The Impact of Formal and Informal Channels on Mexican Migrant Remittances

The Impact of Formal and Informal Channels on Mexican Migrant Remittances A Thesis submitted to the Graduate School of Arts & Sciences at Georgetown University in partial fulfillment of the requirements for the degree of Master of Public Policy [in the Georgetown Public Policy Institute]. By Kheng Mei Tan, B.A. Washington, DC April 14, 2006

The Impact of Formal and Informal Channels on Mexican Migrant Remittances Kheng Mei Tan, B.A. Thesis Advisor: Sarah Gormly, Ph.D. ABSTRACT As Mexico s remittances steadily rises, Mexican migration flow into the U.S. is decreasing. This indicates that migrants are remitting more money than in previous years. With the migrant demographic population unchanged, the exponential growth suggests that other forces are at work. This paper argues that the remittance growth is influenced by the development of the financial industry that has raised competition, reduced transfer fees, and opened new remittance services to allow for cheaper remittance transfers. Though the trend is encouraging to policymakers that support formal remittances, are migrants likely to send more money formally than informally? This study hypothesizes two things: 1) migrants remit more money through formal sectors and 2) among sending options, migrants remit the most money through financial institutions. This paper finds that migrants are likely to remit greater amounts informally, suggesting that transfer costs remain too high for migrants to spend on secured transfers. ii

Introduction Within the last decade, remittances grew rapidly, garnering widespread attention from policymakers and the developing world. In 2004, global remittances amounted to $125 billion, second to Foreign Direct Investment (FDI) and largely outpacing Official Development Aid (ODA). (Mahmood 2005) This was an increase of 73 percent from 2001, when remittances were $72.3 billion (Ratha 2003). As a result, its enormous growth led policymakers to promote remittances as a tool for development (Massey and Basem 1992; Ratha 2003; Cordova 2004; Yang 2004a; Adams 2005). Recent studies on remittances show that they positively impact human capital investments and economic outcomes (Meyer 1998; Durrand, Parrado and Massey 1996a; Woodruff and Zenteno 2001; Cordova 2004; Adams 2005). Human capital investments, such as those in health and education, are crucial in the long term since they create a healthier and more educated population to raise productivity. In tandem, economic growth, spurred by consumption and investments, is important to immediate and future production since it enhances jobs, incomes and development. Currently, the literature on estimating remittance behavior is fairly rich: several studies have determined that a migrant s age (Desipio 2000; Lowell, Bump and Fedewa 2005), gender (Lowell, Bump and Fedewa 2005), education (Desipio 2000), skills (Osili 2004), income (Desipio 2000; Osili 2004), number of dependents (Massey and Basem 1992; Despio 2000; Lowell, Bump and Fedewa 2005), wealth (Massey and Basem 1992; Osili 2004), time in the U.S. (Desipio 2000; Lowell, Bump and Fedewa 2005), and social networks (Massey and Basem 1992) establish the amount of money a migrant is 1

likely to send home. Yet, sparse in the literature are studies on formal and informal sending networks that are likely to influence remittance behavior. Thus, examining money channels are important to remittances since different channels can affect remittance behavior. For example, the cost to transfer money varies across money gram to postal services; and it is possible that migrants compute these costs when determining their remittances. From a policy perspective, the sum of money a migrant remits is important since empirical research demonstrates that remittances positively impact growth. Moreover, policymakers advocate for formal remittances since informal remittances are subject to financial abuse and mitigate capital transparency (Orozco 2003; Ratha 2004; Johnson and Sedaca 2004). Thus, they support policies to reduce transfer fees and increase banking accessibility to strengthen and incentivize formal channels. These policies are integral to raising migrant remittances, and benefit banks and credit unions that can profit from a large, untapped market. This thesis studies how remitting through formal and informal channels impacts migrant remittances. Formal channels are defined as money gram services, banking institutions, automatic tellers (ATMs), credit unions, or cash cards (pre-paid cards that allow migrants to remit money through participating ATM tellers). Informal channels are defined as remittances made through mail, or a relative. This thesis hypothesizes two things: 1) remittances sent through formal networks are greater than remittances sent through informal networks and 2) migrants are likely to remit more money through financial institutions than any of the other sending options. 2

In this study, I use data from the Pew Hispanic Center that surveyed migrants in seven Mexican consulates across the U.S. I find that migrants are likely to send less money through formal sectors, which suggest that current transfer fees in formal mechanisms remain too high. Moreover, the affects of financial institutions on remittances are not statistically significant, which might be due to the lack of randomness in the dataset used for the study. Section two reviews the current remittance studies and findings on sending options. Section three explains the conceptual model. Section four and five summarize the plan of analysis and model limitations, respectively. I discuss my results in section six. In the final sections, I provide the implications of my study and concluding thoughts. Literature Review Currently, the literature on estimating remittance behavior is fairly rich. Previous studies determine that a migrant s age (Desipio 2000; Lowell, Bump and Fedewa 2005), gender (Lowell, Bump and Fedewa 2005), education (Desipio 2000), skills (Osili 2004), income (Desipio 2000; Osili 2004), number of dependents (Massey and Basem 1992; Despio 2000; Lowell, Bump and Fedewa 2005), wealth (Massey and Basem 1992; Osili 2004), time in the U.S. (Desipio 2000; Lowell, Bump and Fedewa 2005), and social networks (Massey and Basem 1992) establish the amount of money a migrant sends home (Massey and Basem 1992; Desipio 2000; Orozco 2003; Osili 2004; Orozco, Lowell, Bump and Fedewa 2005). Moreover, theories of altruism (Maggard 2004; Osili 2004) also explain remittance behavior. Yet, literature on formal and informal networks relative to remittances is sparse. Still, policymakers agree that sending channels impact 3

