Effects of Remittances on Household Expenditure Patterns of Rural Mexico. José Jorge Mora Rivera ITESM, Campus Ciudad de México

Transcription

1 Effects of Remittances on Houseold Expenditure Patterns of Rural Mexico José Jorge Mora Rivera ITESM, Campus Ciudad de México Jesús Arellano González El Colegio de México Te Macmillan Center at Yale University Hewlett Foundation Marc 2009

2 1. Introduction Te number of people living in a country different from teir birtplace as been growing and accounts for approximately tree percent of te global population. According to te United Nations, in 2005 around 200 million people were migrants (UN, 2006). In 2000, Mexico was first place wit net migration of 2 millions of people, followed by Cina wit 1.95 million people living abroad (World Bank, 2005). Tis migrant flow of migrants all around te world creates an international labor market tat leads to a flow of monetary resources tat, in certain proportion, is reallocated to te origin countries in te form of. In 2004, at a global scale, te total amount of reaced billion dollars of wic 64.2% was sent to less developed regions being Asia, East Europe and Latin America te geograpic regions wit te iger proportions; 38%, 32% and 18% respectively (UN, 2006). Te relationsip between te total amount of international and te less developed countries is noticeable. From 1990 to 2003 te growt of as been extraordinary. Latin America contains several countries tat ave experienced a considerable growt in received during tese 14 years (World Bank, 2005). In te case of Mexico, according to El Banco de México, in 1990, international received were around 2.5 billion dollars. By 2007, tis number increased to 26 billion dollars. Tis represents an annual growt rate of 15.2%. Because of te current global recession, uncertainty concerning remittance flows towards developing countries as increased. Trends are not clear so far. In October 2008 remittance flows to Mexico reaced an istorical maximum of 2.6 billion dollars. In contrast, te lowest value ever recorded was seen in January 2009 (1.6 billion dollars) as a result of te global economic turndown. Total in 2008 are 3.6% lower tan tose in However, te decline in remittance flows is expected to be lower tan tat of private and official aid flows. Migrant flows are also expected to decrease but not te current stock of international migrants. (Rata et. al., 2008). Te nature of remittance flows (motivated mainly by altruism towards family, especially in ard times) and devaluation of currencies in several recipient countries are very likely to maintain important for developing economies. Te extraordinary growt of international migration and te monetary flow associated wit tis penomenon as motivated a great number of social researcers to study te diverse effects tat migration migt ave in origin and destination countries. A basic question is if weter or not ave an effect on te economic development of migrant-sending countries. Several studies conclude tat tere is no automatic mecanism troug wic migration and te inflow of monetary resources in te form of elp to improve te economic development of origin regions. How and to wat extent migration and can better perform tis function is a question tat must be present in any researc agenda about te subject. Remittances sent to te origin country represent an important resource tat can be devoted to te creation of pysical and uman capital and tus, a mean to promote te development of origin regions. But beyond teir quantitative importance, te possible impact of 2

3 sould be viewed in terms of teir use in a diverse context since tey can ave multiplier effects on te local economy and even modify te migration dynamics. A fundamental issue wen studying te effects of migration on rural development in origin countries is to know te impact of internal and international on te expenditure patterns of remittance-receiving ouseolds. Te decision making process of ow to spend a limited budget can be different wen ouseolds receive no. Te impact tat te migration penomenon may ave over te incomes, expenditures and productive activities of rural ouseolds as been of particular interest among researcers. Tere are several questions to be answered. One of tem, wic constitutes te main purpose of tis study, is to determine ow affect te monetary resources allocated to certain expenditure categories, especially tose measuring pysical and uman capital investments. Moreover, our interest is to distinguis te impact of internal versus tat of international, mainly from te United States. A key question tat must be also considered is te impact tat may ave over productive investments since tey ave been considered part of te growt and development engine of rural communities and could be viewed as a strategy to decrease te need for future migration flows. Some studies ave concluded tat are devoted mainly to current consumption instead of being invested, wic little impact on te productive use tat suc resources may ave (Cami et al., 2003; Taylor et al., 1996; Durand and Massey, 1992; and Papademetrious and Martin, 1991). In te literature tere are at least tree views on ow are spent and impact economic development. Te first is based on remittance use surveys, wic ask remittancereceiving ouseolds wat goods and services tey spent teir on. Most of te time, a distinction between te effects of remittance income and oter income is not made, assuming tat a dollar increase in remittance income as te same effect of a dollar increase of wage or farm income, and te contribution of to development will be te same as tat from any oter source of income. Remittance-use studies make te mistake of assuming tat ouseold income is completely fungible. Houseolds can distinguis te nature of different income sources attributing tem to different uses and managing tem using separate accounting (Duflo and Udry, 2004). Because of tis, tey provide little insigt into te ways in wic actually influence expenditure patterns in remittancereceiving ouseolds (Cami et al., 2003). A second view argues tat te receipt of can cause beavioral canges at te ouseold level tat may lower teir development impact relative to te receipt of income from oter sources (Baram and Boucer, 1998). Because of a moral azard problem between remitters and recipients, te dependency on tese transfers induces recipients to use as substitute for oter income sources. External socks may lower income from oter sources increasing te dependency on remittance transfers in te future, and since tey do not represent a capital flow, tis may reduce economic activity and growt (Cami et al., 2003). A tird more recent set of studies uses an econometric approac, adding remittance income as an explanatory variable in a system of ouseold demand equations. Tat is, demand is 3

