Intra-household resource allocation and gender-based returns to loans in rural Bangladesh

Transcription

1 Intra-household resource allocation and gender-based returns to loans in rural Bangladesh Syed Saad Department of Economics University of North Carolina at Chapel Hill Preliminary and Incomplete November 14, 2007 Abstract This paper examines gender-specific effect of the income generated from borrowing on the intra-household resource allocation and decision making process in rural Bangladesh. The loan sources vary from micro-credit to government and commercial loans. I exploit the gender eligibility restrictions in micro-credit villages and other relevant instruments to identify the impact. Using a bargaining model framework and endogenous bargaining power, I show how the income generated can effect consumption through the changes in the individual weights and I find that when the income is the result of female borrowing, households tend to spend more on female-oriented goods. I also find that when women are given income earning opportunities, they are empowered. This empowerment comes in terms of being able to make major household decisions, travel, and understand her legal rights. These findings are inconsistent with the unitary model of identical preferences. Moreover, they suggest that even in patriarchal countries where women may not have formal labor labor, women s empowerment and thus improvement in her bargaining can still be obtained through policies and programs that target women like micro-credit. 1

2 . 1 Introduction Empowerment of women is a human right. The benefits of empowering women not only extend to women but also to their children and their children s children. As former UN Secretary General Kofi Annan states, women s empowerment is precondition for development, reduction in poverty, and good governance. When society succeeds in empowering women, all of society benefits and the succeeding generations are given a better start in life. One way to measure empowerment is through the changes in intra-household resource allocation toward goods that are valued by the woman. Yet another measure of empowerment is a women s decision making power and their political, legal, and educational awareness. In this paper, I estimate women s empowerment using both types of measures when women are given the opportunity to earn income in rural Bangladesh. The traditional models of the household behavior are not able to explain the issues regarding household allocation and bargaining power. These models assume that all household members have identical preferences or that preference of the dominant individual dictates the household preference, thereby treating the household as a single entity. In other words, what goes on within the household is matter of little importance. The assumption of identical or dominant preference is erroneous since individuals are different with heterogeneous preferences and thus should be modeled accordingly. And even though many decisions such as fertility, education, labor-force participation are made at the household level, they may emphasize preferences of one member of the household over another. Nevertheless, many studies in development economics use the household as the unit of analysis, with little attentions paid to the internal decision making processes. They fail to illustrate how the decisions are made and how the resources are allocated within the household, both of which may be influenced by the relative power of individuals. Indeed, how much power a woman has in the household may differ across households and be determined by the level of her contribution to the household income. From the policy makers perspectives, analysis of intra-household resource allocation is important. Individual s well-being may be influenced by the way money is injected into 2

3 the household. Ten dollars allotted to the male household head can have different effects on the child health, labor, and education and on tobacco and alcohol purchase than when the same amount is allotted to his wife (Kanbur and Haddad, 1994). Moreover, different individuals may have different welfare within the same household. Some households with average per capital income above the poverty line may have members whose standard of life may be below the poverty line due to inequality in resource allocation (Haddad and Kanbur, 1990). These issues are interesting and important when studied in terms of husbands and wives in rural areas of third world countries where women are usually disempowered and may lack formal labor market opportunities. Under these circumstances, programs that target women may improve their welfare relative to the husbands by altering household resource allocation. There have been considerable evidence that child education and health improve or expenditure on child clothing rises when mother s non-labor income or asset increase or when the mother is the welfare recipient (Thomas et. al, 2002; Lundberg et. al, 1997). Thus, examining intra-household allocation of resources may reveal unanticipated and sometimes unintended consequences. The micro-credit programs in rural Bangladesh provide the ideal condition for the study of intra-household resource allocation and the relative power of the spouses. Rural Bangladesh maintains a predominantly Islamic religion and culture with purdah norms that restrict women s mobility. It also lacks a formal labor market for women. Therefore, most women depend on their husbands for their livelihood. However, over the past decade, micro-credit programs have been increasingly targeting women and dispensing small micro loans to poor women without any collateral. The borrowers may use the loans to start their own micro-enterprises. Thus, these programs provide income-generating opportunities for women who previously had none. In addition, they provide training, legal and family planning educations to empower women. All these may increase her say or power in the household matter and thus change the manner in which resources are allocated. This is evident in the story of Sakina Begum. When her husband lost his job, she enrolled in a micro-credit program in her village and used the small loan to purchase a cattle so that she can sell the milk at the local market. Gradually, she expanded her livestock and land, and paid off her debt. With the increased income, she was able to provide for herself and her children; she put her three sons through secondary school. Sakina states that previously, 3

