Economic Impact of Migration on a Rural Area in Bangladesh

Economic Impact of Migration on a Rural Area in Bangladesh Pierre Yves Beaudouin February 2, 2006 Abstract The main objective of the paper is to analyze the effects of migration on sending countries. The objectives are to analyze the direct and indirect effects of migration on the migrant household income, to measure the opposed effects and to discuss the policy implication of our results. The study is based on a Three Stage Least Squares estimator to determine and measure the net impact of migration on the household income. The empirical study show a negative effect of the loss of labor on the total income. However, this effect is compensate by remittances sent home by migrants. When we decompose the income in three sources of income (farm, self-employed and wage), we find these two opposed effects for the farm income, but no effect of remittances on wage income and only a negative effect of migration for self-employed income earned by the household. Keywords: Migration, Remittances, NEML, Bangladesh JEL Classification: O15, R23, D13 Centre d Économie de la Sorbonne, Université Paris 1 Panthéon-Sorbonne, CNRS, 106-112 boulevard de l Hopital, 75647 Paris Cedex 13, France. Tél: +(33)144078243. Fax: +(33)144078337. Email: pierre. beaudouin@malix.univ-paris1.fr, Homepage:http://team.univ-paris1.fr/teamperso/beaudouin. I thank participants of the AFSE seminar Economic Development and Transition on 19-20 May 2005 for helpful comments. 1

1 Introduction The United Nations estimates the stock of international migrants about 175 million in 2000 1. The proportion of migrants in the world is growing, international migrants represented 2.5 per cent of the world population in 1960 and 2.9 per cent in 2000. This trend should not change in the next decades despite migration laws, institutional structures and tightening of procedures. Despite these small figures, migration may have some important developmental effects on developping countries. One of the main canal of transmission of migration on sending countries is remittances flow sent back home by migrants to their family. Remittances are now view as a possible source to finance development. Dilip Ratha estimated that remittances going to developing countries represents $150 billion in 2004. This amount is bigger than foreign aid and is the first source of foreign capital for many countries. This flow of formal remittances 2 from migrants to their relatives in their country of birth has exhibited a high rate of growth. The doubling of remittances flow 3 during the last decade has increase the study of remittances. The literature on the impact of migration is biased on the receiving countries. Most of the studies on migration are on the receiving countries economies and the effect on sending countries has been neglected. One of the reason is the absence of data on migration. But the increase of remittance flows last years, which is one of the most important channel of the impact of migration on sending countries, has developed a recent growing literature on the effects of migration on sending countries. If non poor household participate to the migration process, like it seems to happen, a study without controlling for selectivity bias will overestimates the impacts of migration. Most of the literature study the effects of migration only on one part and the effects of remittances in another part. Most of the time, migration studies does not take into account the effects of remittances when studying the effects of migration on sending countries. Likewise a lot of papers on the effects of remittances does not take into account the real or potential selectivity of migrant. As migrants may represent a non randomly chosen subset of the overall sample, an econometric estimation by Ordinary Least Squares (OLS) becomes problematic if migrant and non migrant individuals differ systematically in their productive capabilities (individuals characteristics like education, experience, 1 Figures are from the World Migration Report (IOM, 2005, p.379) 2 Official statistics only take into account remittances flows transiting by formal channels (bank or wire transfer companies like Western union or Money gram). As some surveys found that the part of remittances transiting by informal ways is important, remittance flows seems to be underestimate. Another limit of recorded remittance flow is that because these figures comes from the balance of payments, by definition they only take into account remittances send by overseas migrants. 3 This increase of remittances flow is more the consequence of a improvment of the statistics than a real increase of money send back to family. 2

