Policy Research Working Paper 9014 Impact of International Migration on Labor Supply in Nepal Lokendra Phadera Poverty and Equity Global Practice September 2019 Policy Research Working Paper 9014 Abstract This paper analyzes the differential impact of migration women realign their priorities and reallocate their time from on labor supply of the left-behind household members market employment to self-employment and home produc- in Nepal, where international migration for employment, tion, possibly filling in the roles vacated by the migrants. In predominantly a male phenomenon, increased substan- contrast, the income effect dominates the impact of migra- tially between 2001 and 2011. Using the Nepal Living tion on the left-behind men; that is, men value their leisure Standard Survey data, the paper extends the analysis by more because of the remittances from abroad and decrease incorporating the impacts on the extensive and intensive their overall supply of labor. Additionally, the research finds margins. The study also answer the question: if they are significant heterogeneity in the supply of labor by age, skill, not wage-employed, in what activities are the remaining and household head status among the left-behind women, household members engaging instead? The paper finds pointing toward intrahousehold bargaining. that, in response to out-migration of some family members, This paper is a product of the Poverty and Equity Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The author may be contacted at lphadera@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team ∗ Impact of International Migration on Labor Supply in Nepal † Lokendra Phadera Poverty and Equity Group, World Bank September 2019 Keywords: Migration for employment, Labor supply, Well-being of left-behind JEL Classication: F220, O150, J220 ∗ Acknowledgments: Earlier draft of the paper was circulated under the title International Migration and its Eect on Labor Supply of the Left-Behind Household Members: Evidence from Nepal. I gratefully acknowledge the nancial support provided by the ACES Oce of International Programs, Arlys Conrad Fund and the ACES Education Abroad Oce. I would like to thank participants from the 2015 Advance Research and Scholarly Communication (ACE 561) and International Policy and Development (IPAD) seminars at the University of Illinois Urbana-Champaign. In particular, I would like to thank Mary Paula Arends-Kuenning, Alex E. Winter-Nelson, Benjamin Crost, Rebecca Thornton, Hope Michelson, Kathy Baylis, Craig Gundersen, Charles H. Nelson, Philip Garcia, Mark Borgschulte, and the two anonymous proposal referees for their comments and advice. Any mistakes are my own. † E-mail: lphadera@worldbank.org 1 Introduction One-fth of the 30% poverty reduction in Nepal occurring between 1995 and 2004 is attributed to work-related international migration and remittances sent home (Lokshin et al., 2010). More than two million prime-age (mostly male) Nepalese are working outside the country and the inow of remittances accounts for 30% of the country's GDP (Ministry of Finance, 2014). Decrease in labor stock and substantial income from abroad are likely to have profound eects on the labor market and, yet, the impacts of the migration on the non-income dimensions in Nepal remain relatively unexplored. The paper addresses this issue by documenting the dierential impact of international migration on labor supply of the left-behind family members. Traditionally, the literature on the household-level impacts of migration has focused on income and consumption of the left-behind families. There is a general consensus among this literature that temporary out-migration for employment helps increase income and reduce poverty (Adams and Page, 2005, summarize the results of micro-level analysis in several countries). 1 However, a relatively new strand of the literature focusing on the non-consumption dimension of left-behind family members' well-being provides more mixed evidence. These new studies suggest that male migration decreases non-migrating women's labor market participation and increases their labor supply in farming and unpaid family work (Lokshin and Glinskaya, 2009, in Nepal; Binzel and Assaad, 2011, in the Arab Republic of Egypt; Mu and van de Walle, 2011, in China; Mendola and Carletto, 2012, in Albania). Similarly, migration has a negative eect on the left-behind elderly parents' health (Antman, 2010) and on children's educational attainment (Antman, 2011; McKen- zie and Rapoport, 2011). So, how benecial is international migration for the left-behind members? Murard (2016) provides a more complete theoretical framework incorporating both household con- sumption and labor supply. Using data from Mexico, the paper nds that temporary migration for employment leads to both increase in consumption expenditure and farm labor of household members staying behind. The consumption gains, however, are too large to be purely explained by the endogenous increase in non-migrants' labor supply. 1 Remittances help smooth consumption (Yang and Choi, 2007), provide mutual insurance (Stark and Lucas, 1988), relax credit constraints (Yang, 2008) and alleviate liquidity constraints (Taylor et al., 2003) allowing non- migrating members to engage in higher remunerative activities. At the same time, other community members may benet from positive spillovers: Amuedo-Dorantes et al. (2010) nd that children in communities with more migrants have greater school attendance and McKenzie and Rapoport (2007) nd that migration reduces inequality in sending communities in the long run. 1 This paper contributes to the growing literature of migration's impacts of non-income based outcomes and, in particular, this is among the rst papers to investigate the impact of international migration on time allocation and leisure consumption by gender in Nepal. To my knowledge, Lokshin and Glinskaya (2009) is the only other study that explores this topic in the Nepalese