4 her husband dominated her and disregarded her voice in the family matter. She states, I did not dare to go against him... and would often tolerate his tortures. But I am no more a meek and helpless person. I feel (empowered). Through the organization s different training and orientation programs and her income contribution to her family, Sakina gained economic, social, and cultural emancipation, empowering her as a woman. [BEES] 1 Thus, in this paper, I examine how the income generated from micro-credit program participation differentially influences the intra-household resource allocation, based on the gender of the participant. The allocation is measured by the gender-oriented expenditure patterns of rural households from Bangladesh, some of whom participate in micro-credit programs. If program participation by the wife leads to allocation of resources that is relatively skewed toward the goods that are of primary interest to her, this may be evidence of empowerment. The difference in resource allocation would result from the increase in female labor income, which depends on her labor hours. Thus, her bargaining power is determined endogenously because it is now a function of her labor income. Past studies have treated the determinants of bargaining power as exogenous. One of the main contributions of this paper is that it provides for the endogenous determinants of the power by using a two-stage model and allowing for the determinants to include choice variables. Thus, this paper adds to the literature of bargaining and intra-household allocation. The second contribution is to evaluate the micro-credit programs from a different perspective. Thus far, most of the evaluations of these programs have been confined to pecuniary benefits accrued to the participants. However, female empowerment through changes in the intra-household allocation provides yet another perspective to justify these programs. The rest of the paper is organized as follows. Section II provides the background for the micro-credit programs and literature review. Section III and IV gives the theoretical and empirical models, respectively. Section V discusses the data, followed by the results in section VI. Section VII concludes. 1 Bangladesh Extension Educational Services - organization provides financial support through its credit program, vocational trainings, and awareness of health, nutrition, rights etc. 4

5 2 Background 2.1 Brief history of micro-credit Micro-credit programs are poverty alleviation tools that try to provide access to loans and savings for poor people who otherwise lack any savings or substantial land- assets that are typically used as collateral. Credit programs, such as Grameen Bank, started in the 1970s and generally target women in the rural Bangladesh. Dr. Mohammed Yunus, an economics professor, founded the first micro-credit organization, Grameen Bank, in Bangladesh in The bank is based on the belief that credit is a human right. It provides loans to the credit-constrained and landless poor who are defined as those owning less than acre of land. The bank uses peer monitoring as a substitute for collateral. Each member self-selects into a group of five. Furthermore, even though loans are given to each member, everyone in the group becomes ineligible (lose access to loans) if one defaults. Thus, collateral is replaced by group pressure in ensuring repayment. Peer monitoring however provides some additional benefits. Since banks do not observe the individuals with high risk of default or this knowledge may be costly to obtain, banks transfer the liability to the peers. It is often a sunk cost for the peers to gain knowledge of the individual s characteristics. Therefore, not only will the members refuse to group with someone they know will be risky, but they also will make sure that group members utilize the loans effectively. In addition to keeping their credit intact, members may decide to repay someone else s loan and mentor each other about timely repayment. This may explain the high loan recovery rate, at approximately 98%, for the Bank. However, the downside of the approach is that one s financial problem becomes the group s problem. This may lead to a domino effect where if one member defaults, then others may default as well knowing that they would lose loan access regardless. To prevent this, the Bank has started to disburse loans at different times. At first, two members of the five- member group receive a loan. If they repay regularly for the next six weeks, two more members can apply for a loan. The chairperson of the group is usually the last borrower of the five. Most loans last exactly one year. Therefore, due to different timing, some members would have to repay before they know the default status of other 5

6 members. In addition, those who have already repaid will have a greater incentive to urge their peers to repay. However, this strategy is not effective in reducing the domino effect if the first borrower defaults. It can take several months for a group to be recognized or certified by the Grameen Bank. In order to be recognized, all five prospective borrowers must undergo at least seven days of training on Grameen Bank s policies and demonstrate their understanding of those policies in an oral exam administered by a senior bank official. Once all members pass the exam, the first loan is issued. Grameen s total number of borrowers is approximately 2.6 million, 94% of whom are women. There are 1,181 branches in 42,127 villages. The total amount of loans disbursed by the Bank since its inception is about 3.9 billion, of which 3.6 billion has been repaid. Most loans last one year with weekly mandatory installments. Repayment starts one week after the loan is issued. The interest rate is 20%. Repayment amounts to 2 taka per week for every 1,000 taka of the loan amount. Total weekly household expenditures average about 700 to 900 taka. The loan amount for the Bank ranges from 1,000 to 16,000 taka, while the average is about 5,000 taka. The average loan accounts for about 1.2% of a household s total annual income. These micro-credit organizations usually have some form of mandatory savings to help borrowers break the vicious cycle of poverty and become self-sufficient. There are also methods of buffering idiosyncratic shocks to some degree. Grameen requires its borrowers to deposit 5% of each loan in a group fund. Any borrower can take an interest free loan from the group fund not exceeding half the fund s total, provided that all members approve of the amount and its usage. Members can withdraw their savings account only after ten years of membership. This provides an added incentive to avoid default on subsequent loans since the savings balance grows the longer one stays with the program and proves her credit worthiness but is frozen once one defaults. By the end of 1998, group savings reached $162 million, of which $152 million were saved by women (Grameen Bank). The other two programs addressed in this study are BRAC and BRDB. BRDB is the largest public sector agency in micro-credit lending in Bangladesh. Between 1991 and 1998, BRDB distributed billion Taka as micro-credit to 1.6 million beneficiaries through 0.12 million groups in 440 thanas or districts (Banglapedia). While Grameen and BRAC both target women largely, the ratio of male to female borrowers is evenly divided. BRAC, 6