gender... ). But there is few studies which take into account the linkages between migration and remittances. The size of remittance flows do not allow us, for the study of the effects of migration, to use the classical migration theories (Lewis, Harris and Todaro) as these theories did not take into account remittances in the migration strategy. Analysis on remittances does not take into account the determinants of migration. The only way to take into account in the analysis the migration decision and remittances effects is to use the NELM framework. This paper is based on the framework developed by the new economics of labor migration (NELM) which analyzes migration as part of a strategy to diversify the sources of income of the household in presence of market failures (Stark 1991; Stark and Bloom 1986). Futur remittance flows are taken into account in the decision of migration. And migration is viewed as an allocation of labor like non-farm labor or crop production. Another particularity of this paper is to analyze the effects of migration on all the economic activities of the household and not only on a particular sector, like it is done in the previous literature. For example, Azam and Gubert (2002) study the impact of migration on the agricultural sector. There is also some empirical study on the question of the links between migration, remittances and business creation. Amuedo-Dorantes and Pozo (2002) showed that remittances has no statistically significant impact on business investment in the Dominican Republic Using a particularly rich household survey on Bangladesh, the objective of this paper is to test the impact of migration on rural communities. Does migration reduce crop income and/or crop yield? Does migration have an impact on other sources of income? If yes, is it the same effects than for crop production? Finally, are the NELM hypothesis validate? The paper is organized as follows. Section 2 presents the framework of the NELM. Section 3 presents the data used in this paper and describe some descriptive statistics of our household survey. Section 4 presents the empirical strategy to test our model. Section 5 presents the results and section 6 concludes the paper. 2 Theoretical background The literature on migration is not new. Already in 1889, Ravenstein presents some laws of migration. The analysis based on the work of Lewis (1954) and Harris and Todaro (1970) have permit the study of some aspects of migration like the determinants of migration, the effects of migration on the receiving countries or the impact of migration on the labor market of the sending countries. But unfortunately theses models are not 3

relevant to analyze the effects of migration on sending countries. The trademark of the NELM is that migration decisions are taken by larger units of related people - typically the househould, in the contrary to neoclassical theories who take the migrant as the unit of decision(stark, 1991) not only to maximize the expected income, but also to minimize risks and loosen constraints created by a variety of market failures, including missing or incomplete capital, insurance, and labor markets. The persisting bounds between migrants and their households of origin lead to a reject of the individual-level migration decision model (like the implicit migration model of Lewis (1954) or Harris and Todaro (1970)) and to opt for a household model. The household provides the funds to finance the cost of migration (transportation, fees charged by recruitment agencies, fees to obtain a visa and work permit, maintenance while searching for work... ) and in return, once migrants become established, the migrant share a part of its income by sending back some money and goods. These repeated interactions lead us to prefer household model to explain the decisions of migration Taylor and Martin (2001). So NELM focuses on a particular kind of migration: the short-term labor migration, with migrants keeping strong tie with their family of origin and who are expecting to return rather than settle in the host country (Sana and Massey, 2005). Taking the household as the unit of analysis is a mean to understand migration as a risk minimisation strategy. The migration of labor is analyzes as an exportation of a part of the labor factor of the household. Migration is now an element of the diversification of the sources of income of the household. In rural area of developping countries, crop is still the first income generating activity. But crop production suffers from many hazards (aléas de la production et fluctuation des cours des produits agricoles). The migration decision is no more view only as an income maximisation but also as a risk minimisation. In presence of market failures or missing of credit and insurance markets, household members should find an informal method to insure against risk. According to the NELM, the migration of labor could be a method to insure against risk. An informal contracts are conclude between migrants and their households of origin. The second role of the migrant is to be a financial intermediary. Remittances could help to relax the liquidity constraint of the household by increasing investment in new technologies of production, for example by introducing high yield varieties or by developing non-farm business (Massey et al., 1993). Migration by generating some income can help to increase the agricultural productivity. For the NELM framework, income differential is no more a necessary condition to the migration contrary to the standard approach initiated by Lewis (Stark and Katz, 1986). 4