4 modeled as a function of not only income, prices, and socio-demograpic variables but also te amount of remittance income ouseolds receive. Tis view of argues tat actually increase investments in uman and pysical capital at te margin, relative to oter forms of ouseold income. Examples include Adams (2005 and 1998), Edwards and Ureta (2003), Yang (2005) and Alderman (1996). Our researc argues and offers empirical evidence tat (internal and external) resape rural ouseolds expenditure patterns in direct and indirect ways. Te modeling approac we employ controls for censoring on ouseold consumption categories wile testing for differences in expenditure patterns between ouseolds receiving and tose tat do not. Tese models are estimated for bot external and internal. Te data to estimate te model is from te Encuesta Nacional de Ingresos y Gastos de los Hogares (ENIGH) Tis paper is organized as follows. Section two presents some insigts about te possible relationsip between and te expenditures tat rural ouseolds make. Section tree offers an overview of possible ways of estimating te impact of on different ouseold expenditure categories. Te fourt section presents te distinct specifications proposed for te empirical model. Te fift section contains a detailed description of te ENIGH data set. Te sixt section presents te main results. Section seven displays our conclusions. 2. Remittances and Expenditures in Rural Houseolds Empirical researc on expenditures in migrant-sending ouseolds often as contributed to a pessimistic view of te impact of migration on development in migrant-sending areas. Suc studies conclude tat are consumed instead of invested and tus are not put to productive use in migrant-sending areas (for reviews, see Cami et al., 2003; Taylor et al., 1996; Durand and Massey, 1992; and Papademetrious and Martin, 1991). Tis past researc on remittance use offers a partial and possibly distorted view of ow influence demand, due to te fungibility of income. Moreover, it often rests on arbitrary definitions of wat constitutes productive investments. For example, scooling often is absent from te list of productive investments. Tis is probably because expenditures on educating family members usually do not create direct, immediate employment and income linkages witin migrant-sending economies. Housing expenditures are also not considered productive investments in many studies, despite teir potentially important effects on mobility, family ealt and teir stimulus to village construction activities. Reported use of for productive investment at times can be significant. In teir review of studies carried out in Mexico, for example, Durand and Massey (1992) found tat te relative sare of spent on productive activities, altoug always under 50 percent, fluctuated considerably from place to place and often reaced substantial levels. Remittances enabled many communities to overcome capital constraints to finance public works projects suc as parks, curces, scools, electrification, road construction, and sewers (Reicert, 1981; Massey et al., 1987; Goldring, 1990). Oter studies report tat ave been critical to te capitalization of migrant-owned businesses. Escobar and Martinez (1990), for example, found tat 31 percent of migrants surveyed in Guadalajara 4

5 used U.S. savings to set up a business. Massey et al. (1987), in teir survey of te same city, put te figure at 21 percent; and in a survey of businesses located in tree rural Mexican communities, Cornelius (1990) found tat 61 percent were founded wit U.S. earnings. A number of studies from oter world regions eco tese findings (for a detailed review, see Taylor, et al., 1996.) Under te rigt circumstances, ten, a significant percentage of migrant and savings may be devoted to productive enterprises. Rater tan concluding tat migration inevitably leads to dependency and a lack of development, it is more appropriate to ask wy productive investment occurs in some communities and not in oters. Durand and Massey (1992) conclude tat, in Mexico te igest levels of business formation and investment occur in urban communities, rural communities wit access to urban markets, or rural communities wit favorable agricultural conditions. Negative findings on te productive impacts of may be attributable in part to poor researc designs tat do not consider te direct and indirect ways in wic may affect rural ouseold expenditures. Recent empirical models ave been designed to overcome tis problem. 3. Estimating te Impacts of Remittances on Expenditure Categories Most models of ouseold expenditures assume tat ouseolds allocate teir budgets across expenditure categories so as to maximize te utility obtained from te consumption of goods and services, eiter presently or, in te case of investment expenditures, in te future. Wit te exception of a new empirical literature on intra-ouseold resource allocation models, most consumer models assume tat ouseolds pool teir income. Tis leads tem to ignore income-source effects. Te solution to suc a consumer model is a set of expenditure functions of te following form: e = f ( P, Y, Z ) + u (1) i i were te subscripts and i refer to ouseold and expenditure category, respectively; denotes expenditure on good i by ouseold ; P is a vector of prices faced by te ouseold; Y is ouseold income; Z represents oter variables influencing marginal utilities and constraints on ouseold beavior, and ui is an error term tat is assumed to be 2 approximately normally distributed wit mean zero and varianceσ. In te standard consumer model, for a ouseold wit K diverse sources of income (including ), income is te pooled sum of income from tese sources: e i K Y = y k k= 1 (2) 5

