DETERMINING THE MOTIVATIONS FOR INTERNAL REMITTANCES IN UTTAR PRADESH AND BIHAR. March 15, Jeffrey Erfe

Transcription

1 DETERMINING THE MOTIVATIONS FOR INTERNAL REMITTANCES IN UTTAR PRADESH AND BIHAR March 15, 2007 Jeffrey Erfe Economics Stanford University Stanford, CA under the direction of Prof. Aprajit Mahajan ABSTRACT Remittances, financial transfers between family members across borders, have become a substantial form of private aid to developing countries in recent decades. In contrast with the remittance literature that has focused on international migration from wealthier regions, this paper analyzes two extremely impoverished areas within India that provide considerable levels of internal migrant labor. Using the Survey of Living Conditions conducted in Uttar Pradesh and Bihar from , this paper explores the determinants of internal remittances and their effects on marginal consumption and savings among households in these regions. By quantifying remittance elasticities of consumption and savings, this paper finds evidence that altruism may be a stronger motive relative to interhousehold insurance considerations in the aggregate. A subsidiary finding is that remittance transfers to regional households increase the relative proportion of household expenditures on consumption categories and decrease the relative proportion dedicated to investments in human capital, calling into question the notion that remittances are primarily used for productive reinvestments in developing economies. Keywords: India, remittances, poverty, migrant workers, private transfers, economics Acknowledgements: Prof. Aprajit Mahajan, Prof. Geoffrey Rothwell, Joanne Yoong, Colleen Flaherty.

2 1. Introduction Jeffrey Erfe 1 Remittances have become an increasingly significant source of external financing for the development of emerging economies in recent decades as private capital flows have declined. According to development economist Dilip Ratha (2003), remittances constitute the secondlargest source, behind foreign direct investment (FDI), of external funding for developing countries. In 2001, remittance receipts in developing countries amounted to $72.3 billion, dwarfing estimates for FDI and private non-fdi flows. Even among developed countries in the same year, remittance flows reached unprecedented levels: 42% of total FDI flows. Ratha has argued that remittances may in fact represent an advancement over private capital as a method for aid transmission. While private capital tends to move pro-cyclically, improving incomes during booms but failing to cushion incomes during downturns, remittances covariance with business cycles is less pronounced. Remittances may even rise in response to economic cycles in the recipient economy. For this reason, Ratha expects remittances to rise significantly in the long term, assuming that labor markets in the G-7 economies are strong and efficient procedures for scrutinizing international travelers are in place. Remittances have become an irrevocable fixture within modern economies, producing a host of academic studies intended to ascertain their determinants, their macroeconomic effects, and from a normative perspective whether to encourage remittance flows for the purpose of international development. In the past few decades, economic analyses of remittance flows have both utilized and developed new economic methods and approaches. According to Docquier and Rapoport (2004), studies of the microeconomics of remittances have focused since the early 1980s on how information and social interaction explain transfer behavior. Studies of the determinants of remittances have since grown to acknowledge the strategic, multi-period

3 2 interactions in transfer relationships in addition to more traditional motives suggested in the past. On the macroeconomics side, Docquier and Rapoport (2004, p. 1) argued that while previous research in the 1970s and 1980s focused on the short-run effects of international transfers within the context of static trade models, the analytical focus has since shifted toward longterm considerations such as inequality and development. This paper provides an economic analysis of remittances in India, the country with the highest levels of remittance inflows in the world. The characteristics of these remittance transfers and the negligible proportion of remittances to total macroeconomic activity in India render a macroeconomic analysis difficult to execute and almost impossible to generalize. Instead, this paper attempts to fill a gap in the existing microeconomics literature through an empirical test of the motives that produce remittances. Chishti (2007) provides a number of statistics on Indian remittances. The Reserve Bank of India (RBI) reported that the Indian migrants transferred $24.6 billion to India in fiscal year India continues to maintain a dominant position as the largest recipient of remittances in the world; however, these figures were not always as high. Remittances to India have risen steadily over the past decade. From , remittances amounted to $12.3 billion but rose to nearly $22 billion (not deflated) in Despite these staggering figures, remittances remain a negligible proportion of India s GDP. Although remittances today have grown astronomically since , they still represent only a small fraction (3.08%) of the country s GDP. Given that remittances comprise only a sliver to total macroeconomic activity in India, it is difficult to assess their impacts on growth and development. This idea is compounded by the fact that only a small percentage of