7 a NGO, was originally set up in 1972 by its founder executive director Fazle Hassan Abed to provide relief and rehabilitation assistance to refugees returning from India after the War of Independence in Later, BRAC turned its focus to poverty alleviation in rural areas. In addition to micro-credit, BRAC provides a number of other services to the rural poor, including healthcare, education and training programs. It places particular emphasis on the training of its members in trades and income generating activities. BRAC s credit program was initiated in As of December 2002, the organization had disbursed $1.8 billion among its members to develop income generating projects or small enterprises. Members are also encouraged to save regularly. The savings deposited with BRAC amounted to $86 million in 2000 (Banglapedia). 2.2 Discussion of Intra-household resource allocation There have been considerable debate and conflict over the notion of treating household as a single unit, where preferences are aggregated, or as a collection of individuals with different preferences. The traditional approach assumes that a household behaves as a single decision-making unit. Thus, given a household budget constraint, one can derive household demands by maximizing household preferences. This refers to the unitary model 2. Paul Samuelson and Gary Becker were among the first to provide justification for this type of model. Samuelson (1956) sets structure on the household decision-making process such that the household utility function collapses to a unitary one, i.e. in which preferences and resources are aggregated. He assumes a weakly separable household utility function as a function of individual utility functions or subutilities. This aggregation would be achieved by consensus among household members, although he never discusses how this consensus maybe reached. Becker (1974, 1981) uses a similar approach. However, he suggests there be a benevolent dictator or head in the family who takes other members preferences into account and allocates resources accordingly. In his rotten kid theorem, he posits that any one member will not try to gain at the expense of the benefactor or any other members in fear that the benefactor will lower his transfer to that member by more than the potential gain the member can achieve. Ergo, no member will behave rottenly and the preferences of the head will represent the preferences of the household. 2 Other names have been proposed: benevolent dictator model etc 7

8 Formally, a household with two partners or two working-age adults will maximize the aggregate welfare function similar to Bergson-Samuelson welfare function W = W [U m (X, l; µ), (U f (X, l; µ)] (1) where U m is the male partner s utility and U f is the female partner s utility that are strongly quasi-concave, increasing, and twice continuously differentiable functions in their arguments. These include household consumption vector X = (x 1,, x n ) and leisure l = (l m, l f ). Finally, µ is the household specific heterogeneity. Household resource constraint can be shown through a budget constraint: P X + W m l m + W f l f Y m + Y f + Y h + W m T + W f T (2) where P = (P 1,, P n ) is the price vector, W i and Y i are the wage rate and nonlabor income for individual i, respectively, Y h is the total household nonlabor income that cannot be assigned to any member, and T is the total time endowment. The unitary model of the household may have two different interpretations. The simplest one implicitly assumes that individuals have equivalent preferences. This implies that the sub-utility functions in (1) are identical. Another interpretation is that of the benevolent head expounded by Becker in which one individual makes all the resource allocation decisions. In this case, the aggregate welfare function W (.) represents the utility of that member where he assigns zero weights to all but own-utility function. Consequently, heterogeneity in preferences would not affect resource allocation and thus household demands would depend on the total household income, not individual income. Then maximizing (1) subject to (2) will result in n + 2 differentiable Marshallian demand functions for goods and labor supply: 3 X = X(Y t + W m T + W f T, P ) (3) 3 The demand functions must satisfy i) Walrus law ii) homogeneity iii) slutsky symmetry, and iv) negative semidefiniteness. However, empirically, only walrus law has been proven. The other three restrictions have been rejected repeatedly in various empirical studies. See Deaton and Muellbauer (1980), Blundell (1988), Fortin and Lacroix (1997), Browning and Chiappori (1998]) etc. for further discussion. However, these rejections have been attributed to data and specification problems or functional form rather than erroneous theory of consumer behavior 8