6 Combining equations (1) and (2), it is evident tat a marginal cange in income from a given source k (say, ) as te same effect on expenditures as a marginal cange in any oter income source: ei f ( P, Y, Z ) Y f ( P, Y, Z ) = = y Y y Y k ' k ' (3) Oter tings being equal, an increase in from migrants sifts remittancereceiving ouseolds budget constraints outward by te amount of te remittance transfer. Tis raises (decreases) te demand for normal (inferior) goods. In tis model, te influence of migrant is assumed to be limited to indirect effects operating troug total income; income-source effects are ruled out. Recent studies by Adams (2005 and 1998), Zarate-Hoyos (2004) and Alderman (1996) add a new explanatory variable to te rigt-and-side of equation (1): ouseold income from migrant R, were R is also included in Y and can contain bot internal and external as single or independent variables. Tat is, e = f + (4) i ' ( P, E, Z, R ) ui were as in most demand studies, total expenditures marginal effect of a cange in remittance income, good i is tus: ei f ( ) f ( ) = + y E y k ' k ' E are used in lieu of income. Te, on ouseold s expenditure on f ( ) Tis is te same as only if tere are no direct effects of on expenditures. E In practice, a dummy variable indicating ouseolds receipt of, rater tan te level of, is used. Following tis approac and including interactions between te remittance-receipt variable and oter variables in equation (4), Adams found evidence tat te spending beavior of rural Guatemalan ouseolds wit was significantly different tan tat of ouseolds witout. Specifically, ouseolds wit remittance income spent less on consumption goods tan oterwise similar ouseolds witout remittance income, dispelling te notion tat are conspicuously consumed. Tis implies tat te second term on te rigt and side of equation (5) is nonzero. Similar results are reported in Adams (1998) and Alderman (1996) using data from oter less developed countries. Constraints on ouseold expenditures include not only income but also information, uncertainty, risk aversion, and preferences. If migrants provide ouseolds wit information, tis may ave various effects on expenditures, for example, by broadening te consumption set, creating a demand for new goods (e.g., nutrition), or altering ouseold production tecnologies (i.e., better ways of producing goods at ome). Information from migrants y k ' (5) 6

7 in tis way may loosen uman capital constraints on ouseold production, investment, and consumption activities, wile peraps influencing preferences, as well. Even if migrants did not contribute to income, teir contact wit an economy and society foreign to te village migt influence village preferences and demands. Consumption is saped, at least in part, by reference groups and identities. As rural peasants are brougt into te global economy bot troug teir participation in wage work and increasing reliance on from oter family members, and troug teir increased consumption of non-local commodities teir expenditure patterns cange, reflecting bot te influence of new cultural standards and a reorganization of finances witin te family farm. If te ouseold is risk-averse and are not perfectly correlated wit oter income sources, te effect of on consumption and investments in an uncertain world is likely to be different from te effect of te income wit different risk profiles. For example, ouseolds would be expected to allocate income from a risky source, like crop production, more conservatively tan income from, if te latter are viewed as more certain. Differences in te effects of income from different sources in tis case would reflect te influence of risk and uncertainty on ouseold utility from various consumption and investment coices. Even if te variability of migration income is greater tan te variability of farm income, income from migration neverteless may reduce total ouseold income risk troug a low (or peraps negative) correlation wit farm income. Te effects tat internal and external ave on risk and uncertainty of ouseolds may also be different. It is reasonable to assume tat internal are associated wit a lower level of riskiness tan external. Tis is because of te lower variability tat internal may reflect due to te nature of internal versus external migrants (internal migrants are typically more educated tan external migrants) or te macroeconomics socks wit wic external flows are associated (e.g. revaluations of local currencies). Remittance income may be perceived as less transitory tan income from oter sources (Suarez and Avellaneda, 2007). A permanent flow of may encourage ouseolds to invest in goods wose use and upkeep require additional purcases in te future (e.g., fuel for a new veicle). Income from migrants also may be controlled by different ouseold members tan income from oter sources. In tis case, a non-unitary ouseold model migt predict differences in marginal expenditures across income sources, reflecting te preferences and influence witin te ouseold of tose wo receive income from a given source (e.g., see McElroy, 1990; Scultz, 1990; Udry, 1996). Te data and empirical modeling approac to determine te possible effects of on ouseold expenditure patterns in rural Mexico are described below Data Information on ouseold expenditures and income sources was obtained from te Encuesta Nacional de Ingresos y Gastos de los Hogares (ENIGH) carried out by te 7

8 Instituto Nacional de Estadística, Geografía e Informática (INEGI). It is a nation-wide survey collected on a two-year basis from 1992 until ENIGH was also collected in 2005 giving us 8 years of information. Te sampling design of te ENIGH guaranties a representative cross-section of Mexico at te national and rural/urban levels. Te rural sector is defined as localities wit no more tan 2,500 inabitants. Tere are altogeter 37,505 ouseolds located in rural localities and observations can be grouped into Mexico s 5 geograpic regions: souteast, center, western, norteast and nortwest. It includes a very wide disaggregation of expenditure and income records on weekly, tree-mont-long and six-mont-long periods. A very ric dataset on ouseolds socio-demograpic caracteristics is also available. We are aware tat tis survey is not designed to properly study te migration penomenon and tus it poorly defines te concept of. Wat we call in tis study is recorded in te ENIGH as income coming from oter countries or income coming from oter Mexican ouseolds. Toug we ypotesize tat most of tis income represents sent by migrants, we are aware tat tis concept may also include oter kinds of income suc as money earned abroad by professionals or even money coming from a next-door ouseold. Furtermore, tis concept is not restricted to income coming from te United States. Neverteless, we empasize tat our focus is on te rural sector and tus, it is very likely tat te traditional Mexico-US migration penomenon is present in te sample and, accordingly, most of tis income is expected to be coming from te United States. Anoter sortcoming of te ENIGH data set is te lack of information about ouseold members abroad tat migt be sending money. Te only ting we see is te amount of money received by te recipient ouseolds but we know noting about te caracteristics of te sender (age, gender, education, place of living, etc), a ouseold member wo migrated internally or internationally. It would be ideal to ave suc information on migrants caracteristics and deal wit te selectivity issue affecting te decision to migrate. However, our focus is restricted to study te way in wic ouseolds decide to spend, wit te processes generating migration and taken as given under te assumption tat would be treated just like any oter income and taken into account in te ouseold s expenditure decisions. We take te standardized values of te tree-mont-long period income and expenditures provided by te ENIGH as well as teir non-monetary counterparts to define income and expenditure categories (including totals). Te reason to include te non-monetary measures is tat self-produced goods represent a very important part of income and consumption. Income and expenditure records were adjusted to 2002 prices and divided by ouseold size to obtain tree-mont-long period measures in per capita terms. Table 1 presents some summary statistics for te sample, wit ouseolds divided by teir remittance-receiving condition. In all, 5,465 ouseolds received internal wile 3,071 ouseolds received external. Tese numbers represent 14.5% and 8.2% of te sample, respectively. We found 456 ouseolds (1.2%) tat receive bot kinds of. 8

