entry will adversely affect remittance flows. This consider- ation connects with, and should thus impinge on, the political economy of migra- tion legislation, especially procedures aimed at sorting workers. Empirical inquiries concerning the relationship between transfers and the charac- teristics of recipients have produced conflicting results. Some studies conclude that remittances are motivated by altuism; others point to exchange or insurance as motives. And still others see remittance behavior as governed by a combination of motives.9 The conflicting results may have to do with a possible misspecification of the empirical remittance functions that stems from ignoring the strategic motive for migrant transfers. Suppose that the donor is the dominant player in the remittance arrangement, so that transfers from migrant to nonmigrant are given according to equation 12. This equation contains three variables: the migrant's actual earnings, the recipient's actual eamings, and the recipient's potential earnings in country R. The equation predicts that the coefficient for actual recipient earnings is -1. Suppose an empirical remittance equation is estimated that contains the actual earnings of Stark 379 migrant and recipient but not the recipient's potential earnings. The coefficient for recipient earnings in such an equation would be affected by omitted-variable bias. The direction of the bias is upward because the recipient's potential earnings have the effect of raising remittances, and the recipient's actual earnings and potential earn- ings in R are likely to be positively correlated. As a consequence of the omission, the positive effect of potential earnings in R is attributed to actual earnings, resulting in an algebraically higher coefficient. The possible omitted-variable bias might explain why empirical remittance functions produce conflicting sign patterns for the impact of recipient income on transfers, as the values of the omitted-variable-bias terms are likely to vary from one instance to another.10 Our analysis also forges a link betrween remittances and the ease with which employers can make judgments about individual skill levels. Consider the effect of occupational licensure on remittances. If migrants possess some certification that is recognized fairly quickly by employers in R, the informational asymmetry problem becomes less severe and might well disappear. If so, migrants worling in occupations that are commonly licensed would have a diminished incentive to remit Similarly, if a migrant is self-employed, informational asymmetries will not affect his or her earn- ings much, and again, incentives to remit will diminish. Thus the strategic model pre- dicts that remittance flows will be sensitive to the occupational structure of the migrant group. In particular, we expect that occupational status would be significant in empircal renittance equations even after controlling for the preremittance incomes of donors and recipients. Further, if we were to broaden our analysis to include occupational choice, we might expect that informational asymmetry would give migrants an added incentive to become self-employed. Indeed, a testable impli- cation is that the higher the dispersion of skill levels at origin (and, therefore, the greater the potential wage erosion), the more likely that high-skill migrants (who seek to avoid both pooling and remitting) will self-select into self-employment. Recognition Costs and Performance This section of the artide seeks to account for the empirical finding that migrants often outperform the native population. The underlying idea is that how migrants fare, absolutely and relative to the indigenous population, depends on group attrib- utes rather than on individual abilities and skills. It is postulated that characteristics of the market environment and trade technology, rather than returns to traditional characteristics of human capital, help explain-this outcome. Trade as a Game with Recognition Costs A population consists of two groups: migrants and the native born. Each group consists of agents who trade cooperatively, C, and agents who trade noncooperatively, NC In the model members of each group trade only with oilher members of their own group (see the appendix for a relaxation of this assumption). Agents do not know whether those with whom they trade are C or NC (again, information is imperfect), but they can 380 Frtmier Isues in Intiond Mgration obtain such information at a CosL The idea is that the cost to migrants of information about fellow migrants is lower than the analogous cost to nonmigrants. In this simation the equilibrium share of C-type agents in the migrant population is