remittances (Orozco 2003; Suro, Bendixen, Lowell and Benavides 2005); and presently, there is one paper by Amuedo-Dorantes and Pozo (2002) that evaluates sending options. The current literature on sending networks is primarily descriptive, and is based on smaller, migrant surveys and compilation of sending information for exploratory research. Overall, what is consistent in the literature is that accessibility, familiarity and information from social networks determine the sending method used to remit money (Amuedo-Dorantes and Pozo 2002; Orozco 2003; Suro, Bendixen, Lowell and Benavides 2005). For example, even with the recent entry of banks and credit unions that enable cheaper, sending options, barriers, like language, documentation and unfamiliarity prevent more migrants from utilizing these services (Orozco 2003; Amuedo-Dorates and Pozo 2002). Though formal channels offer regulation and security, some migrants remit informally because of high transfer cost and lack of English skills (Amuedo-Dorantes and Pozo 2002; Ratha 2004). This indicates that migrants lacking in income, education and English skills are likely to remit informally. Thus, how migrants choose to send their money possibly can impact their remittance amounts. When examining sending methods, money gram services overwhelmingly dominate, followed distantly by banks, credit unions and informal channels (IADB 2004). This is due to the expansion of money gram services in the 1990s that enabled migrants to remit money to most areas in Mexico. As their popularity and accessibility increased, services like Western Union captured a large number of once, informal remitters who had no other sending option (Orozco 2003). Though remittance channels are more formalized today, migrants use these services due to their familiarity, comfort 4

or unnecessary documentation (Orozco 2003; Ratha 2004). Unfortunately, wire transfer services, among all formal sending options, are the costliest. In a paper by Suro, Bendixen, Lowell and Benavides (2005), they estimate that a 5 percent reduction in transfer fees would raise $1 billion for households, arguing that sending channels greatly impact remittance amounts. Thus, migrants that utilize money gram providers, instead of financial institutions, are likely to remit less money due to their transfer costs. Time in the U.S. also can affect the use of various remittance methods (Amuedo- Dorates and Pozo 2002). For example, though the recent technology of cash cards provides cheaper remittances than wire transfers, they are unknown to migrants who primarily rely on social networks for their information (Orozco 2003). A recent study on sending options by Amuedo-Dorantes and Pozo (2002) demonstrates a decline in wire transfer usage over time. In their descriptive analyses of migrant data from the years 1993-2000, they show a decline of informal and wire transfer remitters by 14 percent and 22 percent over the seven years, respectively. Conversely, the number of migrants remitting through banks increased to 338 percent. This indicates that migrants are more incline to utilize other sending methods with increasing knowledge. In the same study, Amuedo-Dorantes and Pozo estimated the impact of remittances on sending options using a multinomial logit. They found that migrants who remitted through the banking system sent $38 more than migrants who remitted using wire transfer services, arguing that the security of banks enabled higher remittances. Surprisingly, they found that informal and wire transfer remitters sent on average, a similar amount. Though they offered no explanation, this might be due to financial 5

accessibility and familiarity of other sending options. Thus, with exception of wire transfer services, these migrants might be unfamiliar with other sending methods, and remit informally since wire transfer services do not reach their homes in Mexico. Unlike Amuedo-Dorantes and Pozo s paper that examines the choice of sending methods as dependent on the level of remittances, my thesis adds to current, remittance studies by estimating remittances as dependent on formal and informal networks. By constructing my model to explain remittances, I can demonstrate how sending channels influence remittance amounts. Thus, I can determine if there is a relationship between remittance growth and sending networks. Conceptual Model There are three reasons why evaluating formal and informal streams are important to remittances: 1) remittances, by basic mechanics, are transmitted through one of these channels where a migrant can choose from several, sending options, 2) remittances have grown in Mexico despite a decline in migration which suggests that something else is at force and 3) the recent change in financial markets has raised competition, reduced transfer fees, expanded the number of banks, and raised the number of remittance services (Orozco 2003; FDIC 2004; Mahmood 2005; TAPARIA 2005). Firstly, migrants utilize formal and informal streams to remit money, where both options have associated tradeoffs between security and price. For example, by remitting formally, migrants are subjected to transfer fees, but have security that their money reaches their intended recipient. Moreover, within formal channels, migrants can choose 6