The theoretical model developed in this section is based on the NEMT framework developed par Oded Stark. The analysis below is based on the works of Rozelle et al. (1999). The role of financial intermediaries play by the migrants can be illustrated by the figure 1. For simplification, we will use a unitary model of household. In this model, the household head is the only decision taker of the household. Consider a household who could invest a fix sum T, such as agricultural land, labour or factory. The household can invest in two possible production activities with a difference in returns; such Qi, with i = 0, 1, the production of the low and high-return activity, respectively. The production of each activities depend on the investment and of the individual characteristics Z (for example education or age), Q i = f(t i, Z). The total income is Y = g(q 0 + Q 1 ). We can draw a production possibility frontier (PPF) in a graph with x-axis Q 0 and y-axisq 1. For example, for a relative price of P = P 1 P 0, the household will invest all the sum T in the high-return activity. The output will be Q = f 1 ( T, Z y ) and the household income will be Y = g(q ). What are the effects of migration and remittances in this framework? In an environment of perfect market, remittances have no effect on the household production. The household model in the case of perfect markets is separable (Singh et al., 1986). The decisions of production are independent of the dotations and preferences of the household. On a theoretical point of view, production and consumption decisions can be analyzed as being taken successively. Firstly the household decides its production level, and after it deduce its consumption level from the production level decided before. The negative effects of migration can imply a negative impact on production (decrease of the labour force has for consequences the decrease of the production). But in this model, remittances have no effect because as an simple income transfer Remittances have an effect on the consumption of the household by allevitaing the budget constraint but have no effect on the production level of the household 4. One of the main determinant of migtration for NELM is the presence of market imperfections. In presence of imperfections, the household is credit constraint and can only invest T, with T 1 < T. For exemple, the household owns an arable land. To cultivate this land, the household needs to spend some money ex ante. The credit constraint is relaxed by the migration process. If the household could not spend this amount because of a lack of savings or imperfections of the credit market, then the household production would be constraint. The production constraint will be Q c 1 and Q c 0. Migration could 4 The remittances do not change the profit maximisation condition which determine the level of production. 5

Figure 1: Migration effects on production 6

help to relax this constraint. In the presence of market imperfections, migration could be a solution. The model is non separable in presence of market imperfections. Shadow prices are evaluated for non-market activities. These prices linked the production and the consumption spheres. Remittances could have now some effects on the household production sphere. The relaxation of the credit constraint could be represented by a shift upward of the resource constraint line Q c 1. However the net effect of migration is may be not positive. The constrained output is Q c 1 = f 1 (T 1, Z) for the high-return technology and Q c 0 = f 0 ( T T 1, Z) for the technology with low productivity. The constrained household income is so: Y c = g(q c 1 + Q c 0) with Y c < Y the non-constrained income. The overall effect is ambiguous as the partial effects of migration and remitttances, c(m) and c(r), are undetermine. However, we can make a hypothesis on the value of the coeeficients. The non-separability of the model involve the non nullity of the coefficients c(m) and c(r). Test if these coefficient are statisticalley significant than zero would be a support or an unfirmation of NELM. Few tests exist of the NELM hypothesis. However we can cite Lucas (1987); Taylor (1992); Taylor and Wyatt (1996); Rozelle et al. (1999); Taylor et al. (2003). This paper will test the NELM hypothesis by developping an estimator. 3 Empirical Methodology As mentioned below, migration is a strategy to increase the income of the household or to insure the household against risks. But the household can also diversify its production by developing non-farm activities or taking an employment. Migration is in competition with other strategies to attract assets. But migration could alleviate constraints on production. In developing countries, household are usually constrained by imperfection of markets like a labor constraint, a lack of liquidity or or a credit constraint. Remittances could have a positive effect on the household by alleviating these constraints. Migration can extend the investment realize by the household. Migration can also help the household to diversify its activities. Migration has a potential negative effect. The lost of labor can decreases the production of the household if the productivity do not increases. Agricultural activities still represents the major part of the income of the household. However this activity can particularly suffer from migration. This negative impact of migration on crop production has been found by Azam and Gubert (2004) in their study on the Soninke migration and also by Mochebelele and Winter-Nelson (2000) in their analysis of migration in Lesotho. Both papers conclude of a larger farm technical efficiency of non-migrant household. 7