9 Several interesting contrasts emerge wen comparing different groups. For instance, te proportion of ouseolds eaded by a female is iger in remittance-receiving ouseolds (above 25%) tan in non- ouseolds (9.6%). 78.4% of non- ouseolds are eaded by a member wo reads and writes; in contrast, tis proportion is always below 63% for any of te remittance-receiving groups. Houseold eads in te non group are more likely to work in agriculture tan tose in te remittancereceiving categories. Interestingly, all of te education indicators (ead s scooling, maximum scooling, members wit secondary and ig scool completed) are iger for te non-remittance receiving ouseolds. Tis seems to be in conflict wit te New Economics of Labor Migration (NELM) (Taylor et al., 2001) wic posits tat better educated people find a iger reward in migrant labor markets and tus are more likely to migrate. However, it is possible tat te remittance-senders ave iger education tan te rest of te family members wo stayed beind. Again, it would be ideal to ave information on migrant caracteristics to test tis ypotesis. Educational levels are extremely low in rural areas. Despite te fact of aving iger levels of education compared to remittance-receiving ouseolds, ouseold eads in te non-remittance category reac on average only 4.25 years of scooling. Even te average maximum scooling acieved is only enoug to ave completed primary education. Remittance-receiving ouseolds ave more members over age 59 and fewer cildren below 7 tan non-remittance ouseolds; furter, ouseold eads are older in remittancereceiving ouseolds. Tis compliments wit te NELM wic claims tat migrants are younger tan teir non-migrants counterparts. Access to medical insurance is iger for non-remittance ouseolds; on average 0.29 members ave medical insurance (eiter public or private) as part of teir job benefits. External ouseolds ave, on average, more veicles tan non-remittance ouseolds, 0.40 against 0.28, respectively. In general, tere is evidence tat remittance-receiving ouseolds ave better access to services (public water, drainage, electricity and pone service) tan non-remittance ouseolds. Finally, ouseolds receiving bot kinds of seem to ave a better access to credit tan non-remittance ouseolds as 14.7% got loans versus 9.7% for te latter category. Te income and remittance figures by ouseold categories are summarized in Table 2; data is presented in montly approximations. External-remittance ouseolds seem to be igly dependent on monetary resources coming from abroad, wit external representing on average 40% of teir total income, receiving pesos per capita 1 per mont. Tis figure is lower for internal-remittance ouseolds wic are dependent on for 24.4% of teir total income and receive on average 261 pesos per capita. Houseolds receiving bot kinds of seem to diversify dependence, but still teir sare of external in total income is sligtly iger, representing 25% versus 14.9% attributed to internal. It is interesting to note tat te total income of external-remittance ouseolds is iger tan te total income of non-remittance and internal-remittance ouseolds in most of te 1 Around US$39.2 per capita at te current excange rate (13.37 pesos per dollar). 9

10 years and, on average, over te complete period. For all years, ouseolds receiving bot kinds of ave te igest total income of all ouseold categories. Concerning te evolution of tese magnitudes over te years, we see a drastic reduction of total incomes immediately after te 1994 macroeconomic crisis tat affected Mexico. Tis income reduction was about 20.1% for te total sample. Mirroring Mexico s economic performance, internal also fell for internal-remittance ouseolds and ouseolds receiving bot kinds of. Conversely, external didn t fall eiter for external-remittance ouseolds or for ouseolds receiving bot. Tis mitigated te fall in total incomes wic only decreased by 5.1% and 12.7%, respectively. It is possible tat tis beavior is also due to te devaluation of te Mexican peso during tose years. From tis period onwards, we see a slow recovery of total incomes along wit a moderate increase in external and internal received. As mentioned previously, tis trend is still uncertain under te current global crisis conditions. For tis study, ouseold expenditures records ave been divided in ten categories: Food, Healt (medical services, medicines, ealt insurance), Education (tuition, materials, transportation) Durable Goods (furniture, ouseold equipment, audiovisual equipment, veicles) Non Durable Goods (ouseold cleaning items, personal care items, cloting) Oter (transportation, personal services, culture and entertainment, veicle services, fuels and services, gifts, oter expenditures and transfers), Patrimony (additional constructions/renovations, purcases of ouses and land), Business (purcases of macinery or animals for te production process), Savings (deposits, currencies, metals, stocks and bonds) and Out transfers (lending, debt service, insurance, ineritance, etc). Table 3 presents average budget sares for eac of te ten expenditure categories defined above. As expected, food occupies te igest proportion of total expenditure for all of te ouseold categories, wit 41.31% for te entire sample. However, tere seem to be interesting differences in budget sares across ouseold categories. Remittance receiving ouseolds devote significantly more of teir total expenditure to ealt care (always above 5%) tan non-remittance ouseolds but less to education. Houseolds receiving external and bot types of spend more on durable goods (3.62% and 4.21%, respectively) tan non-remittance ouseolds (2.67%). Houseolds receiving external ave a iger sare devoted to te patrimony category (2.25%) compared to non-remittance ouseolds (1.11%). Tere are no significant differences concerning te business category across ouseold categories, altoug non-remittance ouseolds seem to allocate a iger sare tan internal- ouseolds. It seems tat external create incentives to save as sown by te significantly iger sare of savings for te external and bot remittance ouseolds, 5.97% and 4.27% respectively, compared to 3.50% for nonremittance ouseolds. Tese differences suggest tat expenditure beavior across ouseold categories is affected by remittance perception. To explore tis possibility, models depicted in te next section are applied to te ENIGH data for rural Mexico. 5. Our Empirical Model 10