from several services that range in prices. Money gram providers, for instance, are the costliest, while banking services are among the least expensive (Meyer 1998; Orozco 2003; FDIC 2004). Conversely, by remitting informally, migrants may pay less in fees, but have no security that their money arrives to their household. With these tradeoffs, it is possible that migrants, given their demographic traits, such as income, education and age, consider these costs when remitting money. Secondly, the negative relationship between aggregate remittances and migration streams suggest that sending options impact remittance behavior. The current growth rate of Mexico s remittances is rising faster than the U.S. migration population as shown in Table 1. Table 1: 1999 2000 Comparison of Population and Remittance Growth Rates and Percent Changes in Mexican Migrant Flows Year Population Pop % Change Remittance Remit % Change 1999 496-2% 5.9 -- 2000 530 7% 6.6 12% 2001 437-18% 8.9 35% 2002 378-14% 9.8 10% 2003 369-2% 13.4 37% 2004 459 24% 16.6 24% Compiled from IADB (2005) and Passel and Suro (2005) Table 1 compares the growth rates of the U.S. Mexican migrant population to Mexico s remittances for the years 1999-2004. On net, the migration growth rate is declining over time compared to total remittances, which are steadily rising (even after inflation). This means that migrants are remitting more money each year. 7

Thus, what is causing this remittance growth? Before proposing that it is the recent change in the financial industry that is raising remittances, I exhaust two, possible explanations for this trend: 1) existing migrants in the U.S. are still remitting home, and thus, their remittances are reflected in the sum and 2) current migrants are better educated, and therefore, secure higher wage jobs that allow them to send more money home. Though both explanations are possible, current empirical findings suggest that their occurrences are not likely (Orozco 2003; Suro, Bendixen, Lowell and Benavides 2005). Orozco (2003), in examining remittance patterns, finds that on average, the extended duration of a migrant stay is negatively related to his remittance transfer. Thus, as a migrant assimilates into the U.S., his liabilities such as personal investments, living expenses and commitment to stay in the U.S. increases with time, while his remittances to Mexico decline. Further, he finds that after ten years in the U.S., migrants remit below-average amounts. Though it is likely that such existing migrants still remit home, their transfers might not be substantial enough to account for the growth. Secondly, though current migrants might be better educated, and thus, generate higher incomes to send more money home, in evaluating these demographic traits, a descriptive study on current Mexican migration by Suro, Bendixen, Lowell and Benavides (2005) reports otherwise. They find that the demographics of Mexican migrants remain largely unchanged, where the bulk of migrants are poor, low educated and low-skilled. Thus, present demographic traits do not explain the rise in remittances. 8

A feasible explanation for the remittance growth is that the recent change in financial institutions has raised competition, reduced transfer fees and increased remittance services within the last, several years. This large inflow of remittances not only captured the attention of policymakers, but the banking industry as well (FDIC 2004; BANSEFI 2005). During the last, several decades, money gram services, like Western Union, monopolized and profited from the remittance market that were once unknown to banks and credit unions. Unlike wire transfers, migrants without proper identification were precluded from accessing these institutions (Orozco 2003). But, in 2002, the U.S. Treasury Department allowed the Mexican Consulate to issue ID cards to certify the Mexican nationality of migrants entering the U.S (TAPARIA 2004). The U.S. endorsement expanded banking services to migrants and, since 2004, these ID cards were recognized by 178 banks including Wells Fargo, Citibank, and Bank of America in 33 states (TAPARIA 2004). With 80 percent of remittances from migrants without legal status, these cards enabled them access to financial institutions and sending methods (TAPARIA 2004). With the ID cards, migrants received a variety of competitive, wire transfer services that ranged from electronic wiring, ATMs and cash cards from banks and credit unions (Orozco 2003; FDIC 2004). Moreover, large banks, like Wells Fargo, have purchased Mexican banks to open financial accessibility between the U.S. and Mexico. (Orozco 2003; FDIC 2004; BANSEFI 2005) 9

In sum, I believe that the large remittance growth is explained by the recent change in financial climate. Thus, I want to answer two questions in this study: 1) Do informal and formal networks significantly contribute to the amount a migrant chooses to remit and 2) what forms of sending options enable a migrant to remit the most money? I hypothesize that formal and informal networks affect remittances, where migrants remit more money through formal streams. I believe that a migrant s decision to remit involves a trade off between costs and security, where remitting formally, versus informally, has greater financial costs, but is more secure than remitting informally. Thus, with the decline in transfer fees, a migrant will remit more money through formal channels since he saves in sending costs and is secured that his money reaches the intended recipient. Comparatively, some migrants still may choose to remit informally. For instance, a migrant may live in an area that banks do not reach. Here, a migrant might send less remittances because he is less sure that it would reach his household of origin. Secondly, I am also interested in determining how different sending options impact migrant remittances. I hypothesize that migrants who remit through financial institution services, such as banks, credit unions, ATMs or cash cards, remit more money than through informal and money gram channels since they value the security provided by these institutions. I believe that with the entry of the financial sector that typically offer cheaper remittance services than money gram providers, their transfer costs are low enough for migrants to remit through these institutions for greater security. Methodology 10