9 where X = (X, l ) denotes consumption and leisure bundles and P = (P, W m, W f ) is the price vector. Y t = Y m + Y f + Y h is the aggregate nonlabor income. It is important to note that the unitary model implies income pooling hypothesis where individual nonlabor income are pooled into Y t. This asserts that the source of the income does not affect intrahousehold resource allocation problem, i.e. changes in different income have the same effect on household demands : dx/dy m = dx/dy f = dx/dy h. This assertion, along with equality of cross-substitution effects on labor supply, has been strongly rejected in various empirical studies (Ashenfelter and Heckman, 1974; Bourguignon et al., 1993; Browning et al., 1994; Fortin and Lacroix, 1997; Lundberg et al., 1997; Thomas, (1990, 1993) among many others). As an alternative to the unitary model, two other approaches have been proposed that take into account the individual role in household decision-making, i.e. demands, with heterogeneous preferences. The first of these approaches models household behavior in a non-cooperative framework (see, e.g., Ashworth and Ulph, 1981; Ulph, 1988; Carter and Katz, 1997; Lundberg and Pollack, 1993; Browning, 2000; and Chen and Woolley, 2001). This framework assumes that individuals cannot enter into binding and enforceable contract and thus maximize their utility, taking actions of other members as given. This Nash equilibrium results in more restrictions on household behavior than the unitary approach and suboptimal or Pareto inefficient resource allocation. The second approach presents a cooperative framework with cooperative game theory concepts and axiomatic bargaining theories. Household members cooperate with each other to divide the gains from living together, i.e. gains from cooperation, and a Pareto efficient allocation is obtained depending on the bargaining power of the household members. Thus, this type of model imposes more structure on the household allocation process. Manser and Brown (1980) consider the Nash and the Kalai-Smorodinsky solutions to derive the implications on intrahousehold resource allocation, while Bjorn and Vuong (1985) (cited in Thomas, 1993) consider a Stackleberg concept. McElroy and Horney (1981) focus on Nash bargaining solution to derive conditions for Nash demand system that collapse to traditional unitary model. In their model, the household maximizes the utilities less the threat-point utility, which is the highest utility a member would get if he or she withdraws from the household. Therefore, in a two person 9

10 setting, the household utility if given by the following utility-gain product function : N = [U m (X, l) V m ( P m, Y m ; α)][u f (X, l) V f ( P f, Y f ; α)] (4) where V k represents the threat point utility. Therefore, household maximizes the product of the gains from marriage. The α are the relevant shift parameter or the extraenvironmental parameters (EEPs). These are the changes in the opportunities outside marriage such as changes in divorce and alimony laws, sex ratio, assets at marriage, parental income, or income opportunities in developing countries. Any changes in alpha would affect the threat point and thus would shift the bargaining power of the individual within marriage, where the threat is dissolution of marriage. The household commodity and leisure demand functions are obtained by maximizing (4) subject to (2) with respect to X: X = X(Y m, Y f, P ) (5) A criticism of this approach is the imposition of specific structure on allocation decisions which are based on the assumed threat point. Since a particular bargaining concept has to be chosen to model household behavior, if the empirical results are rejected, then it is impossible to determine whether the particular choice itself or the bargaining setting is rejected (Vermeulen, 2002). Therefore, Chiappori (1988a, 1992) and Apps and Recs (1988) suggest an alternative model of collective framework. According to this framework, household utility is defined as the weighted sum of the utilities of individual members in the households. These Pareto weights can be interpreted as proxies for bargaining power for each individual in the household. The only assumption of the collective model is that the solution to the household allocation process is Pareto efficient. There is no restriction on which point will be chosen on the Pareto frontier by the household. Since so little structure is imposed, the model has testable restrictions. Numerous studies have applied the bargaining theories for empirical tests in an attempt to disprove the unitary model. Bourguignon et al. (1993) have rejected the income pooling hypothesis to find that share of husbands and wives own income affect the consumption structure within the household significantly. Lundberg et al. (1997) use exogenous EEPs 10

11 such as the transfer of child allowance from men to women in U.K. in late 1970s and find that such a transfer increases women s and children s clothing expenditure relative to that of men s. Thomas et al. (2002) and Quisumbing and Briere (2000) find that mothers with current asset and more assets at marriage (as indications of more power) will devote more resources toward their children s clothing and education and their children s health, respectively, than would fathers. Other empirical evidence suggest that households in which women have higher levels on non-labor income have healthier children conditional on total income (Thomas,1990; Schultz, 1990; Dufflo, 2003). Using a Brazilian budget data, Thomas (1993) finds that higher income ratio of women to men results in less spending on food but more spending on human capital such as health and education, as well as leisure activities such as recreation. These tests of income pooling hypothesis provide compelling evidence to reject the traditional unitary model. In addition, improving women s status in the household benefits household members, especially children and thus justifies many social programs practice of targeting women. One limitation to these household bargaining models and the empirical studies is that they assume the bargaining power or Pareto weight is determinant of exogenous factors only. Indeed, this assumption is necessary for the Pareto efficiency. However, in reality, the bargaining power of a woman need not depend on exogenous parameter. Consider a household with two members- a husband and wife- and define the Pareto weight of the wife as θ(z) in [0,1]. 4 In the traditional collective model discussed above, Z denotes set of variables that are exogenous to the household. Some examples include female assets, nonlabor income, changes in the divorce law, and sex ratio. However, Basu (2001) criticizes the exogeneity assumption and argues that θ may depend on the households choice variables. For example, the share of the household income earned by the woman would certainly influence her power over resource allocation. This share is endogenous since it depends on her labor supply, which is determined within the model. The endogenous bargaining parameters, Z, fail to achieve Pareto optimality; any solutions to these models would not be Pareto optimal. In this paper, I argue that income generated from micro-credit program participation will increase women s bargaining power or empowerment as evidenced from 4 Thus, the husband s bargaining or Pareto weight is given by 1 θ(z). This also represents the balance of power 11