Previous literature testing the effect of migration on sending countries suffer from some limitations. An approach used in the literature is based on descriptive statistics. Authors use the answer to the question of the use of how do the household use remittances. Surveys usually find remittances are used in housing construction. However, remittances can free up other sources of income that may be used for other means. A survey which find that the principal use of remittances is in the housing construction can conclude of an unproductive use of remittances. But because remittances are fungible, The presence of fungibility of the resources prevent us using this method. Another approach is to examine an outcome of interest by comparing households who receive remittances with households that do not. We can estimate the impact of remittances on an outcome of interest and controlling for a set of migrant, household and community characteristics through ordinary least squares (OLS) regression. Outcome i = δ + φ Remittances i + γ X i + ɛ i However, this approach could suffer from omitted variable bias. If the only effect of migration on the outcome of interest is through remittances, the results are not biased. But if migration has other impacts on the outcome in addition to its effect through remittances, then the error terms contains omitted variable that are correlated with remittances and with the outcome. This will result in a bias of the results. That is why we replace in the previous equation the remittances variable by a variable of migration, to test the overall impact of migration. Outcome i = α + β Migration i + χ Remittances i + η X i + ɛ i The coefficient β then captures the joint impact of remittances and of other consequences of migration through the migrant variable. The production is constrain because of the presence of market imperfections. This as for result that (Y c ) depends on migration (M) and remittances (R). A vector of individual, household and village characteristics may also affect the income. In order to test if migration and remittances have different effects according to the source of income, the empirical analysis will distinguish three distinct sources of income: crop production Y c, self-employed income Y s and wage Y w 5. (1) Y c k = α 0k + α 1k M + α 2k R + α 3k Z k + ɛ Yk ; k = a, s, w As mentioned before a potential bias of endogeneity is possible. We control for this bias by the instrumental variable method 6. 5 The total income is the sum of the three sources of income and the remittances, Y = Y a +Y s +Y w +R 6 A test of endogeneity has been completed. A joint test that the coefficient of the residual are statically different from zero 8

Migration depends of the individual, household and village characteristics, (Z M ): (2) M = β 0 + β 1 Z M + ɛ M The dependent variable is the number of migrants. (Z R ): Remittances are a function of migration, individual and household characteristics (3) R = γ 0 + γ 1 M + γ 2 Z R + ɛ R After the presentation of the theoretical model and our empirical method to test the model, the next section will present the specification of the system of equation. 3.1 Variable specification The survey use the following definition to define a member of a household. A householder is a person who has been living in the household for at least 6 months or less than 6 months but plans to stay in the household for more than half a year. So a person who has been living at least 6 months outside the household 7 or who moved out of the household for marriage of migrated overseas will be defined as a migrant. Remittances include cash and in-kind transfers 8 Following the NELM theory, migration and remittances variables are linked with the household income. To control this endogeneity bias, we choose the instrumental variable method. The migration network is an instrument in the migration equation(taylor and Wyatt, 1996). Village migration network is approximate by the number of migrants by village. We postulate that the variable have a positive effect on the number of migrants but have no effect on the income nor on the remittances (minus the number of migrants of the household). In our preferred specification, we decompose the network variable into an internal migrant network composed by migrants living in Bangladesh and external migrant network composed by migrants living outside the country. In the remittances equation, the norm of the village to remit is an instrument. We use the annual amount of remittances received by a household of the village to proxy the norm of the village. This instrument is proposed by Taylor et al. (2003). This variable should affect the level of remittances received by the household but has no effect on the income of the household. 7 without plans to stay more longer in the household 8 whatever the method of transfer. The great advantage of using a survey is that remittances include transfer transiting by banks or money trabsfer firms like Western Union or Money gram (formal transfer) and money transiting by hand, hundi (informal transfer). 9

3.2 Functional form Equations (1) through (3) constitute an equation system. Error terms ɛ i, i = Y, R, M are normally and identically distributed with variance σi 2. A correlation of the error terms is likely to occur between the income equations and remittances. An exogene shock can have an effect on the five equations. To account for contemporaneous correlation, we choose to estimate our model by the three-stage least squares. This estimation method consist of an estimation of each equation by the two-stage least squares and a correction of the error terms to take into account the possible correlation of the error terms. 4 Data The data use in this paper, Matlab Health and Socio-Economic Survey (MHSS), have been collected by RAND Corporation in 1996. In spite of the goal of the survey was not migration, this survey includes a lot of information on our subject. The dataset contains informations on non-resident household members 9 (their relationship with the household head, age, marital status, education level, location, occupation, remittances send back to home... ). The dataset includes four separate surveys with different samples. The main survey consists of household- and individual-level information on 4,364 households. The second survey is on the determinants of natural fertility survey. The third survey is on internal migrants and the last survey consists of a community survey. This paper use the main survey which consists of 4,364 households distributed on 140 villages. Matlab is a rural area in the south-west part of Bangladesh. Since 1966, a program of demographic surveillance under the Centre for Health and Population Research (ICDDR B, formerly known as International Centre for Diarrhoeal Disease Research, Bangladesh) is in place in this area. The sample of MHSS benefit from this program as the sample is based on previous surveys and census realized by ICCDR B. Moreover the 9 We have information on the spouse, children, parents or sibling of the household head who are migrants. Table 1: Number of household migrating Status Household Percentage Number of household with at least one migrant 1,992 45.65% Number of household with at least one migrant abroad 609 13.96% Any migrant 2,372 54.35% Total 4,364 100% Source: MHSS (1996) 10