The dataset used to estimate my model is publicly available from the Pew Hispanic Center that conducted random surveys across seven Mexican Consulates in Los Angeles, Fresno, Dallas, Chicago, New York, Atlanta, and Raleigh from July 12, 2004 to January 28, 2005. Respondents applying for an ID card were asked about basic, demographic characteristics and their remittance practices. Because the purpose of the consulates is to issue ID cards, this means that the population surveyed is undocumented workers. The strength of this dataset, in comparison to others, is in its descriptive questions on sending methods that enable formal and informal channels to be studied. My sample is of the remitting population. Out of 4,836 Mexicans surveyed, there were 3,794 remitters. The missing observations among my independent variables were not on average, statistically different from the means of the valid independent variables. Thus, dropping the missing observations would not significantly impact my model. This left 1,067 valid observations of undocumented remitters. To estimate the relationship between remittance amounts, and formal and informal channels, the literature on migrant remittance behavior overwhelmingly uses probit models (Massey and Basem 1992; Desipio 2000; Maggard 2004; Osili 2004; Orozco, Lowell, Bump and Fedewa 2005). My dependent variable is defined as the level of remittances a migrant sends home. The Pew survey does not asks migrants to specify their remittance amounts. Rather, it asks migrants to check off one of six categories that vary from $0 to over $500, where each group was partitioned by $100 increments. Thus, the dependent 11

variable is not continuous. To estimate my model, I create the remittance variable as an indicator to equal 1 if a migrant remitted over $300, or 0 if otherwise. I chose the $300 cutoff point from a working paper by Orozco (2003) that compared international remittances. Orozco found that the average Mexican remittance returned home was over $300. In comparison to my dataset, this amount was close to the mode at 2.77, where 3.00 represented migrants who remitted greater than $300. Thus, I believed that it was a reasonable cutoff point. Before proceeding, I note that there are limits to estimating my model. The distribution of remitters for my cutoff is skewed: only 25 percent of migrants in the sample remit greater than $300. This means that the model can have attenuation bias where the coefficients are underestimated. Further, by imposing a discrete cutoff in my dependent variable, I only capture the difference between remittances greater and less than $300, thus, limiting the information that I can interpret from the regression. Still, the model allows for the study of formal and informal sectors on remittances. In the following equation, I estimate the probability that a migrant remits above the $300 average (where a description of the variable names are in Table D): (Remit>300) = Ф β 0 + β 1 Tradregion + β 2 Female + β 3 Age + β 4 Married + β 5 lessthanhsgrad + β 6 Secondary + β 7 AvgEnglishSkills + β 8 SpouseUS + β 9 KidsUS + β 10 Ownership + β 11 incomeless199 + β 12 inc201 + β 13 inc301+ β 14 inc401 + β 15 First_US + β 16 ExpectedstayUS_12m + β 17 ExpectedstayUS_1_5yrs + β 18 ExpectedstayUS_over5y + β 19 Rel_inUS + β 20 Socialclub + β 21 Formal 12

My main explanatory variables of interest are formal and informal channels. In my model, I plan to estimate my main explanatory variables two ways. In the above model, I collapse migrants who remit formally (through any of the formal, sending options) into one indicator variable, where those who remit formally are represented by 1, and those who remit informally are indicated as 0. In a second model, I create indicators for each of the formal sending options where the baseline is informal remitters. Thus, in the first model, I want to establish a general relationship between sending channels and remittance amounts. I expect the coefficient sign for the formal variable to be positive because I believe that migrants who remit formally send more money due to greater security. Thus, its coefficient states that formal remitters will have a probability of remitting greater than $300 by X percentage point versus informal remitters. In the second model, I examine the impact of money gram providers and financial institutions separately. I categorize money gram as its own indicator, and collapse bank, credit union, ATM and cash card into the financial institution indicator. The baseline category is informal remitters who transfer funds using mail or a relative. (Remit>300) = Ф β 0 + β 1 Tradregion + β 2 Female + β 3 Age + β 4 Married + β 5 lessthanhsgrad + β 6 Secondary + β 7 AvgEnglishSkills + β 8 SpouseUS + β 9 KidsUS + β 10 Ownership + β 11 incomeless199 + β 12 inc201 + β 13 inc301+ β 14 inc401 + β 15 First_US + β 16 ExpectedstayUS_12m + β 17 ExpectedstayUS_1_5yrs + β 18 ExpectedstayUS_over5y + β 19 Rel_inUS + β 20 Socialclub + β 21 MoneyGram + β 22 FinancialInstitutions 13

Here, I expect all of the coefficient signs for formal sending options to be positive since I believe that they raise the amount a migrant remits relative to informal channels. Moreover, I also anticipate that migrants who remit through financial institutions send more money compared to any of the other sending options because of its heightened security. The coefficients for each of the sending options are interpreted with respect to informal channels. For example, migrants who remit through money gram providers have a probability of remitting greater than $300 by X percentage point over migrants who use informal sectors. My probit model also contains typical control variables found in the literature for evaluating remittance behavior (Massey and Basem 1992; Desipio 2000; Osili 2004; Lowell, Bump and Fedewa 2005). The control variables are sectioned by demographics, financial literacy, number of dependents, asset levels, time sensitivity and network connections. Key demographic variables, like gender, age, and marital status are controls for responsibility. I expect the signs to be positive. Females, for instance, may remit more money because of stronger family ties. Migrants who are older and married might have more responsibilities than those unmarried and younger. Moreover, the type of region that a migrant comes from also impacts remittances. For example, migrants from traditional sending regions such as Jalisco, Michoacán and Zacatecas may remit more money than migrants from other regions because they have a culture of migration and remittances. 14