12 changes in the resource allocation toward female-dominant goods when the participant is female. Thus, the determinant of the power is own-income, which is endogenous because it depends on her choice of labor supply. One of the contributions of this paper is that it provides for the endogeneity of the bargaining power, θ(z), by using a two-stage model that allows the Z vector to include endogenous factors, namely the income generated from the micro-credit program. As discussed in the previous section, micro-credit programs in rural Bangladesh are ideal for this study because they offer opportunities for women to earn an income. Since women do not usually work outside the house (social norms and lack of female labor market preclude rural women from working), they do not have any income and thus are dependent on the husbands, sons, and male relatives. However, programs such as micro-credit target women in rural areas and provide loans for entrepreneurial activities, while relaxing some of the rigid social barriers to women s mobility with their group-based lending scheme. 5 Over the past decade, there have been some household level studies regarding microcredit. The critiques of micro-credit programs claim that the loans are often controlled by men, regardless of the gender of the borrower. In this case, the program should not increase women s bargaining power and indeed may actually decrease it because they would still be responsible for repayment. Goetz and Sen Gupta (1996) find a negative relationship between micro-credit and women s empowerment by focusing on who controls the loan. They find that a significant portion of the married women as compared to widows have lost control of the loans, while some have no information as to the usage of the loans. In contrast, Hashemi et. al. (1996) show a positive relation between credit participation by women and women s empowerment in Bangladesh, in terms of increased ownership of assets, ability to make decisions and purchases, and decreased domestic violence. Other studies find a positive effect of micro-credit participation on contraceptive use (Schuler and Hashemi, 1994; Amin et. al, 1994; Schuler et. al, 1997), while Pitt et. al (1999) fail to find any significant effect. program placement and self-selection into the program. Most of these studies do not correct for nonrandom Pitt and Khandker (1998) use fixed effects and a quasi-experimental design to control for heterogeneity and selection bias 5 Most group members tend to be neighbors and friends who are frequently of same gender. Households are more willing to allow women to travel to group meetings and participate in such cases. In addition, most of these households are very poor and thus less averse to relaxing the social norms. 12

13 at the individual and village level. They find that the programs have greater impact on total household expenditures, savings, and assets when females borrow than when males borrow. However, these studies do not examine the changes in intra-household resource allocation due to credit program participation. The current paper provides an alternative to evaluating the credit programs. Saad (2007) examines the pecuniary and non-pecuniary impact of borrowing from the credit programs on the self-employment profits of the borrowers. The paper finds that borrowing from micro-credit programs lead to lower profits, even compared to borrowing from private sources. Considering these outcomes, policymakers may conclude that government and donor funds could be better spent in other programs. Thus, we need to extend the evaluation to other facets. Do micro-credit programs empower women, increase their power or say in the household, and change allocation? The purpose of this paper is to answer this question by examining the differential changes in the expenditure pattern, i.e. household resource allocation, as a result of the income generated by the gender of the participant of the credit program. In addition, previous studies use participation choice or the amount of credit to determine the impact. Neither can reveal the impact accurately since they do not take into account the interest rate and other pecuniary and time costs associated with the credit participation. If the loan is used for consumption, then ultimately, they would need to repay the loan with interest, which may leave them worse than before. In addition, if the loan is issued to women who use it for consumption, then the burden would be on the wage earner (usually the male) to repay back the loan. In such case, women may lose bargaining power and may have to reduce expenditure on goods that are of value to her. Even if the borrowers choose to invest the loans into some income-generating activities, using the amount of credit to study the impact may again be inaccurate. In case the profit generated from such activities is less than the total amount of repayment, the program may attribute undue benefit to the loan amount. The present study attempts rectify these problems by using net-income generated from the use of the loan for different purposes. 13

14 3 Theoretical framework The bargaining models discussed in the previous section assume that the bargaining power of the spouses is determined by parameters that are exogenous to the decision makers. In this section, I present a model that explains why and how intra-household bargaining influence resource allocation, i.e. decision making process, in the presence of endogenous parameters of the bargaining power. 6. It is assumed that the relative bargaining power of a woman depends on her share of total household income when she participates in incomegenerating activities. Specifically, higher share of total income would lead to greater power in household resource allocation. Since the bargaining power is endogenously determined according to individuals labor income, they take into account how their labor-supply decisions or income affect household decisions regarding consumption through the bargaining process. Thus, the household decisions can be modeled as a two-stage process. In the first stage, individuals or spouses decide on their labor supply by playing a non-cooperative or pure Nash equilibrium game, while realizing how their decisions jointly will affect their consumption decisions in the next stage. In the second stage, they choose their respective consumption levels using a collective model framework and maximizing a weighted utility function. Consider a household with two adult members-a husband and a wife-and that preferences are caring, i.e. one cares about other s consumption and leisure level (Becker, 1981). 7 The equilibrium outcome can be obtained by solving the system via backword induction, starting with the second stage where spouses make household decisions taking as given each other s labor supply from the first stage. In second stage, the household s optimization problem is given by the following: Max U = µu f (X f, X m, l f, l m, e) + (1 µ)u m (X f, X m, l f, l m, e) (6) subject to 6 The model in this paper was adopted from Iyigun and Walsh, (2002) 7 Note that an egoistic preference would include only own level of demand in the utility function (Chiappori, 1988) 14