setting. However, they limit their analysis to women's labor supply for wage-employment and exclusively on the extensive margin. Using the 2003/04 round of Nepal Living Standard Survey (NLSS II) data, Lokshin and Glinskaya (2009) nd that men's migration from Nepal has discouraging eect on women's labor force participation. In contrast, this paper uses the 2010/11 round of NLSS, when work-related international migration was at record high, and incorporates both extensive and intensive margins by gender. This is especially important given the traditional roles of rearing children and household chores women assume within the household and society, the left-behind women are expected to be aected dierently than men. One of the most salient features of the employment related emigration in Nepal is that it is predominantly a male phenomenon. With males migrating abroad, the authority over household decision making may shift to women. However, the shift may be accompanied by extra responsibilities requiring extra hours of work or it might compel women to give up their jobs to assume the new roles. An additional important distinction between the papers is the timing of the surveys: while the NLSS II was implemented during a critical phase of the Maoist insurgency in the country, the NLSS III data were collected well after the end of the civil conict. The primary identication strategy of the paper relies on an instrumental variable (IV) approach using a popular instrument in migration literature, that is, historical migration networks. Specif- ically, using the 2001 Nepal Census, I compute the percentage of international migrants from a village as an instrument for migration in 2010/11. A decade lagged migration shares are unlikely to aect local economic conditions and hence, labor supply decisions. Furthermore, I use GDP growth between 2001 and 2011 of the most popular destination countries interacted with the local level decade lagged migration shares as an additional instrument and the results are robust. The analysis has four major ndings. First, solely on the extensive margin, having international migrants in the family discourages the left-behind members from participating in wage-employment. This is true for both male and female members. However, female members signicantly increase participation in self-employment, almost entirely through subsistence farming. Whereas, having 2 migrants in the household does not aect male members' decisions to participate in self-employment. Second, both self-employed and wage-employed adults decrease their weekly hours of labor supply. Third, while women staying behind signicantly increase labor time in household activities, I observe no such impact among men. Fourth, when analyzing the aggregate time allocation (wage work + self-employment + household activities), I nd 0 eect of migration on women's overall labor supply and signicantly negative eect on men's. Therefore, women staying behind realign their priorities and reallocate their time from wage-employment to farm and household activities, while men value their leisure more because of the remittances from abroad and decrease their overall supply of labor. These are reasonable ndings in a country with the traditional household norms and social culture that is likely to see women as subordinate to men. In order to understand the intra- household bargaining channel for labor allocation decisions further, I limit the sample to staying behind women members only and observe that unlike other women in the family, household heads are less likely to participate in wage-employment and do not increase participation in self-employment either. On top of already having greater say in family decision making, household heads are the likeliest recipients of transfers from abroad, which in turn, may further increase bargaining power. This paper contributes to the existing literature in the following ways. First, it complements Lokshin and Glinskaya (2009) by extending the analysis of women's wage-employment on both ex- tensive and intensive margins. Second, the paper also includes males' labor supply to investigate the presence of dierential impact between men and women in Nepal. The most important contribution of the paper is that it analyzes the time allocation of left-behind members beyond their time in self-employment and market work. That is, it answers the question of if they are not employed, what do members of migrant-sending households do instead? This is an important question that the literature, including the studies in other country settings, has mostly ignored. Furthermore, by adding the total time spent in the market, self-employment and household activities, the paper investigates whether migration and remittances increase the consumption of leisure as the microe- conomic theory predicts. The answer may have important implications towards the eect on an individual's welfare gains, which have been rarely explored in previous studies. By dissecting the analysis in these multiple ways, this study provides one of the most complete pictures of migration's impact on labor supply of the remaining family members. The results presented in the paper have important policy implications. They highlight the need 3 for tailored policy initiatives targeting specic subpopulations. Male-dominated migration pushes females to give up wage-employment and increase labor supply in their own farms. Labor markets in rural Nepal tend to be incomplete and not fully integrated due to information asymmetry, lack of mobility, and lack of strong institutional implementations. Policy initiatives should be focused on these aspects to improve the rural wage labor markets thus allowing households to hire workers to replace those who migrate. The rest of the paper is organized as follows. Section 2 provides background of out-bound migration from Nepal, a brief scenario of the current labor market in Nepal, and the motivation for the paper. Section 3 describes the data set used for the analysis, while the empirical strategy and identication are discussed in section 4. The ndings of the analysis are presented in section 5 and section 6 concludes the paper. 