higher than the equilibrium share of C-type agents in the nonmigrant population. And since, by con- stn7ction, the payoff matrices of each subpopulation are the same, the per capita pay- off of migrants is higher than that of nonniigrants-the migrants outperform the nonmigrants.1I If their cost of information is not lower, however, migrants will fare no better than nonmigrants. We proceed as follows. Let a prisoner's dilemma type of table represent the payoffs from cooperation and noncooperation for two agents, E and E matched at random: Agent F C NC C C. T) (R. U) Agent E NC (.R) (s. S) In this payoff matrix U > T > S > R > 0 (and 2T > U + R; total payoffs are nax- niized when both agents cooperate). The share of type-C agents in a given group is PC and the cost of finding out the type of another agent is K Ž 0. In this environ- ment there is no memory-every trade is conducted as if it were the first-and C- type agents act first If all C-type agents engage in trade without determining the type of the trading partner, the payoff to a given C-type agent is Ilc = PCT + (I - PC) R. I£ however, a type-determining cost K is incurred, the payoff will be lc = T - X12 The cost will be incurred if nc > fIc, that is, if K < T- R)(1 - Pc) = K. Thus for values of K < K*, a C agent wili have a payoff of T-Kwhile an NC agent will have a payoff of S. Assuming for the rest of this article that (14) T-K>S (that is, the cost is never so large as to swamp the difference betwee the payoff from joint cooperation and the payoff from joint noncooperation), the C agents will have an edge, and their share of the population will rise.3 E, however, flic < Ic, that is, if K > K*, the C agents will trade randomly. In this case, though, the payoff to an NC agent will be IINc = PCU + (1 i PC) S The NC agent will have an edge if rINC > n that is, if PCU + (1 - P5 S > PcT + (1 - Pc) & which indeed holds since U > T and S > R14 Then, the share of the NC-type agents in the population will rise. We see that equilibrium obtains when K = K*; rhat is, when (15) PC = K1- T-R Two comments are in order. First, the equilibrium is stable since if the share of agents of a given type happens to be larger than the equilibrium share, their payoff will Stark 381 be lower than that of agents of the other type (and their share in the population will shrink), and vice versa. For example, if Pc happens to be lover than its equilibrium level, KI must maintain K* > K since aKlPIPc c 0. Hence the inequality NC > flc will hold, that is, the payoff of the C-type agents will be larger than the payoff of the NC-type agents and the population share of the C-type agents will increase. Second, since K 2 0, T > R and K < T- S < T- R (the first inequaliy is due to equation 14, the second to the payoff matrix), KI(T - R) is a fraction between 0 and 1. Therefore, Pc must maintain 0 < Pc < 1. This means that except for the two boundary cases, in equilibrium the population is a mixture of C-type agents and NC- type agents (such an equilibrium is called polymorphic). The two polar cases are as follows: If K happens to be as large as T - R (that is, as large as the difference for a cooperating agent between the payoffs from trading with a cooperator and with a noncooperator) there will be no cooperators; Pc will be sero. (If they incur the recog- nition cost, the C-type agents will have a payoff R,- because R is less than rIC for all values of PC, however, the C-type agents will be driven otu) On the other hand, if K is as low as zero, PC = 1; the noncooperators, who will always have a payoff of only S (< X), will be driven out. Equation 15 entails the following first result: the equilibrium share of the C-type agents in a population is inversely related to the cost of establishing the type of a party to trade. The proof is aPCIaK = -1i(T - R) < 0. What are the payoffs to C-type and NC-type agents at the equilibrum point? For a C-type the payoff is T- -1 and for an NC-type it is S. Therefore, the per capita pay- off is y = PC -) I 4- - PC) S. This entails the following second result the larger the share of the C-type agents in the population, the higher the per capita income. The proof is qy/aP, = T- K - S > O where the inequality sign is due to equation 14.15 Complementary Remarks The cost of establishing the type of a partner to trade helps account for the perfor- mance of migrants compared with that of the indigenous population. Typically, migrants originate in a closely linked group, constitute a more homogeneous and cohesive group than nonmigrants do, live in doser proximity to each other, and con- stitute a minority share of the population they join. These attributes render it cheaper for a migrant to trace the type of a fellow migrant. This cost advantage results in a larger equilibrium share of cooperating agents, which in turn leads to a higher per capita payoff, 1The findings of Chiswick (1986a and 1986b) and Bloom and Gunderson (1991)-that migrants who have been in the destination country for some time often have a higher mean income than the indigenous population-are explicable in this light'7 The higher incomes are not the result of superior skills and human capital or unobserved abilities and innately higher productivity, but of a trade and exchange environment that induces more cooperation. 