Table 2: Distribution of migrants by relationship and remitting status Relationship Number of migrants Remit Total Internal External Total Percentage Spouses 353 203 158 284 80.45% Children 3,883 3,572 1,332 2,143 55.19% Parents 1,015 950 91 83 8.18% Siblings 15,164 12,954 2,210 2,620 17.28% Total 20,415 5,130 25.13% Source: MHSS (1996) experienced field organization and the respondent population accustomed answering to surveys lead to very low attrition rates Matlab is located about 55 kilometers Dhaka, the country s capital and one of the city with the highest growth rate of the world 10. In spite of this geographical proximity, a trip between Matlab and Dhaka takes six hours preventing return trips. This is a cause of urban migration (Kuhn, 2001). Destinations of overseas migrants 11 can be basically separate in three groups: Developing countries, Persian Gulf and Southeast Asian countries. These destination countries for migrants have changed over the years. The first destination was Great Britain. The second destination was Middle east countries (Saudi Arabia, Kuwait, UAE... ) in the seventies. The last destination is southeast countries (Malaysia, Brunei... ). In contrast to the migration to developing countries, migration to the Middle East and South East Asia has been characterised by short term employment, with specific job contracts and migrants returning home after completion of the contract period IOM (2005). Informal migration is so very difficult, the main way to become an illegal migrant is to stay in the country after the end of the labor contract. This survey collected data on such diverse topics as income, expenditures, education, employment, food consumption, health and nutrition, landownings, assets, migration and rural credit.... The particularity of the dataset that is interesting for us is the information on migrants: education, remittances, employment, location.... The survey also include a village survey. 4.1 Descriptive statistics A little bit less than half of households in the survey have at least one migrant, and 13.96% have at least one migrant abroad (table 1). The data does not allow us to know precisely in which country resided migrants, but in general Bangladeshis migrate to Middle East 10 The population of Dhaka was 1.3 million before the independance (1971) and 8.5 million in 1997. 11 Figures are available on the SAMReN website http://www.samren.org/facts a nd F igures/bangladesh/1.1.htm 11

Table 3: Remittances Internal External Total Spouses 4395.733 30002.18 12921.53 Children 2412.394 6106.651 9109.04 Parents 21.910 7.412 110.735 Siblings 569.027 104.573 939.772 Total 2541.451 12531.9 7166.73 Source: MHSS (1996) countries (Saudi Arabia, Kuwait, Oman, Qatar... ) and more recently to South East Asian countries (Malaysia, Singapore... ). The dataset included information on different type of migrant: spouse of the household head, its children, parents and siblings. Table 2 shows statistics on remittances received by surveyed households. The incidence of remittances vary a lot. 80% of household head spouses send money to their spouse. More than half of the children of the household head remit. This figure fall to 17.28% and 8.18% for parents and siblings of the household head. This lead us to restrict our analysis to spouse and children migrants. A migrant household is defined, in the following of the paper, as a household having at least one person (the spouse of the household head or its children) who was previously a member of the household has left for more than six months to live or work elsewhere, either in Bangladesh or abroad. The average respondent received tk7,166.733 per year from migrants (Table 3). But this figure vary a lot between the destination of migration and the relationship of the migrant. International migrants send more money back home as they earn more abroad. Also, the spouse of the household head or these children send more money than parents or siblings. A household who actually had a migrant son or a migrant spouse received tk4,395.733 from internal migrant and tk30,0002.18 from international migrants. In contrast, siblings living outside the district or the country account for a very little sum of transfer. Table 4 shows the number of migrants per household. On average a household count 4.68 migrants, with a minimum of 1 and a maximum of 29. Table 5 presents the means of variables used in the regression analysis, broken down by migration status. The household size and the composition of the household are not significantly different between migrant household and non-migrant household. The human capital is more important in migrant household as non farm assets, but it is not the case for farm assets and the agricultural land per capita own by the household. A significant difference comes out from comparing the two sub-samples regarding the amount of remit- 12