11 A common problem wen dealing wit micro data, and especially wit disaggregated expenditure categories, is te existence of a large number of zeros in te dependent variable. Te reasons for te presence of zeros (see Garcia and Labeaga, 1996) are summarized as follows: 1) Infrequency of purcase: an issue in individual or ouseold surveys covering a sort period of time. 2) Abstention: due to individual or ouseold unobservable caracteristics tat prevent it from participating in a given market (selection model). 3) Corner solution: individuals and ouseold decide not to purcase a particular item because of active budget constraints, i. e. aving a genuine zero expenditure on te item. Wile putting aside te debate about te various reasons possibly limiting observations, we propose te use of tree different approaces to modeling ouseold demands, assuming tat all ouseolds wit zero expenditure are actually in a corner solution. First, we apply a standard Tobit model specification as follows: ' e * i / E = α i + β1 i ln( E ) + β 2iZ + β 3i Rr + ui (6) e i / E = 0 if e * i / E 0 e / E = e * / E if e * / 0 > 0 i i i E were e i / E is te sare of ouseold s expenditure on good i, and α i, β ki, k=1,...,3, are vector parameters. Te r subscript indicates internal and international. e * i / E is te corresponding latent variable governing te observability of a positive expenditure sare. Te use of a Tobit model in a single-equation framework affected by censorsip is straigtforward. However, in a system approac, suc as te one we adopt, censored regressions ave correlated error terms and estimation must be done jointly. Applying te Tobit tecnique to eac equation separately leads to inefficient estimators since it fails to take into account te interrelations across equations. So, in a context of a system of equations wit limited dependent variables te modeling of te data must be different. Teoretical literature about te subject exists. However, most employs a censored demand equation approac to model ouseold expenditures witout testing for remittance effects (Heien and Wessells (1990), Sonkwiler and Yen (1999), Perali and Cavas (2000), Gould et al. (2002), Lazaridis (2003) and Jabarin (2005)). We adopt te next two systems of equations approaces and utilize tem to capture remittance effects on te demand system. Te first specification involves a system of equations in wic te dependent variables, ouseold expenditure sares, are censored by unobserved latent variables influencing te decision to spend income on given consumption and investment goods. In bot systems, expenditure by ouseold on good i is observed (i.e., e i > 0 ) only if te ouseold's total desired expenditure on te item exceeds some tresold. Tis tresold will depend on te lumpiness of te good as well as te opportunity cost (te satisfaction or utility tat te ouseold would enjoy by spending tis tresold amount on some oter item). Bot, te 11

12 decision to spend income on a specific category of goods and te amount spent depend on te variables in equation (4) ( P, E, Z, R ). Assuming tat te stocastic errors are approximately normal wit zero means and a finite variance-covariance matrix tat is constant over all observations tat is, iid te system of expenditure equations can be estimated using Lee s (1978) generalization of Amemiya s (1974) two-step estimator to a system of equations. In te first stage, a probit is estimated for participation in eac expenditure category. Te dependent variable in eac probit is equal to 1 if e i > 0 and zero oterwise. Te probit models are used to calculate a set of Inverse-Mills ratios, one for eac expenditure category in wic censorsip is likely to be a problem: IMR i = -φ(x )/ Φ(X ) (7) were φ (X ) denotes te standard normal density function and Φ (X ) denotes te normal distribution function, and X is a vector containing E, Z and R. In te second step, te Inverse-Mills ratios are included as rigt-and-side variables in te corresponding expenditure equations to correct for censoring. We estimated te expenditure system using te Almost Ideal Demand System (AIDS) metod, extended to include demograpic caracteristics (Deaton and Muellbauer, 1980). Unfortunately, prices were not available in te data set and for tat reason tey are not included in our estimation. Te first system of equations estimated as te form: ' i / i β1 i ln( ) β 2i β 3i r 4i IMRi i e E = α + E + Z + R + β + u (8) e i / E is again te sare of ouseold s expenditure on good i, and i were α, β ki, k=1,...,4, are vector parameters. Te set of equations depicted in 8) was estimated using te seemingly unrelated regression (SUR) tecnique as proposed in Sonkwiler and Yen (1999). Tis functional form displays a number of advantages for tese purposes. It is flexible enoug to allow expenditure patterns to cange wit te total expenditure level. It permits us to estimate te marginal effect to as well as te marginal effect of oter variables on expenditures for eac category of goods. It also controls for some (lumpy) expenditure categories. Finally, it as attractive properties from a teoretical point of view, e.g., restrictions are easily imposed so tat it conforms to adding-up, omogeneity, and symmetry properties derived from standard demand teory (Lazaridis, 2003). Te second system approac consists of a version very similar to tat depicted previously wit some canges proposed by Perali and Cavas (2000) sowing tat it is possible to find a more efficient two-step estimator. Accordingly, te second system of demand estimations takes te form: e E = Φ( X )[ α + β ln( E ) + β Z + β R ] + β φ( X ) (9) i / i 1i 2i 3i r 4i 12