15 P mx m + P f X f + P e e Y + N k + W m L m (7) l m + S m + L m T (8) l f + S f T (9) (X m, X f, l m, l f, e) 0 (10) µ = f(n m (S m ), N f (S f ), W m L m ; α) (11) Where U f (.) and U m (.) are twice continuously differentiable, strongly concave utility functions with the consumption vectors X m = (x m1,..., x mn ) and X f = (x f1,..., x fn ) for male and female goods, respectively. The l f denotes female and l m denotes male leisures and e denotes the level of education for their children. U k is assumed to be increasing in all arguments. Y is the household non-labor income and T is the total time endowment. S m and S f denote the time investments into a self-employment activity by male and female, respectively, while L m represents the time in market labor activity for the male. Since women do not have a formal labor market in rural Bangladesh, female time is divided between leisure and self-employment activity. Thus, any market labor income is due to the husband working and any female income is due to the self-employment returns. The time at work and time in self-employement activities are fixed at L m, S m, and S f in the second stage. The prices for the male goods, female goods, education, and male labor are given by P m, P f, P e, and W m, respectively. The N k denotes the net return from the self-employment activity from micro-credit when the participant is of gender k (m, f). This is a function of the time inputs, which are the choice variables in the first stage but taken as given in this stage. Finally, µ represents the bargaining power or welfare weights of the female within the household in the decision making process. As µ increases, more weight is placed on her utility relative to her husband s. In case µ = 1, the household 15

16 utility is solely determined by her utility. Conversely, when µ = 0, household utility equals the male utility. These limiting cases collapse the model to the traditional unitary model. The bargaining power, µ, need not be constant and may depend on exogenous variables as advanced by the traditional collective models and other bargaining models such as those with Nash solutions. Historically speaking, it has been a function of prices, wages and other exogenous variables. For example, changes in wages and non-labor income, assets at marriage, education, divorce law, sex ratio all may shift the power from one spouse to another by putting more weight on his or her utility and influencing allocation. Consider the Nash bargaining model given in equation (4). The household maximizes the utilities in marriage less the threat-point utility, which is the highest utility a member would get if he or she withdraws from the household. Anything that increases the threat point utility would lower the gains from marriage, thus giving the spouse power to negotiate his or her allocation. Otherwise, the threat is the dissolution of marriage. Since access to and participation in micro-credit programs may give women opportunities to earn income and thus increase her threat point utility, the Nash bargaining model has potential in this paper. 8 However, the net return from female participation, henceforth female return, is endogenously determined. This will increase her share of household income and thus her bargaining power, µ. In Nash bargaining model, an endogenously determined bargaining power results in Pareto efficiency breakdown. Same problem ensues with the traditional collective model. Thus, extending the collective model into a two-stage model provides an alternative in which the bargaining power can be endogenously determined. The equation (3.6), therefore, presents µ as function of the female returns, male returns, male labor earning, and α which contains other exogenous parameters such as education ratio. The Solution to the optimization problem would yield the following second stage demand functions: X m = X(S m, S f, L m, N k, Y, P, W, µ(n m (S m ), N f (S f ), W m L m ; α) (12) 8 The divorce rate has been on the rise in Bangladesh, both in urban and rural areas. It is more prevalent in poorer families. Social status and stigma is less important to these groups of people. Therefore, whereas women from affluent families stay in marriage (when they want divorce) to maintain status and avoid stigma, the women from lower class families have stayed in marriage for self-preservation because there were little opportunities for women s labor in which they can earn enough to support themselves and their children. As more opportunities have become available, they do not need to stay in the marriage if the gain from marriage is small. 16

17 X f = X(S m, S f, L m, N k, Y, P, W, µ(n m (S m ), N f (S f ), W m L m ; α) (13) e = X(S m, S f, L m, N k, Y, P, W, µ(n m (S m ), N f (S f ), W m L m ; α) (14) l m = T S m L m (15) l f = T S f (16) With (12)-(16), the couples can now solve a Nash equilibrium game, taking each other s labor supply as given. In the first stage, each spouse of gender k maximizes the following: U k (X m, X f, l m, l f, e ) (17) Substituting (12)-(16) into (17) and maximizing with respect to the labor supply variables yield three reaction functions: S f = s(s m, L m ), S m = s(s f, L m ) and L m = l(s f, S m ). Thus, the household equilibrium is as follows: S f = s(s m, L m) S m = s(s f, L m) L m = l(s f, S m) (18) X m = X(S m, S f, L m, N k, Y, P, W, µ(n m (S m), N f (S f ), W m L m; α) (19) X f = X(S m, S f, L m, N k, Y, P, W, µ(n m (S m), N f (S f ), W m L m; α) (20) e = X(S m, S f, L m, N k, Y, P, W, µ(n m (S m), N f (S f ), W m L m; α) (21) l m = T S m L m l f = T S f (22) 17