4 3 remittance flows, as computed by the RBI, can be deemed true remittances in the conventional sense: income transfers from one household or individual to another. Table 0.1 Total Remittances as Percentage of GDP Year Remittances Percent GDP (US$ billions) (P) 3.08 Note: (P) notes a projection. For conversion purposes, it has been assumed that US$1 = 45 Indian rupees. Source: Chishti, The RBI divides remittances into two categories: local withdrawals from Non-Resident Indian (NRI) deposit accounts and inward remittances. Transfers received from NRI accounts typically denote local cash withdrawals by Indian diaspora living abroad. The Indian government created NRI accounts in the 1970s in order to attract foreign capital for the purpose of strengthening foreign-exchange reserves. In order to attract investors, NRI depositors can choose to hold deposits in either Indian rupees or foreign currencies. Like interest-bearing debt accounts, NRI depositors can repatriate principal and interest in foreign currency at their discretion. The RBI counts local withdrawals from Rupee-denominated deposits as unrequited transfers and therefore adds them to remittance totals in calculating its statistics. NRI accounts, however, are more of an investment vehicle, a financial innovation in an age of globalized

5 4 electronic financial markets, than a true source of remittance inflows. Chishti (2007, p. 2) notes that many analysts claim India s remittance boom is largely a massive withdrawal surge. 30 Table 0.2 Types of Remittance Receipts, in $US Billions Total Remittance Flows 25 Local withdrawals/redemptions of NRI deposits Inward remittances for family maintenance * 20 US$ Billions Fiscal Year *Note: Projected amount for Source: Chishti, Inward remittances, on the other hand, are more akin to the typical definition of remittance transfers. Inward remittances denote direct transfers of funds from one person, situated abroad or locally, to another in India to provide financial support. Recent RBI initiatives have created two transfer schemes intended to facilitate the receipt of inward remittances (RBI, 2006). These two schemes, the Money Transfer Service Scheme (MTSS) and the Rupee

6 5 Drawing Arrangements (RDA), have gained popularity over other banking channels because of their relative speed and ease of operation. MTSS connects money transfer agents abroad with RBI-approved agents in India: Authorized Dealers, Full Fledged Money Changers, registered NBFCs or International Air Transport Association (IATA) approved travel agents with a minimum net worth. MTSS allows the RBI to track inward remittance flows with ease because only personal remittances intended for family members or foreign tourists in India can be transferred (RBI, 2006). Alternatively, Rupee Drawing Arrangements (RDA) are private contracts for channeling inward remittances between Indian banks and Private Exchange Houses in Singapore, Hong Kong, and the Gulf Region. Inward remittances through RDAs typically take the form of personal remittances from NRIs to relatives in India (RBI, 2006). Surprisingly, little has been written regarding the economic determinants of inward remittances in India. Much of the microeconomics literature that does exist takes a theoretical perspective, constructing models to predict behavioral patterns in the transfer relationship between two households but neglecting to execute data-driven tests of the empirical implications of those models. On the other hand, studies conducted on Indian interhousehold transfer relationships usually focus on a particular subset of migration and its resulting income transfers (e.g. Stark and Rosenzweig (1989)), failing to acknowledge as a whole the workings of a more general class of internal remittances that result from labor migration. Most importantly, since the RBI does not distinguish inward remittances that traverse international boundaries from purely internal inward remittances, no studies have analyzed the determinants of remittances produced by internal migrant labor in India. There is reason to believe that these determinants may differ. In general, standard human capital theory identifies a number of determinants for worker mobility (e.g. age, education,

7 6 distance) that suggest different types of workers may self-select for internal or international labor migration. With this initial self-selection process in mind, it is not unreasonable to explore the notion that international labor migrants may remit for reasons that largely differ from their internal labor market counterparts. A divergence of motivations could translate into a divergence in the nature and characteristics of remittances and the transfer relationship. This study attempts to fill this research gap by analyzing the theoretical determinants of remittances using the Uttar Pradesh and Bihar Survey of Living Conditions. A large proportion of the respondents in this survey receive remittances from internal migrant workers in India. In the literature review, four different models that aim to explain the motivations for remittance transfers are presented and their implications derived. Two models, the altruism and insurance models, imply contrary predictions on the timing and cyclicality of remittances. According to standard microeconomic theory on consumer behavior, the marginal propensity to consume or to save income transfers depends largely on the timing of these transfers. With this theoretical foundation, the methodology develops a strategy to calculate the percentage change in the marginal propensities to consume or to save remittance receipts in order to infer the timing of these transfers. In turn, the relative strength of the altruistic or insurance motives can be determined from the economic results. These results suggest that the altruistic motive seems to dominate the insurance motive in the aggregate among Indian households in Uttar Pradesh and Bihar. Section 2 provides an overview of the existing literature by presenting several utilitymaximization models and their predictions on the remitting behavior of migrants. More specifically, this section proposes four models that may explain the motivation to remit: altruism, pareto-improving transactions, strategy, and insurance. Section 3 develops an econometric

8 7 methodology that uses the Working-Leser Model to develop estimates for marginal income shares and remittance elasticities. These elasticities are used to test empirically the timing implications of the altruism and insurance models in order to determine which effect proves stronger in the aggregate among Indian households. Additionally, the remittance elasticities and their implications are compared to certain empirical observations of the Adams (2005) study in Guatemala upon which the methodology of this paper is largely based. Section 4 presents a number of summary statistics regarding the dataset to be analyzed. Section 5 presents the results of this analysis. Section 6 finishes the paper with a summary of the findings and suggestions for further research.