382 Fwntier Isues in International Migration An interesting implication is that various "anti-clustering" processes aimed at assimilating migrar.s may, by raising the cost of establishing the type of a partner to trade, lower rather than enhance the well-being of migrants. Conversely, processes that reinforce cohesion among migrants tend to improve their economic performance. Condusion Under asyrmmetric information pertaining to skill levels, a mixed-skill group of migrant workers will split into low-skill workers, who return, and high-skill work- ers, who stay. Migration will not be selective to begin with, but will be positively selective thereafter. The wages of the migrant workers who stay will rise. This rise occurs not because of an increase in the human capital of individual migrants, bat as a consequence of the increase in the average level of human capital (skill level) of the group of migrants. Recalling the example of skill levels 9I through 04, the pos- sibility that the skill composition of migrants will be first (1 02) then (02), then (02, i., 04 and, finally (02, 04) implies that empirical tests of the selectivity of migration could produce conflicting results merely because a given dynamic migration process is observed at different points on its path.18 If high-skill migrants bind together to buy off low-skill workers, remittances will be made to low-sldll workers at origin and migration will be positively selective from the start. Remittances and selectivity will thus be positively related. Furthermore, remittances will be related to the occu- pational structure of the migrants: the higher the share of self-employment in total employment, the lower remittances will be. Under imperfect information about agent types, the per capita income of migrants belonging to a cohesive group or to a dosely knit ethnic group will be higher than the per capita income of native-born workers or, for that matter, than the per capita income of migrants who belong to less-cohesive groups. Ethnic or group cohesion positively affects measured market performance beyond individual attributes and incentives. Once again, it is not human capital (conventionally defined) that drives a migration outcome. The policy implications of the analysis can also be briefly summarized. Time-an implicit policy tool-will result in the departure of low-skill workers who, under initial asymmetric information, migrated along with high-skill workers. If the coun- try of destination is impatient, however, a precisely defined entry tax can be imposed to screen out the low-skill workers. If remittances as side-payments are fea- sible, the imposition of the tax means that the country of destination will reap a rev- enue that substitutes for remittances to the country of origin. There is considerable social and public policy interest in the assimilation and absorption of migrants. Econonics literature tends to measure assimilation by migrants' relative earmings (LaLonde and Topel 1993). Thus measured, assimilation is served by policies aimed at enhancing the cohesion of a group of migrants at destination rather than by poli- cies aimed at inducing their dispersion. Whereas the literature tends to attribute migration-related phenomena to human capital and to changes in human capital, assuming that information is symmetric and stark 383 perfect, the current artide has followed a reverse track: migratory outcomes have been attributed to states of information and to changes in them, while levels of human capital were held either unchanged or undifferentiated. In real-world migra- tory processes, information and human capital change over time, with the variance in migration outcomes attributable to variations in both. Indeed, how optimal investment by migrants in human capital-including devices and means that affect the cost and lag of skill discovery-responds to informational states lies on the fron- tier of research on international labor migration. Appendix Suppose that trade between migrants and nonmigrants can take place, that an agent can costlessly identify the type of