The variables for financial literacy are education and English skills. These variables are controls for financial access which can impact migrant remittances. I expect the coefficient signs for education and English skills to be positive since greater education and English skills raises a migrant s ability to remit more money. The number of dependents, such as a spouse or children, a migrant has with him in the U.S. also affects remittances. I expect the coefficient signs for these variables to be negative because a migrant with dependents living with him in the U.S. would remit less money since he also utilizes his income to take care of his family. A migrant s level of assets in Mexico also positively impacts remittances. Assets like property, land, and businesses require maintenance capital, and greater remittances. Thus, I expect that migrants owning property will remit more money home. Time sensitivity also affects remittances. Migrants who first visit the U.S. might remit less money because of initial set-up costs to get acclimated. Moreover, Orozco (2004) finds that migrants who stay for shorter periods remit less than migrants who stay for longer periods. This is because migrants that expect their imminent return are likely to go home with a lump sum transfer, rather than pay for remittance services. And finally, social networks positively affect a migrant s remittance amount since they can reduce living expenses and allow a migrant to save more money. Moreover, these social networks can facilitate information on sending methods. Model Limitations 15

Though the dataset contains information on sending options, it has several limitations. Three issues can impact my regression results: selection bias, data credibility, and probit modeling. I address them respectively. Firstly, the survey samples migrants from Mexican consulates in the U.S. The purpose of the consulate is to issue ID cards to undocumented migrants who use them for financial access. Thus, this is not a representative sample, and there might be a selection bias where migrants that obtain these ID cards may place more value on remitting through financial institutions. This means that my variable for financial institutions may result in statistical significance, but it would reflect a bias towards the use of banks and credit unions. Secondly, there may be some data credibility concerns. What is interesting about the survey is that it includes migrants that remit through banks and credit unions. This is unexpected since migrants applying to the consulate are undocumented, and thus, should not be able to access financial institutions. Yet, 11.4 percent and 4 percent of the sample show migrants remitting through banks and credit unions, respectively. Here, three possibilities exist: 1) the respondents may be untruthful 2) the respondents misunderstood the question or 3) the respondents were able to obtain access through relatives with proper documentation. Because I do not know how true either explanation is, I compare my sample of remitters to other institutional findings to determine statistical differences. 16

The results of a 2003 survey on formal and informal U.S. Latino migrant remitters conducted by the Multilateral Investment Fund (MIF) and Pew Hispanic Center demonstrate similarities to my sample population as shown in Table 2. Table 2: Sending Options Compared Across MIF and Pew Survey Sending Option 2003 Survey Pew Dataset Money Gram 70% 74% Bank 11% 11.4% Credit Union 2% 1.3% Mail 7% 1.7% Relative/By Hand 1% 7.6% IADB 2004 From Table 2, there are not large differences between my dataset and the 2003 survey among remitters using money gram, banking and credit union services. Among informal channels, the results are reversed: the 2003 survey show more migrants remitting through mail, and less through relatives than in my dataset. Still, the aggregate sum of informal remitters in the 2003 survey is 8 percent versus 9.3 percent in my dataset. Moreover, because my paper is focused on specific formal sending options versus informal channels (which, recall, is collapsed into one indicator), I am not as concerned by the informal differences. Still, because the survey samples Latin migrants instead of only Mexican migrants, there might be other demographic traits that drive these results. Finally, due to data constraints, I use a probit model to estimate the effects of sending options on the amount remitted, where my dependent variable is based on a cutoff of $300. Thus, my results are interpreted with respect to the cutoff point. Here, two issues arise: 1) though I chose the cutoff using Orozco s paper, the $300 marker is still subjective and 2) this produces vagueness in the results interpretation. 17

In choosing the mode marker of $300, I am theorizing that if sending options impact remittances, then, at the very least, these sending options should enable migrants to remit greater than the average amount. Still, I do not know what the true mean is of the Mexican migrant remitting population since 1) I had previously established that remittances were growing, which implies a dynamic mean and 2) that, even with randomness in my dataset, there is a chance that I am off the true mean, which impacts the results interpretation. Moreover, in creating a discrete cutoff, I am grouping my remittances into categories of high and low values or those above or below $300, which loses information. Thus, this leaves some vagueness in the analyses. Unlike in a model with a continuous dependent variable where I can determine the marginal impact to every dollar remitted through a sending channel, this model estimates the probability of a migrant remitting greater than $300, which weakens my results. Empirical Findings Table A and B present the descriptive results for formal and informal remitters who remit at least $300 or more home. The columns show the variable means, standard deviations and mean differences. I will address the descriptive findings before discussing the regression results. Table A presents the sample means of formal and informal remitters who remit at least $300 home. Because I hypothesize that migrants remitting formally send more money than migrants who remit informally, I expect to see mean differences in my financial literacy and asset variables. I believe education, English ability, physical and 18

financial capital are important skills that enable migrants to access and understand formal institutions. Instead, the means among both forms of remitters are not statistically different with exception of the income range of $300 to $399, and intention of stay for 1-5 years. Formal remitters are less likely to migrate from tradition migratory regions than informal remitters, and are slightly older than their counterparts. Moreover, formal remitters are better educated, bring less children in the U.S., earn higher incomes, expect to stay for longer periods and are less active in social clubs. This intuitively makes sense since I expect formal remitters to have improved financial literacy and higher incomes that would allow them to stay for longer periods in the U.S. and rely less on social networks. The explanation for formal remitters having fewer children in the U.S. is less intuitive. It might be that these migrants with higher education and income value education and want to keep their children in Mexico for school, or have relatives caring for their children. I note that the means for these variables are not statistically different. This might be due to the sample population of only undocumented workers. Thus, there is no variation in my sample among documented and undocumented migrants who can have distinct traits. Because I am examining the variances within an undocumented population, the mean disparities may not be wide enough to show statistical differences. In comparing the mean differences between formal and informal remitters, only income of $300-$399 per week and intention of stay for 1-5 years are statistically different at the 1 percent and 10 percent levels, respectively. 19