13 As above, φ (X ) denotes te standard normal density function and Φ (X ) denotes te normal distribution function. Again X is a vector containing E, Z and R. Te first stage of tis approac is to estimate φ ( ) and Φ ( ), using a Probit specification and ten, in te second stage, te use of functions φ( ) and Φ ( ) to correct te system of equations as depicted in 9). Tis second system of expenditure equations is also estimated wit te SUR tecnique. Te sare of eac ouseold expenditure category in total expenditure is regressed, according to our empirical metodologies previously depicted, against ouseold caracteristics and internal and external. Remittances were included as sares in total income. Dummy variables for years and geograpical regions were also included wit year 1992 and Region Sout as te base cases. Te list of variables used can be found in Table Results Table 5 presents results on te Tobit specification of equation (6). Houseold caracteristics were found significant in several cases and mostly according to expectations. For instance, a one-year increase of a ouseold ead s scooling and a one-member increase of ouseold members wit ig scool completed decreases te sare devoted to food by 0.21% and 0.93% respectively. Te presence of more people over 59 years old is associated wit a lower sare spent on food wile more cildren below 7 years old as a positive relation to it. A one-year increase in te ouseold ead s scooling increases te sare spent on education by 0.22%. Contrary to expectations, te number of ouseold members wit ig scool completed as a negative and significant effect on te education sare. A ouseold ead working in agriculture is associated wit a decrease of te sare devoted to education by 0.72% and a ouseold ead being male decreases te sare devoted to education by 2.02%. Having a ouseold member over 14 years old wo doesn t read and write is associated wit a 0.97% decrease in educational expenditures. Having a ouseold member studying in a private scool as a strong positive relation wit expenditures on education, increasing te sare devoted to it by 10%. Te number of elderly people over 59 years old and cildren below 7 years old in te ouseold increases te sare spent on ealt by 0.98% and 0.75%, respectively. A male ouseold ead and one working in agriculture is associated wit iger levels of expenditure devoted to durable goods. Access to medical security as well as to te provision of public services seems to provide an incentive as well increase expenditures devoted to durable goods, wit estimated coefficients of 0.18% and 0.87%. A one-veicle increase in te number of veicles possessed by te ouseold increases te sare devoted to durable goods by 1.64%. 13

14 Several ouseold variables ave significant effects on te non-durable goods category. For example, a one year increase in ouseold ead s scooling leads to an increase of 0.042% in te sare devoted to tis category. Te iger te number of ouseold members wit ig scool completed te iger te sare devoted to non durable goods. It seems tat ouseolds eaded by a male spend less on tis category tan tose eaded by a female, wit a decrease of 0.21%. Houseolds already paying for teir ouse spend considerably more on te patrimony category, wit an increase of 19.7% of te sare devoted to tis category. Houseolds were te ead works in agriculture spend more on te business category tan oterwise, wit a significant increase of 9.3% of te sare. Tese same ouseolds devote 2.5% more to savings. Also, ouseolds wit a iger presence of members over 59 and below 7 years old are associated wit iger levels of savings. Te loan dummy performs surprisingly well, wit significant effects on several categories, especially on tose tat we migt call pysical investment categories. Having a loan increases te sare of durable goods, patrimony and business by 0.40%, 3.5% and 8.6% respectively. It is wort noting tat it also as positive and significant effects on te education and ealt categories, wit increases of 0.37% and 3.4% respectively. Access to capital markets may also capture te financial position of ouseolds and, not surprisingly, ricer ouseolds spend more on education and ealt care. Year and regional dummies were also found significant in several cases indicating te presence of important time and location effects. Moving to our results wit respect to te central issue of, we find tat an increase in te sare of internal by 1% significantly decreases te sare of food (-0.048%) and of savings (-0.07%) and increases tat devoted to ealt (0.02%), education (0.04%), durable goods (0.01), non durable goods (0.01%), patrimony (0.05%) and outtransfers (0.02%). An increase in te sare of external by 1% significantly decreases te sare devoted to food (-0.11%) and oter (-0.01%) and increases te sare expended on education (0.01%), ealt (0.01%), durable goods (0.03%), nondurable goods (0.02%), patrimony (0.097%), savings (0.07%) and outtransfers (0.06%). Tis result sows tat, weter internal or external, ave significant effects on te ways ouseolds decide ow to allocate teir resources. Bot types of income seem to reduce te sare devoted to food expenditures and allocate more resources to oter categories, suc as uman development investments (ealt and education) and capital investment (durable goods and patrimony) categories. Tis results are consistent wit Borraz (2005), and Hanson and Woodrouf (2003) for education. Hildebrandt and Mckenzie (2005) obtain similar results concerning ealt. Tere is also evidence tat tey affect positively te current consumption of ouseolds partially captured in te non durable goods category in agreement wit Canales and Montiel (2004). We attribute te lack of significant effects of on te business category to te weakness of tis concept to effectively capture te type of businesses in wic external recipients are more likely to invest as documented by Escobar and Martinez (1990), Durand and Massey (1992), Massey et al. (1987) and Cornelius (1990). Te specific question attaced to tis information asks about te purcase of macinery, equipment or animals to be used in te production process. In tis way, te information required is not 14