group a trading partner belongs to but not the partner's C- or NC-type, and that a C-type agent can find out a partner's trait but at a cost. This cost, however, is larger than the cost pertaining to within-group detection. It is easy to show that a C-type migrant will not trade with a nommigrant. If he were to do so, incurring a cost KI > K where K' is the across-group cost and K is the within-group cost, his payoff would have been T - K, which is lower than T - K. If however, he were to trade randomly, his payoff would have been Ic = [a Pc + (I1-a) PM] T +[a(I- Pcm ) +I (- cE) C N )] where a (1 - a) is the share of the migrant (nonmigrant) group in the combined population and PM (PtNM) is the proportion of C-type agents in the migrant (non- migrant) group. This payoff is lower than the payoff arising from a random within- group trade. The proof is nc = PM T+(1-PcM)R=oPcM +(l-a)PcM ]T + [a ( Pc,W )+(I-(1-a(1 - C)] R> [oLPc + (1- a)PCN m] T + [a (1 -Pc) + (I1-a) (I1-e )]R = ITc since (because of the detection-cost advantage) Pcm > p1MN Thus a random trade with nonrnigrants will not take place. Since migrants reject trade with nonmigrants, non- migrants who may have attempted to engage a migrant in trade will be quicldy turned away. language, accent, color of sldn, and other similar traits are recognized cosdessly, correcdy, and immediately. We conclude then that the possibility of intergroup trade need not result in such a trade and hence that the migrants' edge is immune to this possibility. This last case assumes that agents are "hard-wired" as C or NC But what if agents who are C within their own group turn out to be NC when trading with outsiders? The answer is that the foregoing conclusion that trade will not take place holds a for- tiori: now the possible appeal that migrants may have to pursue trade with non- 384 Frontier Issues in International Migration migrants is even weaker since the actual PcM migrants would encounter upon trade would be lower. What if a reverse switch is allowed? In particular, consider the possibility that in order to facilitate trade with migrants, the NC-type nonmigrants will, upon trading with migrants, behave as if they were C-type. This switch cannot erode the migrants' edge either. To see why note that the migrants will now face a group of nonmigrants, all of whom are of C-type. By assuming an NC-type, the C migrants will derive U from a trade with a nonmigrant whose payoff will therefore be R. This is dearly worse than what the nonmigrants can obtain by trading with members of their own group. Hence such a scheme will not work. Notes 1. To make the analyses tractable we assume throughout that the wages in both R and P arc depcndent only upon a worker's siull level and not upon the excess supply of or demand for labor. In this we fol- low the similar assumption made in the optimal tax literature. Thus for example, W. (0) and W. (0) may be linear in e such that WR (6) = rO + rO, >0,r > 0 and Wp t() = po + PO, PO > 0, p > 0. It can be shown that these equations are reduccd equilibrium forms, where in each equation the left-hand side is the equilibrium wage and the right-hand side is the productivity of a worker with skill level 0 (Stark 1991, chap. 12). 2. This may, for example, resu.t from a higher capital-to-labor ratio in R, from superior technology in R, or from eeternalities arising from a higher average level of human capital per worker in R. 3. If employers are risk neutral and production functions are linear in skills, the employer does not suf- fer from his or her ignorance of the true skill level of each worker, so paying the average product per worker is the competitive outcome. These are the commonly accepted assumptions in the screening lit- erature (see, for example, Stiglitz 1975). 4. Equation 4 provides a cut-off condition that results from individual rationality. The arising cquilib- rium is compatible with, indeed ensues from, the other side of the market, namely, the bchavior of firms in the destination R (Stark 1991, chap. 12). S. Since O, and 62 are removed from the avcraging process, we can normalize 93 and 94 to constitute the [0,11 interval, and therefore W. (*) is fully defined as per equation 3. 6. This is Ravenstein's (1885) well-known law of migraton. Indeed, the analytically derived sequential pattrn of migration is also in line with Ravenstein's observation that migration streams have a built-in tendency to increase over time. 7. Borias (1987) provides evidence that the quality of migrant cohorts from Western Europe to the United States increased over 1955-79. However, his measures of quality are diffcrent from the one used in this artide. 