Table 4: Number of migrants and amount of remittances received per household Number of Mean Number Mean Number migrants amount of of amount of of within remittances households remittances households the household received received 0 0 422 0 3,755 1 499 1686 24,221 427 2 538 2749 35,223 123 3 516 4612 62,335 40 4 468 5569 63,884 16 5 418 5715 26,866 3 >5 1503 10674 - - Total 5898 4,364 4,186 4,364 Source: MHSS (1996) tances received by the household. This is not a surprising figure, because children and spouses have the higher propensity to send remittances and they send a larger amount of money than siblings. The fact that we restrict our analysis to overseas migration is also an explanation of this high amount of remittances send back home by migrants. Households with migrants abroad get in general about 83% more income than the others. But this figure is not sufficient to answer the question of the impact of migration. That is why in the following sections we develop an econometric analysis to answer this question. A p-value of less than 0.05 means that the null-hypothesis of equal means for both groups can be rejected at an error level of less than 5 percent [H0: Differences in means = 0]. 13

Table 5: Descriptive statistics Variable Non migrant Migrant t P> t household household (N=3755) (N=609) (*) Household size 5.516 5.84-3.187 0.0014 Dependence ratio 0.403 0.381 2.269 0.0233 Education 2.901 3.804-5.522 0.0000 Experience 38.922 43.656-7.563 0.0000 Non productive assets 10206.81 27970.1-7.663 0.0000 Land per capita 0.091 0.152-1.516 0.1295 Number of cows 0.069 0.080-1.023 0.1889 Farm assets, lagged 501.039 789.557-1.349 0.1774 Non farm assets, lagged 8843.42 38660.83-4.008 0.0001 Remittances 2238.482 32587.45-23.812 0.0000 Total income 40967.45 81102.68-11.1164 0.0000 N 4364 Source: MHSS (1996) 5 Results 5.1 Migration Estimations of the determinants of migration are presented in table 6. Results are consistent with previous empirical findings. Results are quiet similar between each specification. The first specification is a simple OLS, the others are Poisson estimations 12. Specifications 2 and 3 only change in the variable that we belive can act as instruments to identify migration. The first set of regressors are household demographic characteristics. One would expect that the number of migrants would rose as the size of the household rose and fall as the the number of dependents rose. The coefficient of the household size has a statistically significant impact on the number of migrants. However, the dependance ratio has no statistically significant impact. Education and experience variables have both a statistically significant and positive impact on the number of migrants. The coefficient of land per capita is not statistically significant. But the proxy variable for wealth, the value of non-productive assets, has a positive effect on migration. This could reflect the fact that migration has a cost. The migration networks variables include in the regression have a statistically significant impact on the number of migrants. More precisely, the coefficient of the total 12 The OLS estimator is biased, because the migration variable is non-negative and there is a lot of null observation. 14

number of migrants of the village is positive. But when we break down this figure between the number of migrants living in the country and the number of migrants living abroad, we found a different effect between these two variables. The external migration networks has a positive effect on the number of migrants, but contrary to Mendola (2005) who found a positive but smaller effect, we cannot conclude to a effect different from zero of the internal migration networks on the migration variable. The final set of regressors are village characteristics. The negative coefficient of this dummy reflect the fact that a better environment decrease the incitation to migrate. The coefficient of the village size is negative. The presence of markets in the village have no significant impact on the number of migrants. 5.2 Three Stage Least Squares 5.2.1 Remittances Estimations of the Three Stage Least Squares are presented in table 7. The first column represents the remittances equation. Only few coefficients are statistically significant. Wealthier households (measured by the value of the value of the nonproductive assets owned by the household) receive more remittances. The village norms to remit has a positive coefficient. Only for the specification with the different source of income, the coefficient of the dependence ratio is statistically significant and positive. A household living in a village with a stronger norm to remit will receive more remittances than a household staying in a village with a less strong norm. The coefficient of the migration variable is positive. Each additional migrant increase remittance by Taka 17,892. This support the descriptive statistics presented in the previous section. 5.2.2 Total income and sources of income Results are quite similar between total income regression and sources income regression (table 7 and 8). Remittances have a positive effect on income, except on the wage income. Migration do not have any significant effect on the total income. However, crop income falls when migrants leave the household. This negative effect is even bigger on wage income. With the departure of a household member, the labor force of the household decreases, and the income also decreases. Coefficients for the other exogenous variables affect income in ways that are consistent with findings by other researchers. The household siez has a positive effect on both total income and income by sources. It is also the same with the number of years of education of the household head, except for the wage income where its impact is non statistically 15

significant. Stock of capital (farm and non-farm) positive effects on income, like the amount of land per capita. Although the coefficients of the village size are generally insignificant, the population of the village negatively affects crop income. This is the same for large and small markets which negatively affects respectively crop and self-employed income. 16