The income variable shows more formal remitters earning a weekly income of $300-$399 than informal remitters. It is confusing that this income range is the only limit that differed significantly. This might be due to the higher percentage of migrants in the sample who earn $300-$399 per week. Thus, because this income range is capturing a larger proportion of the sample, it is likely that the variation between formal and informal remitters differs, and results in a statistical difference in the means. The intention to stay between 1-5 years differs significantly among remitters, where a smaller proportion of formal remitters intended to stay for this length of time. This coincides with previous literature that explains that migrants with lower education and skills are unable to remain in the U.S. for extended periods due to cost burdens. Moreover, in evaluating the time variables, I observe that more formal remitters (with improved education and work skills) intended to stay longer than 5 years. Table B presents the sample means of remitters who remit at least $300 home grouped by sending options. Because I hypothesize that migrants who remit through financial institutions send the most money, I want to determine if there are mean differences among remitters using different sending methods. Similar to Table A, there are no mean differences among senders with exception of the income range of $300- $399 per week and social club that are statistically different at the 5 percent level. The income of $300-$300 remained statistically different, but now beyond the 5 percent level. The social club variable also differed significantly among the three sending remitters. Thus, the role of social club participation impacts various sending remitters differently. 20

Table C presents the probit regressions at the $300 cutoff, where columns 1-5 are variations of my model. Recall that I hypothesize two things: 1) remittances sent through formal networks are greater than remittances sent through informal networks and 2) migrants are likely to remit more money through financial institutions than money gram providers. To demonstrate that formal channels and financial institutions enable migrants to remit more money, I assume that they must remit greater than the mode of $300 in my sample. Column 1 and 2 are the reduced-forms of my model, column 3 and 4 are models with the formal channel indicator, and column 5 is my model separating formal channels into money gram and financial institution services. The results report the marginal effects and robust standard errors. Column 3 addresses my first hypothesis that formal channels enable migrants to remit more money. Recall that I define formal channels as migrants that remit through money grams, banks, credit unions, ATMs and cash card services. In comparison to column 1, I observe that the χ2 substantially changes when formal channel is added to the model, and is jointly, statistically significant beyond the 1% level. Surprisingly, the sign of this coefficient is negative. Thus, migrants utilizing formal channels are 8.05 percentage points less likely to remit greater than $300 than migrants utilizing informal channels. This might indicate that migrants who utilize formal means remit less money due to high transfer fees. Though transfer costs have reduced substantially within the past years, perhaps their reductions are not significant enough for migrants to pay for secured services instead of remitting through, free 21

informal networks. Thus, this implies that migrants who must utilize formal means to remit money are negatively impacted by these institutional costs. Column 5 addresses my second hypothesis that financial institution services enable migrants to remit more money than money gram providers. In column 5, I separate formal channels into money gram and financial institutions services. The baseline category is informal channels represented by migrants who remit through the mail or a relative. In comparison to my reduced-form model, the χ2 in column 5 significantly changes and is jointly significant beyond the 1% level. Surprisingly, only money gram was significant. Thus, migrants remitting through money gram services are 8.58 percentage points less likely to remit greater than $300 in comparison to migrants who remit through informal channels. This corresponds to my conceptual theory of migrant tradeoffs between transfer costs and informal channels. Though transfer fees have reduced substantially, the results suggest that they remain too high for many migrants who prefer to use free, informal channels. Thus, because of the added, remittance costs, migrants are able to send less money home in comparison to informal means. Moreover, the marginal effect of the financial institution variable was statistically insignificant. This might be related to my sample population. Recall in my Model Limitations discussion that my sample was base on undocumented migrants applying for ID cards at the Mexican Consulates. Without the ID cards, migrants were unable to access financial institutions such as banks and credit unions. Yet, 16 percent of my observations reported that prior to obtaining their ID cards, they remitted through banks 22

and credit union services. Thus, the insignificance of my financial institution variable might be impacted by my sample; and this variable might be robust if my data was based on the same migrants a year after they obtained their IDs. In examining all five models, consistently significant were the explanatory variables female, age, spouse in the US, asset ownership and income. Moreover, the coefficients and their respective signs were stable, even when variables were removed. The χ2 in all five models were significant beyond the 1 percent level. Columns 1 and 2 represented two forms of my basic model, where, in Column 1, I excluded the intended length of stay. This variable was used to proxy the actual length of stay a migrant remained in the U.S. to capture their assimilation. Previous studies showed that the length of time a migrant remained in the U.S. impacted their remittance behavior. The longer a migrant stayed, the more assimilated and attached he was to a country, and the less money he sent home. But in comparing the five models, the intent to stay was a poor proxy for the actual time of stay. From the χ2, I observed that when they were included, these variables did not contribute much to my model; and their marginal effects were statistically insignificant. Moreover, part of the assimilation traits were captured in my other time variable First US Visit, an indicator variable denoting whether it was a migrant s first visit to the U.S. Here, new migrants were assumed to have not assimilated, while experienced migrants had partly assimilated to the U.S. Thus, I chose to exclude intent to stay from my model; and the rest of my explanations addressed the results without the intent of stay variables. 23