8. The artidce's two-skill-level implementation of the asymmetric-information theory predicts that the nonmigrant household members will be low-skill workers. 9. For example, some studics find an inverse relationship between recipient income and remittance amounts (Kaufmann and Undauer 1986, El Salvador; Kaufnann 1982, the Philippines; and Ravallion and Dearden 198 8, rural households in Java). This finding is consistent with altruism; the poorer receive more transfcrs. But other studies find a positive relationship between recipient income and remittance amounts (Lucas and Stark 1985, Botswana; Ravallion and Dearden 1988, urban households in Java; and Cox and Jimenez 1991, urban households in Pcru). This finding is inconsistent with purely altruistic motives for remittan cs 10. For example, the correlation between potential and actual wages is likely to be lower if low-skil workers in country P are subject to binding minimum wages. 11. Migration enables agents to utilize a production technology specific to the country of destination that is superior to that for the country of origin (Galor and Stark 1991). Hence the bencfits to agents from migration are not conditional on migrants trading with nomnigrants. 12. By incurring cost 1, the C-type agent attains a trade with a C-type agent with probability 1. To see why, suppose the C-type agent announces his intention to undertakc the type-detcrmining action. Since this action determincs a type perfcctly, no NC-type agent will approach a C-ype agent, knowing that such a meeting will not result in a trade. The C-type knows that the NC-type knows this, which could tempt the C-type not to incur the cost after alL However, what works against such a temptation is the Stark 385 realization that any failure to pursuc type-determining could result in the NC-type approaching the C- type, which in turn will result in a trade that was considered undesirab!e wben the decision to incur K, rather than trade randomly, was taken. 13. For an explicit evolutionary exposition sec Bcrgstrom and Stark (1993). 14. Suppose that by incurring some costK the NC agents can identify the C agents in an attempt to trade with them rather than to trade randomly. But then the C agents will be rcluctant to trade randomly as this confers a payoff of R which is worse than n1c; the C agents will fare better by incurring K (and will reccivc a payoff flu). Thus invoking the assumption that the C agents 'move" first, the possibility of NC agents incurringR is negated. 15. The assumption that the payoff matrices of each of the subpopulations are the same can be rclaxed without affecting this result. Even if the payoffs to migrants from trade with fellow migrants arc system- atically lower than the analogous payoffs for nonmigrants, the recognition-cost edge could result in the per capita incomc of migrants dominating that of nonmigrants. 16. Ethnic minorities that concentrate in cthnic cnclaves and fare well may succeed not in spite of their concentration but bccause of it. 17. Interestingly, the studies reporting that migrants outperform the indigenous population point out that they do so only somc timc after arrival. Perhaps a time-consuming process of convergencc to an equi- librium Pc accounts for this result 18. We referred to ongoing rcsearch that attrbutes return migration to maximization of lifetime consump- tion which, in turn, leads to disposition at home of savings accumulated abroad. And we developed the argu- ment that retum migtion of low-skill workers arises from the reinstatement of symmetric information. These two explanations may be complementary. Presumably revelation of information is quicker than accu- mulation of savings. Hence in a return process that takces a long time to unravel, there will be an initial bout of information-induced return, followed by retum induced by consumption maximiation. LaUnde and Topcl (1993) report that in the United Statcs much of the total return migration occurs shortly after arrival, with the rest spread over as much as several decades (that is, until about one-third of the migrants return). References Bergstrom, Theodore C, and Odcd Star 1993. "How Altruism Can Prevail in an Evolutionary Environment." American Economic Review 83 (May): 149-55. Bloom, David E., and Morley Gunderson. 1991. "An Analysis of the Earnings of Canadian Immigrants." In John M. Abowd and Richard B. Freeman, eds., Immigrauion, Trak and the Labor Market Chicago: University of Chicago PNs Borjas, Georgc 1. 1925. 'Assimilation, Changes in Cohort Quality and the Earnings of Immigrants." Journal of Labor Economics 3 (October): 46349. . 1987. 'Self-Selection and the Earnings of Immigrants." American Economic Review 77 (September); 531-53. Borjas, Georgc J., and Bcrnt Bratsberg. 