9 2. Literature Review Jeffrey Erfe 8 Studies of the microeconomics of remittances in the past have focused overwhelmingly on theoretical constructions of their determinants. This section reviews the relevant literature that models the motives behind remittance transfers and therefore provides the theoretical foundation for this paper. The central goal of this paper is to test the degree to which the altruism and insurance motives hold for households by quantifying the effects on marginal consumption and savings of remittances. Any theoretical discussion of remittances should first attempt to differentiate itself from a discussion of migration. Readers might wonder whether the determinants of remittances are in fact different from those of migration. In other words, does the probability that a household receives remittances reflect the result of two successive, interrelated self-selection processes, one for whether to migrate and another for whether to remit? Or, does the probability that a household receives remittances reflect a single decision by a household member both to migrate and to remit? Edward Funkhouser s (1995) study of remittances to the capitals of Nicaragua and El Salvador suggest that two self-selection processes are at work. Funkhouser discovered many similarities between the two migrant populations in a number of demographic characteristics such as age, education, gender, and to a lesser extent the number of years since emigration. The differences between El Salvadorans and Nicaraguans in these observed characteristics and the timing of emigration could not explain the differences in remitting patterns. Only when Funkhouser estimated remittance functions could he explain variations in remitting behavior, the probability of remitting and the level of remittances, through both observed characteristics and unobserved characteristics related to the process of selfselection. Although emigrants that remit in both countries are those whose unobserved ability to

10 9 remit is lower than those who do not remit, for emigrants from Nicaragua, this self-selection bias is much larger. Funkhouser therefore inferred that better educated Salvadorans tended to maintain stronger filial and/or patriotic attachments because they were more likely to be motivated by these unobservable characteristics. Furthermore, he discovered a clearer negative correlation between years since emigration and probability of remitting for Nicaraguans, implying that Salvadorans were more likely to return home. Funkhouser s study highlights the importance of separating analyses of migration from those of remittances. While the two undoubtedly share interrelated determinants, the sets of determinants for migration and remittances are not identical. In other words, remittance behavior is not simply a function of migrants observable characteristics. There is an element of choice in the decision to remit; accordingly, a number of theoretical models have arisen to elucidate remittance-sending motivations. The most intuitive theory for the motivation to remit is migrants altruism, their proclivity to extend monetary aid to close kin in a home country. Other motivations include riskdiversification, consumption smoothing, and the intergenerational transfer of income. Docquier and Rapoport (2004) noted that theories that attempt to model remitting behavior as a product of these motivations typically take a game theoretic perspective. For example, if altruism does not adequately enforce remitting behavior, home families may punish the self-interested actions of the remitting member by altering inheritance protocols or enforcing social sanctions. In effect, a remitting member must weigh the rewards to opportunistic behavior against the potential family sanctions that may result. Strategic gaming may result when imperfect information a product of distance and unobservable needs of the home family and the remitting member enters into the transfer relationship.

11 10 In order to understand any game theoretic model of the transfer relationship, one must first analyze the utility-maximization processes at work for a household and the remitting member. Docquier and Rapoport (2004) borrowed from Stark (1995, Chapter 1) a utility model to explain the altruistic motive, the simplest and most intuitive of all possible motives. The following section illustrates the altruism model, borrowing from both Docquier and Rapoport (2004) and Stark (1995, Chapter 1). A. Altruism Imagine a world with two agents: a remitting migrant (m) and a receiving household (h). Let each agent s utility be defined as U i, where i =m, h. I denotes pre-transfer incomes, C denotes consumption, and T represents the amount remitted by m to h; naturally, C + T = I. The level of C affects an agent s own-pleasure (felicity). Docquier and Rapoport assumed each U i is affected by the felicity derived from own consumption, V i (C i ), with V > 0 and V < 0, and the utility of the other. Utility can then be written as the weighted average of own consumption and utility of the other player, where i denotes the level of altruism. Let 0 i 1 to ensure that each agent attaches a non-negative weight both to own felicity and the utility of the other (i.e. the agent is neither masochistic nor envious). U m (C m, C h ) = (1 m )V m (C m ) + m U h (C m,c h ) (2.1) U h (C h, C m ) = (1 h )V h (C h ) + h U m (C m,c h ) (2.2) Solving these two equations for U i in terms of V i (C i ) gives: U m (C m, C h ) = (1 m )V m (C m ) + m V h (C h ) (2.3) U h (C h, C m ) = (1 h )V h (C h ) + h V m (C m ) (2.4)