18 The equations (19) - (21) show that increase in female returns may have different effect of the demand than a increase in male returns. The aim of this paper is to show that an increase in female returns will increase X f by more than a similar increase in male returns. The reason is that with more income, it increases the female s bargaining power and thus would lead to greater allocation toward female-oriented goods. This is evidence of female empowerment arising from programs like micro-credit in rural areas that target mostly women. 4 Empirical model 4.1 Estimation method This paper examines the gender-specific impact of borrowing, especially borrowing from a micro-credit organization, on the intra-household decision-making process. To do this, I estimate two equations; one which determines a consumption decision and another that determines net returns. LogX ijn = H ij λ n + logincome ij λ 1n + N ij λ 2n + N ij G ij λ 3n + G ij λ 4n + V j λ 6n + η X ijn (23) N ij = H ij α + Z N ij α 1 + V j α 2 + η N ij (24) N ij = N ij if N ij > N o N ij = 0 if N ij N o LogIncome ij = H ij β + Z I ijβ 1 + V j α 2 + η I ij (25) H ij is a vector of household charateristics for household i in village j, Income ij is household income from working in the labor market plus any other income generating from household business. G ij is the proportion of female borrower in the household. If this equals 1, then all borrowers within the household are female and equals 0 when all male borrowing. The reason I take proportion is for some households, there are more than one 18

19 borrower, especially if the household is an extended family. V j is a village fixed effect, 1 if village=j j = 1,..., 87 V j = 0 otherwise Z N and Z I are vectors of instruments. I discuss these in the following section. Finally, N ij is the net returns or income generating from investment of the loans for the current year for household i in village j with gender k borrowing. This is different from profit from the business since it takes into account the cost the borrowing, the fact that one would have to pay the loan back with interest and other transaction fees. In addition, I distinguish the return if they have finished paying the loan more than a year ago or otherwise. I assume that a borrowing household has not finished paying if I observe the household still paying or if it finished paying any time within year prior to the time of the survey. The reason for this distinction is that the expenditure is given for the year. Therefore, if a household paid their loan anytime within the last year, for example two months ago, the effect of net returns on consumption will be biased if net return does not account for the cost of borrowing. In addition, I observe different purposes for the loan use, which in turn would affect net returns differently. Households either invest the loans in small business given by an indicator function I(invest = 1)or consume or I(consume = 1); loans in this case would have a consumption-smoothing effect. 9 Thus, π ij [(1 + r)c ij + F ]D f I(invest = 1) N ijk = ( [rc ij + F ]D f ) I(consume = 1) (26) where π ij is the profit from the business and C ij is the amount of loan borrowed by household i in village j, r is the interest rate of the loan, F is the other transaction costs. Finally, D f is a dummy variable that equals 1 if household has yet to finish paying or have paid back within the last year and 0 otherwise. Note that this specification allows for net returns to be negative or zero when the loan is used for consumption as would be expected. I use a two-stage lease squares fixed-effect (2SLS-FE) to correct for the self-selection 9 The net return may be more (less) if the durable goods appreciate (depreciate). I assume no appreciation or depreciation. 19

20 issue and non-random program placement. The self-selection arises due to the correlation between unobservable household-specific effects. For example, if households in which women are more empowered or have more bargaining power borrow, then any positive increase in the bargaining power may be erroneously attributed to the borrowing and thus income generated from the borrowing for households that borrow as compared to the household that do not. That is, the estimates will be upwardly biased. Therefore, the variance-covariance matrix of the error terms in the models (ηijn X, ηn ij, ηi ij ) σ XX σ XN σ XI σ NX σ NN σ NI σ IX σ IN σ II shows that the covariance terms are not zero. Therefore, it is necessary to treat N ij as endogenous. The IV approach solves endogeneity problem. However, there is also a sample selection issue since we do not observe N ij for those households that do not borrow. In addition, N ij depends on the use of the loan, conditional on borrowing. Some households use the loans for consumption purpose; they spend on durable and non-durable goods such as household repair and appliances. Thus, each household has a choice of not borrow, borrow for investment, or borrow for consumption. Sample selection problem is solved by using a Multinomial choice model and calculating inverse mills ratios using results of Heckman (1978) and Lee (1983). First, I estimate the participation decision, which shows the three different choices a household has, and obtain the predicted probabilities from Multinomial Logit regression using maximum likelihood estimation. Once I obtain the estimated probabilities, I can construct inverse mills ratios for each choice using the method suggested by Lee (1983) as follows: λ i = φ[φ 1 (P i )] P i i = 1, 2, 3 With the base group to be not borrowing, I use the two inverse mills ratios as additional regressors in equation (23) and (24). The standard errors will be bootstrapped. A Cramer- Ridder likelihood ratio test was performed to see if the three choices are same as two choices of borrowing and not borrowing. Table reports the test statistic [CM(109) = 433.3], which 20