Beginning with the demographic variables, in my reduced-form model in column 1, females are 6.6 percentage points less likely to remit at least $300 or more in comparison to males, controlling for all the variables at their means. This coincides with previous literature that females tend to remit less money than male migrants. Moreover, as expected, age positively affects remittances. Thus, a one year increase in age leads to a 0.30 percentage point chance that a migrant remits greater than $300. Dependents in the U.S. negatively impact remittances. Migrants with a spouse in the U.S. are 8.82 percentage points less likely to remit greater than $300 than migrants without a spouse in the U.S. Finally, financial assets positively influence remittance behavior. Migrants who own property are 8.93 percentage points more likely to remit greater than $300 compared to migrants who do not own property. Moreover, my income variables are all significant beyond the 1% level. The baseline for my income variables is income over $500. Since they are all interpreted the same way, I will interpret the first income range of weekly earnings less than $199. Thus, migrants who earn less than $199 per week are 23.17 percentage points less likely to remit greater than $300 compared to migrants earning greater than $500. Surprisingly, migrant children in the U.S. are not statistically significant. This does not make sense since having a spouse in the U.S. is important, and having children is not vital. A possible explanation might be in the lack of randomness in my sample that surveyed just seven Mexican consulates. Though the survey was taken randomly, my 24

sample is not a proportionate representation of Mexican migrants. A more balanced survey might show that children in the U.S. do affect remittances. Policy Implication This study finds that formal channels and money gram services negatively impact remittances relative to informal channels, suggesting that current transfer fees are too high for migrants. Thus, for migrants to choose to remit formally rather than informally, financial sectors must incentivize migrants by further reducing transfer fees. If a migrant s choice in sending options is related to his trade off between transfer cost and security, whereby, formally he pays a fee for a secure transfer, while informally he pays no fee and has a much less, secure transfer, financial sectors must reduce these costs to where it is feasible for a migrant to remit formally. Beyond this change, migrants have no incentive to act otherwise. Policymakers are correct in arguing for improved formal channels to raise remittance transparency and to reduce the insecure use of informal networks. The findings of this study suggest that achieving this goal requires further incentive schemes by financial sectors to enable migrants to remit formally. For migrants, formal channels are not the consistent choice. Though transfer fees have reduced in the last several years an optimistic trend for policymakers they remain too high for migrants who value the costless transfer of informal mechanisms. Though my study finds that financial institutions insignificantly impact remittance behavior, I believe that the findings are skewed by my sample. Therefore, this does not mean that financial institutions have little or no role in impacting remittances: if 25

anything, financial institutions gain from opening access to migrants since they can profit heavily from a large, untapped market in need of services. Migrants then benefit through lower transfer fees and improved financial services. Thus, future studies should collect remittance information across migrants with and without banking access to determine how remittances are influenced by financial markets. Conclusion As worldwide remittances grow, they become more important to development. Current studies on the impact of remittances to economic growth demonstrate that remittances raise health, education and productivity outcomes for poor nations. Thus, for impoverished countries like Mexico, remittances are crucial to household and aggregate sustainability. Yet, an area less studied that also impacts remittance growth is formal and informal channels. Policymakers want to strengthen formal sectors because informal mechanisms are abused and less transparent. With formal remittances amounting to $16.4 billion for Mexico (Ratha 2003), policymakers speculate that informal remittances are as substantial. Thus, the desire to reduce informal remittances is a valid concern. In examining the impact of both channels on remittances, this thesis finds that migrants are likely to remit less money formally. This implies that, at least for now, the cost of remitting formally remain too great for poor migrants. Still, policymakers are correct in wanting to strengthen financial access to migrants since it enhances remittance security. But for migrants to utilize formal streams, expenses must be low enough that the gains of security outweigh the cost of the transfer. Here, financial institutions have 26

most to benefit in reducing transfer fees since it would mean access and profit from an untapped market, while migrants gain by receiving lower costs and greater security in their transfers. 27