15 properly captured. Concerning te results on savings, ave significant but opposite effects. Wile external seem to encourage ouseold savings, internal seem to decrease savings by te same proportion, indicating a reallocation of ouseold income. Results for te implementation of equation (8) wit te first two-step estimator are sown in Table 6. Some minor canges emerge concerning ouseold caracteristics. For te particular results of almost all effects are in general te same as above in direction, magnitude as well as significance. Te only surprising cange is te appearance of a very small negative but significant effect of external on te business category, i.e., a 1% increase in external remittance leads to a surprising decrease in te sare of total expenditure devoted to business of 0.002%. Te second two-step estimation of equation (9), sown in Table 7 also originated some canges concerning remittance effects. In particular, external no longer ave a positive and significant effect on te education category. Te number of ouseold members wit ig scool education as now a positive and significant effect on te sare devoted to education. Te positive effect of internal on te patrimony category found in te two previous specifications disappears, leaving only external wit a significant role. Te negative effect of external on te business category remains wile tis time, tere is a positive and significant effect of internal witin te same category. As noted, te more interesting results pertain to te business category. It is also possible tat te number of observations may be too small to obtain reliable estimates. Out of possible observations, only 471 recorded a positive amount of expenditure on business. Results of internal and external for te food, ealt, durable goods, non durable goods and savings categories seem robust to different specifications meaning te te significance levels as well as te sing of te coefficients are te same across different specifications. Regarding results on education, internal seem to ave a robust effect across specifications wile external did not pass te tird specification test. Tus we may conclude tat external ave a positive but weak effect on education. Te same applies for internal in te case of te patrimony category. In tis same category, external ave a consistent positive effect. Also, it is important to remark tat te effect of external on te durable goods and patrimony categories is iger tan tat of internal. Te opposite appens wit ealt and education were internal ave a iger impact. Tese findings are robust and consistent across specifications and suggest a differentiated impact of on uman and pysical capital categories. Internal are mainly devoted to uman development wile external are mainly devoted to pysical capital investments. A possible explanation for tis result relies on te nature of bot types of income. Human capital investments tend to be constant over time, especially tose of education. In contrast, pysical capital investments are not as frequent and most of te times contingent upon te availability of monetary resources, especially for poorer ouseolds. If internal present a lower variability, ten ouseolds may decide to use tem in an expenditure category tat requires a constant injection of resources. On te oter side, if 15

16 external are igly volatile, ouseolds may decide to use tem in eventual investments suc as furnising or renovating te ouse as well as purcasing lands or new ouses. A surprising finding is te significant effect of te loan dummy across expenditure categories and specifications. Access to capital markets seems to sape consumption beavior of rural ouseolds consistent wit standard consumer teory. Tus, it is wort asking ourselves if te remittance-receiving condition as some effect on te probability of a ouseold getting a loan. Out of ouseolds, 3674 got a loan, or around 9.8% of te sample. Table 8 sows te marginal effects of a probit estimation for te probability of getting a loan as a function of ouseold caracteristics and teir remittance-receiving condition. In particular, a ouseold receiving internal increases te probability of getting a loan by Te same appens for ouseolds receiving bot kinds of wit a 0.04 increase. However, surprisingly tere is no evidence tat ouseolds receiving external improve teir credit access. A more accurate measure of te frequency and quality (formal vs informal) of credit as well as borrowing constraints may allow us to develop a more complete and adequate model to test if te remittance-receiving condition improves te financial access of ouseolds (maybe troug a iger collateral represented by a permanent flow of ); a model suc as tat developed in Quisumbing and McNiven (2008). Te ENIGH data set does not allow us making suc distinctions Conclusions Te migration penomenon, in addition to contributing to ouseold income, links village ouseolds to new markets, societies and cultures; it may induce canges in production tecnologies and induce a substitution from ome-produced goods to purcased goods in response to lost labor, tecnology cange and oter effects; it may also alter ouseolds information set, risk profile, and preferences in ways tat affect marginal utilities of consumption and investment. A simple comparison of ouseolds wit and witout reveals tat te former group spends more of teir income on ealt, durable goods, and patrimony. Tere are oter significant differences tat indicate tat consumption preferences may be saped by te remittance-receiving condition. External remittance ouseolds ave iger incomes tan internal or non-remittance ouseolds, on average, and socio-demograpic caracteristics across different ouseold categories differ as well. It is not clear, a priori, weter differences in average expenditures between remittance-receiving and nonremittance ouseolds are due to or to differences in total income or oter variables. In tis work we ave presented different empirical specifications to test for te impact of on expenditure patterns in rural Mexican ouseolds. Te modeling approaces we used are more general tan standard consumer models, remittance-use studies, and recent 16