12 11 where 0 m = and 0 h = cv (2.5) Note that the initial bounds on i dictate that i > 0 and m + h = 1. The utility function of the remitting member can then be rewritten as follows: U m (C m, C h ) = (1 m )V m (I m T) + m V h (I h + T ) (2.6) This utility function exhibits the following properties. All possible degress of generosity are between complete selfishness ( m = 0) and complete altruism ( m = 1). Examining equation (2.5), one sees that these values of m correspond with m = 0 and m = 1, respectively. Carrying these values through to equation (2.6), the extreme values of the utility function arise as V m (I m T) in the case of complete selfishness and V h (I h + T ) in the case of complete altruism. In other words, the migrant s utility corresponds directly with his felicity if he is completely selfish and corresponds directly with the household s felicity if he is completely altruistic. Similarly, the utility function of the recipient household can be rewritten with the same substitution: U h (C h, C m ) = (1 h )V h (I h + T) + h V m (I m T ) (2.7) Repeating the analysis conducted on equation (2.7) shows that the reduced form of equation (2.7) is V h (I h + T) when the household is completely selfish and V m (I m T ) when the household is completely altruistic.

13 12 Returning to the utility function of the migrant, maximizing equation (2.3) with respect to T gives the first order condition: (1 m V m ) + m h V 0, with equality for T > 0. (2.8) C m C h Once one removes the possibility of negative transfers from m to h, and assuming that V(.) = ln(.), one finds that the optimal remittance level is the following: (2.9) Taking first-derivatives results in comparative statics observations that describe the effect on the optimal transfer as income and altruism change exogenously: T * / I m > 0 if m I m > 0, T */I h < 0, T */ m > 0, T */ h < 0. These comparative statics imply that the altruistic transfer increases with the migrant s income and level of altruism and decreases with the home family s income and level of altruism. These results seem relatively reasonable. All else equal, one would expect a migrant to remit more the higher his income because he will have a larger proportion of discretionary income after fulfilling his own consumption needs. Further, a migrant will commit more of his income to remittances the more his utility depends on the welfare of the household. On the other hand, remittances should decrease the richer the recipient household because the marginal value of a remitted dollar is much lower. The rationale for why one might expect remittances to decrease in the household s utility is less intuitive; presumably, since a less altruistic household places a lower value on a migrant s utility, the migrant may in turn place a low place on the household s utility and therefore remit less.

14 13 Docquier and Rapoport (2004) outlined a number of important predictions of the pure altruism hypothesis. Although the altruistic parameters m and h cannot be observed in empirical data, one can test the notion that transfers do not increase with home family s income. A second prediction is that a unit-increase in the migrant s income that is met with a unitincrease in the home family s income should increase net transfers by one unit. This can be seen by the following transfer-income derivative: (2.10) This implication of the altruism motive is typically unverifiable because datasets on remittances do not usually capture data on the remitting member s income. Funkhouser (1995) proposed another behavioral model for remittances that offers a number of predictions similar to the altruism hypothesis. First, migrants with higher earnings potential should return higher monetary levels of remittances. Second, households with lower incomes should receive more remittances. Docquier and Rapoport (2004) point out that the first two predictions are compatible with a range of possible motives, but these two predictions are similar to those from the previous model if we believe that high earnings potential translates into high income for the migrant. Third, the monetary level of remittances should decrease the larger the distance between a migrant worker and his/her home family. Fourth, remittances should increase in the migrant s intentions to return to the home family. Fifth, remittances should decrease in the number of other remitting members from the same household. Sixth, the timing of remittance transfers depends on both the migrant s time-discount factor and the migrant s earnings profile abroad.

15 B. Pareto-Improving Exchanges Jeffrey Erfe 14 A second possible motive after altruism is that of pareto-improving exchanges. Note that remittance transfers in the altruism model are pareto-improving since the utility function of one agent incorporates the utility of the other. Pareto-improving exchanges in the context of this motive denote exchanges that improve the utility of the other agent as a secondary effect; these exchanges can be thought of as more selfish. For example, one example of this type of paretoimproving exchange is the purchasing of various types of services from the home family, necessary favors that improve the remitting member s utility. For example, a migrant may need household members to watch over the migrant s assets such as land or the migrant s immediate family members such as children while the migrant is away. Another pareto-improving exchange could result if the migrant uses remittances to repay the household for monies used to finance the migrant s human capital investments (e.g. education, training) or transaction costs incurred during the migration process. Docquier and Rapaport (2004) adapted the work of Cox (1987) to model the exchange motive by considering non-altruistic agents and fixed amounts of service. Given that the empirical implications are quite complex and this paper does not attempt to test implications of the pareto-improving model, I will let interested readers refer to these papers for a quantitative exploration of this remittance motive. C. Strategic Motive A third important motive to consider is the so-called strategic motive developed by Stark (1995, Chapter 4). The strategic motive underscores the importance of treating the decisions to migrate and to remit as two interdependent decisions in a unified framework. Stark suggested that remittances are one component of a strategic interaction intended to induce