Table 6: Determinants of migration Explanatory variables Number of migrants (1) (2) (3) Household characteristics Household size 0.013 0.060 0.062 (0.001) (0.000) (0.000) Dependance ratio 0.440 0.104 0.121 (0.281) (0.524) (0.460) Education 0.011 0.073 0.069 (0.000) (0.000) (0.000) Experience 0.056 0.028 0.028 (0.000) (0.000) (0.000) Non productive assets 0.000 0.000 0.000 (0.000) (0.000) (0.000) Land per capita 0.002 0.006 0.006 (0.833) (0.850) (0.847) Village characteristics Village size -0.000-0.001-0.000 (0.003) (0.000) (0.006) Large market 0.017 0.098 0.014 (0.396) (0.354) (0.899) Small market 0.009 0.058 0.071 (0.732) (0.472) (0.389) Total Migrant -0.005-0.008 (0.871) (0.946) Internal migrant 0.093 (0.542) External migrant N 4236 Significativity levels : : 10% : 5% : 1% 1.255 (0.001) 17

Table 7: Estimated effects of migration and remittances on total income Explanatory variables Number of migrants Remittances Total income Number of migrants 17892.79-7200.424 (0.000) (0.109) Remittances 1.245 (0.000) Household characteristics Household size 0.062 4702.144 (0.000) (0.000) Dependance ratio 0.119-1494.178 (0.467) (0.362) Education 0.069 1480.925 (0.000) (0.000) Experience 0.028 2.967 (0.000) (0.966) Non productive assets 0.000 0.077 0.311 (0.000) (0.000) (0.000) Land per capita 0.006-157.568 1327.261 (0.833) (0.679) (0.038) Farm assets, lagged Non farm assets, lagged 0.910 (0.000) 0.089 (0.000) Village characteristics Village size -0.000 0.982-4.963 (0.005) (0.502) (0.204) Large market 0.025 543.444 265.778 (0.817) (0.670) (0.946) Small market 0.082 1064.253-9177.352 (.307) (0.248) (0.000) Number of industries Industry External migrant Village remittances 1.271 (0.000) 0.822 (0.000) Remittance facilities -1386.039 (0.274) Mills ratio 1154.361 (0.531) 481.722 (0.038) N 4236 Significativity levels : : 10% : 5% : 1% 18

Table 8: Estimated effects of migration and remittances on total income Explanatory variables Number of migrants Remittances Agricultural income Wage income Self-employed income Number of migrants 17969.6-4130.601-7755.191 245.460 (0.000) (0.000) (0.015) (0.892) Remittances 0.289 0.068 0.151 (0.000) (0.669 (0.047) Household characteristics Household size 0.062 903.461 3079.052 380.850 (0.000) (0.000) (0.000) (0.028) Dependance ratio 0.119-13696.819 (0.467) (0.011) Education 0.069 199.911 1066.357 25.244 (0.000) (0.000) (0.000) (0.825) Experience 0.028 28.802-79.113-65.364 (0.000) (0.034) (0.130) (0.021) Non productive assets 0.000 0.078-0.016 0.210 0.016 (0.000) (0.000) (0.009) (0.000) (0.266) Land per capita 0.006-163.133 2036.253 940.365-159.320 (0.833) (0.669) (0.002) (0.217) (0.700) Farm assets, lagged 0.216 (0.000) Non farm assets, lagged 0.029 (0.000) Livestock -0.030 (0.580) Village characteristics Village size -0.000 0.651-1.256-3.468-1.202 (0.005) (0.656) (0.086) (0.217) (0.429) Large market 0.025-846.649-1129.921-703.346 867.480 (0.817) (0.496) (0.056) (0.812) (0.590) Small market 0.082 880.057-190.590-8099.859-1116.084 (.307) (0.329) (0.665) (0.000) (0.247) Number of industries 703.766 (175.052) Industry 722.295-32.171 (0.000) (0.736) External migrant 1.271 (0.000) Village remittances 0.758 (0.000) Remittance facilities -619.180 (0.581) Mills ratio 1644.284 (0.368) N 4236 Significativity levels : : 10% : 5% : 1% 19