17 work extending consumer models by including direct remittance effects. We control for censoring in demands wile offering a compreensive test of remittance effects on expenditure patterns. Our findings indicate tat tere is evidence of significant effects of internal and external on ouseold expenditure patterns; tese income sources resape ouseold demands in ways tat are independent of total income. Moreover, effects are different for eac type of remittance income. Internal seem to stimulate more categories related to uman development investments, ealt and education, wile external affect positively pysical capital investments. External ave also a solid positive effect on ealt, and less consistent effect on education. Houseold caracteristics do not affect tese results wic agree to wat as been found in previous literature. External, usually considered as part of transitory income, may indeed represent a constant and permanent flow of monetary resource for tose ouseold tat are eavily dependent on suc income source and can be terefore be viewed as permanent income. In tis case ouseolds can make consumption and investment decisions based on tem. Te same is true for ouseolds for wic internal are a significant part of ouseold income. A separated analysis must be made in order to asses weter internal and external are part of permanent or transitory income. Suarez and Avellaneda s findings (2007) support positive and significant effects of remittance income on consumption decisions wen regarded as permanent. Our findings do not support te view tat ouseolds receiving disproportionately spend teir income on current consumption. It is consistent wit te findings reported by oter researcers (Edwards and Ureta 2003, Adams 2005, Lopez Cordova 2004). Besides evidence of positive effects on current consumption (partially measured by expenditure on non-durable goods), our findings reveal tat do indeed significantly influence expenditure patterns in rural areas of Mexico. In particular, te propensity to spend on some investment categories (education, ealt, durable goods, patrimony and savings) appears to be considerably larger for remittance-receiving ouseolds, internal or external, tan for non-remittance ouseolds. As rural incomes rise, expenditure patterns cange. Tis is true regardless of weter te income gains are from migrant or oter sources. Te key question tat sould be of interest to researcers and policy makers is weter expenditure patterns cange differently for ouseolds tat receive, and if so, wy. Tis study as sown tat tey cange indeed, most of te time in a way improving te future well-being of -receiving ouseolds. 17

18 8. - Literature References Adams, Jr., Ricard H Remittances, Houseold Expenditure and Investment in Guatemala. World Bank Policy Researc Working Paper No ttp://ssrn.com/abstract = (Marc) Remittances, Investment and Rural Asset Accumulation in Pakistan. Economic Development and Cultural Cange 47: Amemiya, T Mutivariate Regression and Simultaneous Equation Models Wen te Dependent Variables are Truncated Normal, Econometrica, 42(6): Alderman, H Saving and Economic Socks in Rural Pakistan. Journal of Development Economics 51: Baram, Bradford and Stepen Boucer Migration, Remittances, and Inequality: Estimating te Net Effects of Migration on Income Distribution. Journal of Development Economics 55(2): Borraz, Fernando Assessing te Impact of Remittances on Scooling: te Mexican Experience. Global Economy Journal, Vol.5, pp Canales, Alejandro and Motiel, Israel Remesas e inversión productiva en comunidades de alta migración a Estados Unidos. El caso de Teocaltice, Jalisco. Migraciones Internacionales, Vol. 2, Núm. 3. Cami, R., C. Fullenkamp and S. Jaja Are Immigrant Remittance Flows a Source of Capital for Development? International Monetary Fund (IMF) Working Paper 03/189. Wasington DC. Cornelius, W Labor Migration to te United States: Development Outcomes and Alternatives in Mexican Sending Communities. Wasington, D.C.: Commission for te Study of International Migration and Cooperative Economic Development. 18

19 Deaton, Angus, and Jon Muellbauer An Almost Ideal Demand System. American Economic Review 70(3), Duflo, Ester and Udry. Cristoper, Intraouseold resource allocation in Cote D Ivoire: social norms, separate accounts and consumption coices. National Bureau of Economic Researc. Working Paper Durand, Jorge, and Douglas S. Massey Mexican Migration to te United States: A Critical Review. Latin American Researc Review 27:3-42. Edwards, Alejandra and Ureta, Manuelita International Migration, Remittances and Scooling: Evidence from El Salvador, Journal of Development Economics 72 (2): Escobar, Agustin and Maria de la O. Martinez Small-scale Industry and International Migration in Guadalajara, Mexico. Working Paper No. 53. Wasington, DC: Commission for te Study of International Migration and Cooperative Economic Development. Funkouser, E (1995). Remittances from International Migration: A Comparison of El Salvador and Nicaragua. Te Review of Economics and Statistics, Vol. 77, No. 1. (Feb., 1995), pp García, J. and Labeaga, J. M. (1996). Alternative Approaces to Modelling Zero Expenditure: An Application to Spanis Demand for Tobacco. Oxford Bulletin of Economics and Statistics, 58, Goldring, Luin Development and Migration: A Comparative Analysis of Two Mexican Migrant Circuits. Wasington, D.C.: Commission for te Study of International Migration and Cooperative Economic Development. Hanson, Gordon H. and C. Woodruff Emigration and Educational Attainment in Mexico, Universidad de California, San Diego, (mimeo). Heien, D. and Wessells, C. R Demand systems estimation wit microdata: a censored regression approac. Journal of Bussiness and Economic Statistics, 8, Hildebrandt, Nicole and D. McKenzie Te Effects of Migration on Cild Healt in Mexico, World Bank Policy Researc, WP, num Jabarin, Amer S Estimation of meat demand system in Jordan: an almost ideal demand system. International Journal of Consumer Studies, 29, Lazaridis,P Houseold Meat Demand in Greece: A Demand Systems Approac Using Microdata. Agribusiness 19(1),