16 15 positive selection among migrants. Assuming a world of heterogeneity in skills and imperfectly observable labor productivity among migrants, migrants employers apply statistical discrimination and pay migrants according to the average productivity of the minority group to which they belong. Docquier and Rapoport (2004) noted that cooperative arrangements can arise between skilled and unskilled migrants wherein skilled migrants in effect bribe unskilled migrants in order to keep them at home since their home production is more valuable than their market wages. Docquier and Rapoport (2004) provided a game theoretic illustration of the strategic motive. Let m and h represent potential migrants but h possesses less market skills than does m. Let h s productivity be a proportion, 0 < < 1, of m s productivity. If both m and h choose to remain home, then h earns I h while m earns a higher wage which can be denoted I h /. If m alone migrates, m earns simply his or her marginal product in the destination area, I m. However, if both h and m migrate, then they each earn their average productivity, [(1 + ) / 2] * I m, assuming a world of imperfect information where employers cannot discern individual skill levels. One can then write the following payoff matrix: The strategic motive operates under two conditions. In a game without transfers, (Migrate, Migrate) must be a Nash equilibrium. This holds when:

17 16 (2.11) Assuming that equation (2.11) holds, the level of strategically-motivated remittances should increase both players utilities. This requires: (2.12) and (2.13) with strict inequality for at least one of the two players. From equation (2.12), one can derive algebraically the optimal transfer: (2.14) arises: If one substitutes equation (2.14) into equation (2.12), then the following wage inequality (2.15) Equation (2.14) implies that unskilled migration is efficient only under the condition that unskilled workers are pooled with skilled workers. In this sense, remittances may help in promoting positive selection in the context of strategic interactions between migrants and nonmigrants. Unfortunately, the strategic model suffers from many weaknesses. First, each migrant is motivated to free ride on other migrants efforts to achieve positive selection. Second, if the revelation of individuals skill levels happens quickly then low gains to migration may be at stake. Third, Docquier and Rapoport (1998, p. 14) pointed out that the strategic motive requires employers in the destination area to possess a knowledge of anthropology, a condition that is

18 17 unlikely to be met in the real world because of information asymmetries. Employers will probably view migrants as a subset of foreigners, suggesting that the selection undertaken by one group will benefit the whole group with which it is identified. D. Insurance Motive A fourth motive discussed in the literature is that of insurance and moral hazard. Unlike the previous motives, which largely stem from individual decisions, this motive contextualizes the decisions to migrate and remit within a larger framework of social and familial interactions. In a nutshell, this motive suggests that in a situation of imperfect capital markets, remittances may serve as an implicit contract between the remitting migrant and the recipient household, allowing the household to smooth consumption despite volatility in its income. Stark and Rosenzweig (1989) highlighted the threat of income volatility inherent to agricultural economies of developing nations, particularly India. Stark and Rosenzweig noted that imperfect credit markets necessitate informal family and caste contracts to insure against income volatility. Although Stark and Rosenzweig focused on migration stemming from marriage as an interfamily contract to reduce income risks, their analysis can be carried to remittances. Docquier and Rapoport (2004) noted that migration to other areas is subject to risks generally uncorrelated with agricultural activities in the home household (e.g. crop failure, cattle disease, etc.). These authors suggested that migrants in effect insure the receiving household not only against drops in income, but also unemployment and retirement. To enforce this implicit contract, receiving families must obtain reliable information on remitting members types in order to select the migrants with the optimal mix of a good earnings profile and a high degree of loyalty to the family. The family can threaten the migrant with ostracism, denial of inheritance

19 18 rights to family property, or even banishment from the village in extreme cases. Docquier and Rapoport (2004) argued that rich families will therefore rely more heavily on migration than poor families because these threats, particular inheritance rights, are more credible among richer families. Empirically, whether richer families are more likely to have migrants is ambiguous because this motive is mixed with others in the aggregate, all of which are directly unobservable. Docquier and Rapoport (2004) quantified the implications of the insurance motive with a utility-maximization model. Consider a family with two members living for two periods. Every member receives an exogenously determined first-period wage, I 0, and the second-period rural _ wage is a random walk. This second-period wage is I h with probability p and I h with _ probability 1-p, subject to the constraint I h < I h. This aggregate uncertainty affects all individuals in the same fashion, and a key assumption is that the utility function is identical for all agents and is additively separable. The expected utility of agents is therefore the equation: (2.16) with v > 0 and v < 0 to account for risk-averse behavior. Assume agents have the possibility to migrate to a destination with no uncertainty, and earn a wage I m during the second period. Let (c) denote the transaction costs associated with migration. Docquier and Rapoport (2004) assumed that all agents are credit constrained, so that no individual migrant can finance the cost of migration alone and therefore every migrant requires a familial contract: (2.17)