21 rejects the null hypothesis that the decision to investment and decision to consume the loans are similar. The other source of bias is due to non-random program place. Program officials admit that programs are typically placed in poorer and more flood prone areas of the country. It is possible that they would place programs in villages that may be more liberal in their treatment of women and their roles in the household. If some villages are more openminded and less traditional about women s role in the household, better outcomes would result from targeting women in those villages. Thus, treating program placement as random may lead to erroneous and biased estimates (Pitt, Rosenzweig, and Gibbons, 1993). Thus the estimation must control for village-specific unobservables. I employ village fixed effects (FE) technique to address the bias due to program placement. 4.2 Identification I employ an instrumental variable approach to address the selection bias due to self-selection by the households. Eligible households that reside in a village where one of the programs is offered can choose whether or not to participate into a micro-credit program. The decision to participate is thus endogenous. However, these households with choice do not provide the necessary identification since unobserved characteristics such as high regards for women or empowerment that affect the decision to participate may also affect allocation of resources. Thus, measuring the program impact by comparing expenditures of participants and non-participants may confound participation effect with these unobservables. The main identification comes from the gender-based restriction. Households are exogenously excluded through their residence in non-program villages. Since men can join men only groups and women can join women only groups, the gender-based restriction is enforceable and observable. Therefore, I use the gender-based program design to identify program effects on self-employment profits. A second program eligibility rule is the land restriction. Only households with 0.5 acre land or less are eligible for the programs. In this study, I exclude all households with land more acre to aid the comparison between the eligible participants and eligible non-participants in treated and control group. One potential problem of such exclusion is the issue that this eligibility criterion was not strictly 21

22 enforced. The data contains program households with land amount above the cut off; 25% of the program participants have more than 0.5 of land. Thus, excluding these households may create selection problems. I discuss this in section VIC. Denote the first element of the instruments used in equation (24) as Z 1 ij Z1kj N 1 if village j has s programs of k gender = 0 otherwise s=gb, BRAC, BRDB and k=male, female I also include interactions between Z ij1 and household exogenous variables such as education, age, number of people, land. I use amount of land as an interaction because landholdings serve as proxy for wealth. This term will identify the change in the effect of net of returns as the household s wealth changes. A second instrument is the length of time the program has been available in the village. This should affect household expenditure only through the participation and thus through the income generated from participation. The interest rate and the distance to the lending institution serve as additional instruments. The participants were asked the main reasons they decided to join the micro-credit program. 65% replied that it is because the programs offered relatively cheaper credit; 20% said the access to credit is easier. Another reason for participation is the lack of collateral. There may be a second reason a household may have a female participant rather than a male participant. Men and women on average face different wait time to obtain the loan. Whereas men wait 14 weeks to access the loan, women wait 9 weeks on average. I use the length of wait for the loan as an instrument. The borrowers also have to bear the time cost of joining; they have to attend training and group meetings. These can influence the decision to participate but not the household expenditures on goods and thus serve as additional instruments. The households were asked if they have close relatives who are alive and living in the same village and how many of them own land. I add these as instruments as well because these relatives are potential transferees and may impact their participation decision. To the extent that they do not affect household expenditures by contributing to the income, they are valid instruments. Finally, I include some village level variables such as the nearest distance to the bank, shop, bus stop, male wages, female wages, and child wages as instruments. 22

23 Exclusion restriction holds if above instruments affect the main outcome only through the income generated from participation. They must be correlated with the endogenous variables but uncorrelated with the main outcome. Table 7 shows the relevance test of the first stage. The null hypothesis of no significance is rejected at 1 percent significance level for (24) and (25). The instruments are valid predictors of the endogenous regressors. The second is an overidentification test to show that the instruments are independent of the errors terms in the main outcome equation (23). The table provides the LM and Hausman test and shows that the null hypothesis of proper exclusion restriction, i.e. the chosen instruments are independent of the error, cannot be rejected. Therefore, the exclusion restrictions are valid. 5 Data 5.1 Description The data was collected in an extensive multipurpose quasi-experimental survey in Bangladesh commissioned by the Bangladesh Institute of Development Studies-World Bank in 1991/ 1992 and a follow-up in 1998/ It was designed to study the impact on borrowers welfare from three group lending programs: Grameen Bank, BRAC, and BRDB. The survey interviewed 1,798 households in 87 villages of 29 thanas (sub-districts) selected at random from 391 thanas, 24 of which had at least one of the three credit programs under study in operation, while 5 thanas had none of them. Three villages in each program thana were then randomly selected. Three villages from each non-program thana were also randomly selected from the village census taken by the government of Bangladesh. Only programs that had been in operation for at least three years were included. The survey thus has three groups of people: a target group and a non-target group separated by the exogenous eligibility requirement of land ownership of less than 0.5 acre. Within the target group, there are participants in one of the three credit programs and non-participants who chose not to participate even though they were eligible for the credit. Thus, the three groups are target-participants, target non-participants, and non-target. In a number of the villages surveyed, there were micro-credit programs for both men and women. However, some of the villages had only male or female credit groups. Table 1 shows 23