1992. 'Who Leaves? The Outmigration of the Foreign Born." University of California, San Diego. Chiswick, Barry R. 1986a. 'Human Capital and the Labor Market Adjustment of Immigrants: Testing Alternative Hypothcses." In Oded Stark, ed., Migration, Hwnan CapitaandDevelopment Greenwich, Conn.: JAI Press. - . 1986b. "Is the New Immigration Less Skilled than the Old?" Jouna of Labor Economics 4 (April): 168-92. Cox, Donald, and Emmanuel Jimcnez. 1991. 'Motives for Private Incomc Transfcrs Over the Life-Cycle: An Analytical Framework and Evidcncc for Peru." Boston College, Boston, Mass. Djajic, Slobodan, and Ross Milbourne. 1988. 'A General Equilbrium Model of Guest-Worker Migration: A Source-Country Perspective." Journal of Intemational Economics 25 (November): 335-51. Dustmann, Christian. 1993. "Retum Intentions of Temporary Migrants." Paper presentcd at the Seventh Annual Mceting of the European Society for Population Economics, Budapest Elbadawi, Ibrahim A, and Roberto de Rezede Rocha. 1992. 'Determinants of Expatriatc Workers' Remittances in North Africa and Europe." Policy Research Working Paper 1038. 'World Bank, Transition and Macro-Adjustment Division, Washington, D.C. Galor, Oded, and Oded Stark. 1991. "The Impact of Differcaces in the Level of Technology on Intenational Labor Migration." Journal of Populiaon Economics 4: 1-12. 386 Frontier Inues in Intermational Migration Hill, John IL 1987. "Immigrant Decisions Concerning Duration of Stay and Migration Frequency." journal of Development Economics 25: 221-34. Kaufmann, Daniel. 1932. "Sodda lnteraction as a Strategy for Survival Among the Urban Poor: A Thcory and Evidence." Ph.D. dissertation, Harvard University, Department of Economics, Cambridge, Mass. Kaufmann, Daniel, and David L. Lindaser. 1986. "A Model of Income Transfcrs for the Urban Poor." journal of Development Economics 22 (July/August): 337-S0. Kimball, Miles S. 1993. "Standard Risk Aversion." Econometrica 61 (May): 589-611. LaLonde, Robert J., and Robert H. Topel. 1993. "Economic Impact of International Migration: A Survey of Empirical Findings." University of Chicago, Department of Economics, Chicago, lL. Lucas, Robert EB., and Oded Stark. 1985. "Motivations to Remit: Evidence from Botswana." Journal of Political Economy 93 (October): 901-18. McCall, B.P, and J.J. McCall. 1987. "A Sequential Study of Migration and Job Search." Journal of Labor Economics 5 (October): 452-76. Miyagiwa, KaL 1991. "Scale Economies in Education and the Brain Drain Problem." International Economic Review 32 (August): 743-S9. OECD (Organization For Economic Cooperation and Development). 1990. Continuous Reporting System on Migration (SOPEMI). Paris: OECD, Directorate for Social Affairs, Manpowcr and Education. Polachek, Solomon W., and Francis W. Horvath. 1977. "A Life Cycic Approach to Migration: Analysis of the Perspicacious Peregrinator." In Ronald G. Ehrenberg, ed., Research in Labor Economics, vol. 1. Greenwich, Conn.: JAI Press. Pratt, John W., and Richard J. Zeckhauser. 1987. "Proper Risk Aversion." Econometrica 55 (January): 143-54. Ravallion, Martin, and Lorraine Dearden. 1988. "Social Sccurity in a 'Moral Economy': An Empirical Analysis for Java." Review of Economics and Statitics 70 (Fcbruary): 36-44. Ravensmein, Ernest George. 1885. "The Laws of Migration."Journal of the Royal Statistical Society 48 (June): 167-227. Reilly, Barry. 1994. "What Determines Migration and Return? An Individual Levcl Analysis Using Data for Ireland." Univcrsity of Sussex, Sussex, UK Stark, OdedL 1991. The Migration of Labor. Oxford and Cambridge, Mass.: Basil Blackwell. - . 199Sa. Altruism and Beyond An Economic Analysis of Transfers and Exchanges Within Families and Groups. The Oscar Mowr rstem Memorial Lectures. Cambridge: Cambridge University Press. 1. 995b. "The Path of Labor Migration Whcn Workcrs Differ in thcir Skills and Information Is Asymimetric." Sandinavian joul efEconomics. Forthcoming Stark, Oded, Christi;m Heimenstein, and Ytry Yegorov. 1994. "Migrants' Savings, Purchasing Power Parity, and the Optimal Duration of Migration." Institute for Advanccd Studies, Vicnna. Stiglitz, Joseph E. 1975. "The Theory of 'Screening,' Education, and the Distnrbution of Income." Amerian Economic Review 65 (June): 283-300. U.S. Department of Commerce, Bureau of the Census. 1983. General Social and Economic Characteristics, United States, 1980 Census of Population. Washington, D.C.: Government Printing Office. . 1984. General Social and Economic Charaaeristics, Puerto Rico, 1980 Cnsu of Popultion. Washington, D.C.: Government Printing Office. bWrld Bank. 1984. World Development Report 1984. New York: Oxford University Press.