20 19 During the first period, agents agree on an informal contract to specify how each party bears migration costs and the amounts to be remitted in bad and good states of nature. The set of Pareto-efficient contracts is given by the equation: (2.18) where T p and T 1-p denote the amounts to be remitted in the bad and the good state. denotes the share of the migration cost borne by the migrant, represents the relative _ weight of the remaining agent in the bargaining process, and V h gives the utility of the migrant and the remaining household, respectively, according to the equations: (2.19) (2.20) The first order conditions can be derived as: (2.21) (2.22) (2.23) To illustrate the interaction between migrants and family, let = 1 and I m = E(I h ). In the simple case when * = ½, migration reduces uncertainty by ½, and expected utilities are equalized. However, migration pays off only if the transaction costs associated with migration are below c*, which is a function of the degree of risk-aversion. E. Comparison of Altruism and Insurance The insurance and altruistic motives predict similar signs on the effects of pre-transfer income levels on the amounts remitted. However, according to Docquier and Rapoport (2004),

21 20 these models differ on the predicted effect of familial wealth on the size of remittances and the predicted timing of remittances. Although the altruistic motive implies higher remittances to lower-income households, the bargaining model implies the opposite for two reasons. Since migration pays more for families with sizeable but risky assets and since the exchange model predicts greater familial wealth increases the family s bargaining power, remittances may actually increase with receiving household wealth if one assumes the insurance motive. With regards to the timing of remittances, the insurance motive implies that migration and remittances are more likely in areas with higher income volatility and that remittances should be sent on an irregular basis since the migrant s utility is a function of the migrant s income rather than household utility (see equation 2.19). On the other hand, the altruism model shows that the migrant s utility is a linear function of both the migrant s consumption and the household s consumption (see equation 2.6). Agents acting based on altruistic motives should therefore remit more frequently and regularly than those driven by contractual insurance obligations. Docquier and Rapoport (2004) also argued that altruism should imply a gradual decrease of remittances over time, as long as one assumes altruism is solvable in distance and time, while the insurance motive should imply no increase during a given period (if specified contractually) and a sharp decline after the contract expires. Standard microeconomic theory clearly predicts the effect on consumption and savings behavior. Using a multi-period model that takes current income, future income, and wealth as given, Abel and Bernanke (2005, Chapter 4) use a utility-maximization model for a single consumer and employ comparative statics analysis to predict the effect of an increase in these exogenous variables on consumption and savings. Abel and Bernanke differentiate the effects of permanent and temporary income changes by invoking the permanent income theory of

22 21 consumption and savings. They argue that permanent changes in income, because they last for multiple periods, should have larger effects on consumption than temporary changes in income. Therefore, temporary income increases should be mostly saved, while permanent income increases should be mostly consumed. Remittances motivated by altruism, insofar as they can be counted on by a household to arrive at more regular intervals, can be viewed as permanent income increases because the household effectively incorporates the expected remittance receipts into its budget constraint, increasing the present-value of lifetime earnings for the household. On the other hand, remittances motivated by contractual insurance obligations are more akin to temporary income changes in the household s budget constraint because they arrive at more irregular intervals. They are more likely to be a part of contractual obligations than remittances motivated by altruism. As noted above, insurance remittances should not change drastically in volume during the period of the contract but should decrease sharply afterwards, a characteristic that further suggests these remittances can be viewed as temporary changes in income. It is arguable that the average time horizon for the life of a contract is quite long; if this is the case, this analysis breaks down. However, there is reason to believe that insurance contracts may last for only short periods of time, especially if they are repayments for previous services or favors rendered. In sum, remittances motivated by altruism should mostly be consumed, while those motivated by insurance needs should mostly be saved. The goal of this paper is to test empirically the relative strength of the altruistic and insurance motives for remittances by calculating how remittances are spent among various categories of goods. One can calculate the household savings rate as the amount not spent on consumption goods. Appropriating the Working-Leser model used in a study of Guatemalan

23 22 remittances by Richard Adams, Jr. (2005), this paper calculates marginal and average income shares of consumption and savings to compute remittance elasticities. (For a full explanation of the Working-Leser model, please refer to Economics and Consumer Behavior by Deaton and Muellbauer (1999)). These elasticities quantify the change in the marginal propensity to consume a particular good and the marginal propensity to save, given remittance receipts. If the aggregate percentage increase in consumption at the margin is greater than that of savings, one may conclude that the altruism motive operates more strongly among the households in the sample; alternatively, if the opposite is true than insurance must be a stronger motivation. Previous literature on income transfers among sub-caste networks in India suggests that the insurance motive is probably stronger in the aggregate. Stark and Rosenzweig (1989) examined from a risk-theoretic perspective a major component of migration in agricultural communities, the movement of women for marriage purposes. The authors argued that the spatial distribution and characteristics of matches are in fact manifestations of implicit interhousehold contractual arrangements intended to facilitate household consumption smoothing in an environment with information costs and spatially covariant risks (i.e. farming profits). Analyzing longitudinal data from South Indian villages, Stark and Rosenzweig discovered that marriage migration reduces consumption variability, for a given variability in income from crop production, and that households exposed to higher income risk are more likely to invest in longer-distance migration-marriage arrangements. These arrangements are akin to insurance contracts aimed at smoothing consumption in two households connected through marriage. A more recent study by Munshi and Rosenzweig (2005) examined the persistence of low spatial and marital mobility in India in the context of increasing growth rates and rising inequality in recent decades. This study analyzed Indian panel data and discovered that Indians

24 23 who out-marry and migrate lose the ability to utilize intra-caste loans; mobility is therefore dampened if alternative sources of insurance or finance of comparable quality are unavailable. These findings support the notion that insurance motives in interhousehold income transfers, here intra-caste loans, play a prominent role in migration and remittance decisions.