6 Conclusion This study used a rich dataset from rural Bangladesh to examine the different effects of overseas migration and remittances on the different sources of income of the household. Three conclusions emerge from it. First, our results supports the NELM framework that remittances are an element of a household strategy to alleviate market failures. Secondly the loss of labor affect negatively the household income. Finally the impact of migration is complex. It should be break down into two different effect: the lost of labor and the amount of remittances received. Effects also differ from income sources. Migration have no effect on self-employed income and remittances have no statistical impact on wage income. 20

Annex 1. Variable definition Household characteristics Number of migrants: a person (the spouse of the household head or its children) who was previously a member of the household has left for more than six months to live or work elsewhere Remittances: value of remittances send by migrants Household size: size of the household Dependence ratio: (children 0-14 years + adults > 65 years)/persons 15-64 years Education: number of years of education of the household head Experience: number of years of experience (Experience=Age-Education-16 based on work of Jacob Mincer, Schooling, Experience, and Earnings (New York: Columbia University Press, 1974) Non productive assets: value of non productive assets Land per capita: agricultural land owned by the household per capita Farm assets, lagged: value of farm assets owned by the household in 1995 Non farm assets, lagged: value of non farm assets owned by the household in 1995 Village characteristics Village size: size of the village Large market: dummy variable, 1 if a large market is present in the village Small market: dummy variable, 1 if a small market is present in the village Number of industries: number of industries in the village Total migrant: number of migrant in the village minus the number of migrants of the household Internal migrant: number of internal migrant in the village minus the number of migrants of the household External migrant: number of external migrant in the village minus the number of migrants of the household Village remittances: mean value of remittances received by the village minus the amount of remittances received by the household Remitances facilities: dummy variable, 1 if there is a bank or a post in the village 21

References Amuedo-Dorantes, C. and S. Pozo (2002). Workers Remittances and the Real Exchange Rate: A Paradox of Gifts. mimeo. Azam, J. P. and F. Gubert (2002). Those in Kayes. The Impact of Remittances on their Recipients in Africa. Document de Travail 11, DIAL. IOM (2005). World Migration Report. Technical Report, IOM. Kuhn, R. (2001). Never Far from Home: Parental Assets and Migrant Transfers in Matlab, Bangladesh. RAND Labor and Population Working Papers 01 12. Lucas, R. E. (1987). Emigration to South Africa s mines. American Economic Review. 77 (3), 313 330. Massey, D. S., J. Arango, G. Hugo, A. Kouaouci, A. Pellegrino, and J. E. Taylor (1993). Theories of International Migration: a Review and Appraisal. Population and Development Review. 19 (3), 431 466. Mendola, M. (2005). Migration and Technological Change in Rural Households: Complement or Substitutes? Working Paper 2005-15. Rozelle, S., E. J. Taylor, and A. de Brauw (1999). Migration, Remittances and Agricultural Productivity in China. American Economic Review. 89 (2), 287 291. Sana, M. and D. S. Massey (2005). Household Composition, Family Migration, and Community Context: Migrant Remittances in Four Countries. Social Science Quaterly. June 86 (2), 509 528. Singh, L. Squire, and J. Strauss (1986). Agricultural household models, extensions, apllications, and policy, Stark, O. (1991). The Migration of Labor, Basil Blackwell, Cambridge, U.K. and E. Katz (1986). Labor Migration and Risk Aversion in Less Developed Countries. Journal of Labor Economics. 4 (1), 134 149. Taylor, J. and T. Wyatt (1996). The Shadow Value of Migrant Remittances, Income and Inequality in a Household Farm Economy. Journal of Development Studies. 32 (6), 899 912. Taylor, J. E. (1992). Remittances and Inequality Reconsidered: Direct, Indirect and Intertemporal effects. Journal of Policy Modeling. 14 (2), 187 208. 22

and P. L. Martin (2001). Human Capital: Migration and Rural Population Change. in B. L. Gardner and G. C. Rausser, eds., Handbook of Agricultural Economics, Elsevier., S. Rozelle, and S. de Brauw (2003). Migration and Incomes in Source Communities: A new Economics of Migration Perspective from China. Economic Development and Cultural Change. Octobre 52 (1), 75 101. 23