Bibliography 1. Adams, R. (2005, March). Remittance, Household Expenditure and Investment in Guatemala (Working Paper Series 3532) Washington, D.C.: World Bank Policy Research. Retrieved November 1, 2005, from http://www-wds.worldbank.org/ servlet/wdscontentserver/wdsp/ib/2005/03/04/000012009_20050304091807/ Original/wps3532.doc 2. Banco Del Chorro Nacional Servicios Financieros, SNC. (2005, June). Remittance Markets and the Social Bank Power Point Presentation. Retrieved January 1, 2006, from http://www.iadb.org/mif/v2/files/gavito_remitforum05.pdf 3. Cordova, E.L. (2004, October 1). Globalization, Migration and Development: The Role of Mexican Migrant Remittances Retrieved October 1, 2005, from http://www.ceresuy.org/rin2004/pdfs/globalization_migration _and_development_the_role_of_mexican_migrant_remittances_ernesto_lop ez_cordova.pdf#search='globalization%2c%20migration%20and%20develop ment%3a%20the%20role%20of%20mexican%20migrant%20remittances' 4. Desipio, L. (2000, January). Sending Money Home For Now: Remittances and Immigration Adaptation in the United States Washington, D.C.: Inter-American Dialogue. 5. Durand, J., Massey, D. & Zenteno, R. (2001). Mexican Immigration to the United States: Continuities and Changes Latin American Research Review 30(1), 107-127 6. Durand, J., Parrado, E. & Massey, D. (1996a). Migradollars and Development: A Reconsideration of the Mexican Case International Migration Review 30(2), 423-444 7. Federal Deposit Insurance Corporation (2004, December). Linking International Remittance Flows to Financial Services: Tapping the Latino Immigrant Market Retrieved January 1, 2006, from http://www.fdic.gov/regulations/examinations/ supervisory/insights/siwin04/latino_mkt.html 8. Inter-American Development Bank (2004, May). Sending Money Home: Remittance to Latin America and the Caribbean Retrieved January 1, 2006, from http://www.iadb.org/mif/v2/files/studype2004eng.pdf 9. Inter-American Development Bank (2005, June 28). Bank Remittances and Development: The Case of Mexico Powerpoint Presentation. Retrieved 28

November 29, 2005, from http://www.iadb.org/mif/v2/files /guemezremitforum05.pdf 10. Johnson, B. & Sedaca, S. (2004, March). Diasporas, Emigres and Development Economic Linkages and Programmatic Responses Washington, D.C.: United States Agency for International Development 11. Maggard, K.Q. (2004, November). The Role of Social Capital in the Remittance Decisions of Mexican Migrants from 1969 to 2000. (Working Paper 2004-29) Atlanta, Georgia: Federal Reserve Bank 12. Mahmood, J. (2005, July 7). Development Impact of Remittances. The Nation Retrieved December 12, 2005, from http://www.nation.com.pk/daily/july- 2005/11/bnews6.php 13. Massey, D. & Basem, L. (1992). Determinants of Savings, Remittances, and Spending Patterns among Mexican Migrants to the United States Sociological Inquiry 62(2), 186-207 14. Meyer, D.W. (1998, May). Migrant Remittances to Latin America: Reviewing the Literature Washington, D.C.: Inter-American Dialog 15. Orozco, M. (2003). Worker Remittances: An International Comparison Washington, D.C.: Multilateral Investment Fund of the Inter-American Development Bank 16. Orozco, M. Cost, Economic Identity and Banking the Unbanked. Testimony presented before the Congressional Hispanic Caucus, Washington, D.C. (26, March 2003). Retrieved January 1, 2006, from http://www.the dialogue.org/publications/country_studies/remittances/econ_identity.pdf 17. Orozo, M., Lowell, L.B., Bump, M., & Fedewa, R. (2005, July). Transnational Engagement, Remittance and their Relationship to Development in Latin America and the Caribbean. Washington, D.C.: Institute for the Study of International Migration 18. Osili, U.O. (2004). Remittance and Savings from International Migration: Theory and Evidence Using a Match Sample. Indianapolis, IN: Perdue University 19. Passel, J. & Suro, R. (2005, September). Rise, Peak, and Decline : Trends in U.S. Immigration 1992-2004 Washington, D.C.: Pew Hispanic Center 29

20. Ratha, D. (2003). Worker Remittances: An Important Source and Stable Finance for Development. Washington, D.C.: Asian Development Bank Bank Retrieved November 29, 2005, from http://siteresources.worldbank.org/intrgdf/ Resources/GDF2003-Chapter7.pdf 21. Ratha, D. (2004). Understanding the Importance of Remittances Washington, D.C.: Migration Policy Institute Retrieved January 1, 2006, from http://www.migrationinformation.org/feature/display.cfm?id=256 22. Suro, R., Bendixen, S., Lowell, L. & Benavides, D. (2005, June). Billions in Motion: Latino Immigrants, Remittances and Banking Washington, D.C.: Pew Hispanic Center and the Multilateral Investment Fund 23. Taparia, V. (2005, April). Banking Across Borders Texas Business Review Retrieved January 1, 2006, from http://utopia.utexas.edu/articles/tbr/ remittance.html?sec=business&sub=finance 24. Woodruff, C. & Zenteno, R. (2001, August). Remittances and Microenterprises in Mexico San Diego, CA: University of California in San Diego Retrieved September 26, 2005, from http://sccie.ucsc.edu/documents/working_papers/2001 /Remittances_Mexico.pdf#search='woodruff%20zenteno' 25. Yang, D. (2004, July). Why Do Migrants Return to Poor Countries? Evidence from Philippine Migrants Responses to Exchange Rate Shocks Ann Arbor, Michigan: Gerald R. Ford School of Public Policy and Department of Economics. University of Michigan. Retrieved September 26, 2005, from http://www.fordschool.umich.edu/research/papers/pdffiles/04-003.pdf 30