25 3. Methodology Jeffrey Erfe 24 The econometric methodology entails three major components, all of which are largely based on Richard Adams, Jr. s (2005) study of Guatemalan labor migration and the remittances it produces. Adams analyzes a large set of microdata to determine how the receipt of internal and international remittances affects the marginal spending behavior of households on various consumption and investment goods. In one part of his study, Adams employs a Working-Leser model to calculate marginal budget shares and to quantify the effect of remittances on the marginal propensities of households to consume various expenditure categories. Adams finds that households at the margin dedicate a large proportion of their incomes to investment goods (e.g. education, health, housing) and therefore boost local economic development by creating new income and employment opportunities for skilled and unskilled workers. First, a replication of the relevant components of Adams (2005) Guatemala study is conducted. I perform an ordinary least-squares regression analysis on average budget shares with the same specification of the Working-Leser model that Adams uses in his study. The coefficient estimates from Adams specification are compared to those derived from a more common form of the Working-Leser model that lacks a reciprocal of expenditures term. Adams specification is denoted as the A specification, while the other specification is denoted as the B specification. Second, an instrumental variables regression is run using income as an instrument for expenditures in an effort to control for consumption preferences that may be endogenous to the model. Third, average income shares are regressed onto the B specification to determine marginal income shares and to calculate the remittance elasticities necessary to ascertain the relative magnitude of the altruism and insurance motives in the data.

26 A. Expenditure Shares Analysis Jeffrey Erfe 25 First, I must select a proper functional form to analyze the marginal spending behavior of different groups of households. For a functional form, I choose to use the Working-Leser model as suggested by Adams (2005). The first portion of the results section provides an exact replication of Adams analysis on budget shares. The methodological and econometric reasons for choosing the Working-Leser model and the specific regressors will be discussed in Section C since they are largely identical. However, it is instructive to provide here a quick summary of the methodology used in the 2005 Guatemala study. Adams chooses a Working-Leser functional form for a number of reasons. First, the model provides a good statistical fit to household expenditure on a wide range of goods, such as housing, food and education. Second, since the focus is on expenditure relationships, the chosen form requires a slope that is free to change with consumption expenditures. Since this paper aims to calculate remittance elasticities, the model cannot use a specification that imposes a constant slope (or marginal income share) for all levels of consumption expenditures. The Working-Leser form allows mathematically for rising, falling or constant marginal propensities to spend over a broad range of goods and expenditure levels. Third, the Working-Leser model allows for the additivity criterion. If one assumes for a moment that all households differ only according to their levels of total expenditure, then one can easily see that a linear functional form would be too restrictive for analyzing marginal consumption patterns. The linear Engel curve can be written as: C i = i + i * EXP (3.1)

27 26 where C i denotes expenditure on good i, EXP denotes total household expenditures per capita, and represents a constant intercept. Obviously, the marginal income share of good i ( i ) cannot vary with expenditures according to this specification. For this reason, a nonlinear function is necessary for the analysis. Adams noted that a number of nonlinear functional forms for Engel curves have been explored in previous literature. For example, Prais and Houthakker (1971) experimented with double logarithmic, semilogarithmic and log reciprocal forms. Adams admitted that although each of these forms possesses certain analytical advantages when examining different goods, none of them conform fully with the criterion of additivity. The Working-Leser model meets all of the necessary criteria. It linearly relates budget shares to the logarithm of total expenditure. This paper uses the version of the Working-Leser model introduced by Adams. The model can be written as: C i /EXP = i + i /EXP + i (log EXP) (3.2) An important constraint imposed by Adams is that the sum of all five expenditure categories should equal unity. This constraint is modified in Section C when expenditure categories are divided by income and not total expenditures, loosening the constraint so that the sum of all goods expenditure ratios and the savings ratio should equal unity. However, all the Section A regressions satisfy the following constraint: C i /EXP = 1 (3.3) Adams wrote equation (3.2) as the Engel function: C i = i + i * EXP + i (EXP) (log EXP) (3.4)

28 27 After choosing a number of regressors that should prove significant according to human capital theory, Adams modified equation (3.4) so that each variable can shift both the intercept and slope of the Engel function. Let Z j denote the household characteristic variable j and let ij and ij be constants. The complete Working-Leser model can be written as: C i = i + i (EXP) + i (EXP) (log EXP) + j [( ij )(Z j ) + ij (EXP)(Z j )] (3.5) The model can be written in semi-log ratio form by dividing through by EXP: C i /EXP = i + i /EXP + i (log EXP) + j [( ij ) Z j / EXP + ij (Z j )] (3.6) Equation (3.6) is therefore the complete Working-Leser model for average budget shares and is called the A specification in this paper. An alternative specification more common in the literature is the following: C i /EXP = i + i (log EXP) + j [( ij ) Z j / EXP + ij (Z j )] (3.7) Equation (3.7) will be referred to as the B specification. The B specification is identical to the A specification except that it lacks the i /EXP term. One goal of the first section of the methodology is to test whether the extra EXP term present in the A specification significantly alters the values of the regression coefficients or their significance. A second goal is to compare the coefficient estimates and remittance elasticity values gleaned from these regressions on data from Uttar Pradesh and Bihar with those from found by Adams in his study of Guatemalan households. The results of these two investigations will aid in assessing the validity of Adams results on a more general scale and will determine the specification for the second and third parts

29 28 of the methodology (whether to regress on A or B). If the coefficients, their significance estimates, and the derived remittance elasticities do not largely differ between A and B, then the more parsimonious B specification will be used in the rest of the paper. An analysis of expenditure shares involving households with varying levels of income must account for socioeconomic and locational factors that could determine differences in expenditure patterns. For example, differences in family or household composition, such as the number of children or working-age adults, could influence expenditure share determination. The complete list of these factors includes demographic characteristics, human capital variables and wealth. Each of these factors is assigned a variable that will enter the Working-Leser specifications as a regressor Z j in the component j [( ij ) Z j / EXP + ij (Z j )] of equations (3.6) and (3.7). The functional form should therefore include a number of household demographic variables. In this paper, these variables include the household size, the age of the household head, and the number of children under age five. The size of the household should be a significant determinant of expenditure shares. To illustrate this idea, suppose that two households identical in incomes and all other relevant demographic characteristics differ only in their number of members. Since some consumption goods are more necessary than others for sustaining basic household welfare, these two households would differ in their allocation of total expenditures toward various consumption goods. For example, the larger household would need to allocate a larger proportion of its expenditures for food in order to feed all of its members. The smaller household would be able to allocate the difference between what it spends on food and what the larger household spends on food to other expenditure categories, such as human capital investments like education.

30 29 The age of the household head should be significant given that older household heads, all else being equal, are more likely to have a larger amount of children. Additionally, the variance of the ages of these children should be larger on average since older household heads have had more time to space out the timing of their children s births. Therefore, the age of the household head should have a large effect on the age of other household members, thereby influencing the household s composition. All else equal, households that differ in the ages of their members should allocate income toward expenditure categories differently. For example, a family with several young children might allocate a larger proportion of expenditures for health care (e.g. immunizations), while a family with more teenagers might allocate a larger proportion of expenditures for food, assuming that teenagers eat larger amounts than small children. The final household characteristic variable included in the functional form tracks the number of children under the age of five. The rationale for including this variable is similar to that provided above. It is useful, however, to include a separate variable to track children because they have a unique effect on expenditure shares. Children typically need specialized food and health care (e.g. immunizations); moreover, children under the age of five do not contribute to household income. For these reasons, a household with a larger proportion of children will probably allocate expenditures differently from an otherwise identical household. A specification that includes a variable for children therefore can capture these differences in expenditure shares while a specification that includes only household size cannot. Another important category of regressors to be included in the specification are human capital variables. More specifically, the specification includes variables that track the number of household members with preparatory, primary, secondary, and university education. The rationale for including these human capital variables is twofold. First, education is typically

31 30 correlated with age since older people are more likely to have attained higher levels of education. The ages of household members can affect income shares, as discussed previously. Second, and more importantly, an individual s educational attainment affects his or her earnings potential. In turn, the earnings of each individual member determine the total income of the household. All else equal, a household with lower income per capita will have lower expenditures per capita, necessitating a more parsimonious allocation of income among expenditure categories. For example, households with higher incomes per capita will have a greater ability to spend on capital improvements in the home, since the basic subsistence needs of all members have been met. One final variable necessary for the analysis is a proxy for non-monetary wealth. The wealth of a household, even if it is stored not in liquid currency but rather non-monetary assets, can affect expenditure categories if it is an indicator of high savings or social networks that consistently relay gifts or income transfers. In this data, total durable goods owned by a household will serve as a proxy for aggregate wealth. In order to calculate the Marginal Budget Shares (MBS i ) for the ith good, one must differentiate each expenditure share (equation (3.5)) with respect to total expenditures per capita: MBS i = C i / EXP = i + i (1 + log EXP) + j [( ij )(Z j )] (3.8) Average Budget Shares (ABS i ) are defined as follows: ABS i = C i / EXP (3.9) The remittance elasticity of the expenditure share of good i is then calculated as follows: i = (MBS i / ABS i ) (3.10)