2 Background and Motivation 2.1 Emigration from and Remittances to Nepal With 25.2% of its population earning less than US$ 1.25 per day (World Bank, 2014b), Nepal is one of the least-economically-developed nations in the world. However, with recent international labor treaties, Nepal has been experiencing large outows of migrants and hence, remittance inows from abroad. Figure (1) presents the historical international migration trend from Nepal. Close to 2 million Nepalese, 7.3% of the population, were living abroad during the census in 2011. This is a substantial increase compared to earlier decades. Only 3.2% (0.76 million), and 3.4% (0.66 million) of the population was living abroad in 2001 and 1991 respectively. The rise in numbers of Nepalese living abroad in the last decade is mainly due to low skilled employment related migration. The Foreign Employment Act of 2007 - which was designed to provide security, protect the welfare of migrants, provide migrants with education and training before leaving the country, and monitor the businesses that facilitate migration processes - along with the bilateral labor treaties that Nepal signed 2 have facilitated the migration process. The end of the Maoist insurgency, during which mobility within the country was severely restrained and government oces were destroyed, making 2 Nepal signed major international labor treaties in last decade, with Qatar (2005), the United Arab Emirates and the Republic of Korea (2007), Bahrain (2008), Japan (2009), and Malaysia (started process in 2007), with the aim to protect migrants and facilitate the migration process. 4 it dicult to obtain travel documents, also helped improve conditions to migrate internationally. Figure (A.1) shows migration trends to the top ve destination countries for labor employment. 3 Malaysia, Qatar, and the United Arab Emirates , countries Nepal signed treaties with, are among the most favored destinations for work. India, not shown in the gure, is the largest recipient country of Nepalese workers. Due to the open border the two countries share, it is dicult to track migrants and most workers migrating to India do not report to the Department of Foreign Employment, which keeps the records. Additionally, many workers migrating to countries other than India travel through India, so labor-related migration might be signicantly higher than is ocially reported. With these outows of workers, it is not surprising that remittances have become major nancial ows to Nepal. Remittance income has become a major factor in the economic development of Nepal. According to the Ministry of Finance (2014), Nepalese households received 430 billion and 560 billion rupees, accounting for 25.7% and 29.1% of the GDP in scal years 2012-13 and 2013-14, respectively. About 32% of Nepalese households received remittances in 2004 (World Bank, 2006). At the country level, remittances have helped sustain the balance of payments, covering 169.5% of imports and are equiv- alent to 82.9% of the foreign reserves in 2013 (World Bank, 2014a). This trend is likely to continue in the near future, as the growth in outows of migrant workers is on the rise. Migrant outows grew by 16% between 2012 and 2013 (World Bank, 2014a). Remittances sent through unocial channels could be as large. Thus, resource inows from abroad are becoming an increasingly larger share of household budgets to a growing number of families in Nepal. 2.2 Employment International migration and paid-employment are male-dominated phenomena in Nepal. The labor migration trend by gender is presented in Figure (A.2). 3 Among labor-related migrants, only 6.0% (about 23,000), 6.2% (about 28,000) and 5.6% (about 30,000) of migrants were female in 2011, 2012 and 2013, respectively. In my sample, 15.6% of migrants are female. 3 Similarly, 3 Workers going abroad are required to obtain labor-permits from the Department of Foreign Employment. Migrants can apply on their own or can apply through foreign-employment recruiting agencies. The numbers reported in Figure (A.1) are those who opted to apply through the recruiting agencies and were issued the permit. Many of the females migrating abroad from Nepal tend to migrate along with their male household members and not necessarily for employment purposes. Therefore, NLSS, which asks if the household has members abroad, is likely to have a higher percentage of female migrants in the sample than reported here. 5 there is variation in labor market participation across gender. Among working age (18 - 60 years) males, 49.8% of them reported to be paid employees and 70.9% of them reported participating in self- employment, mostly subsistence farming. Among females, only 23.1% reported to be paid employees while 66.3% of them reported participating in self-employment. This is not surprising given the strong social and traditional family norms in Nepal, which discourage women from participating in paid-work, and where women mostly engage in taking care of children and household chores (CEDAW, 2003). However, with males migrating abroad, the authority over household consumption and investment might shift to female members. Women might become more involved in making decisions on labor market participation as well. Besides continuing to care for children and engaging in household chores, women in Nepal often take up men's roles in family farming and enterprises when male members are abroad (Nandini, 1999). Similarly, women play key roles in deciding the use of remittances and running bazaar economics when husbands are away (Brown and Conneil, 1993). The male-dominated international migration in Nepal may aect non-migrant male and female members in the household dierently. 3 Data and Descriptive Analysis For this study I use the 2010-2011 round of the Nepal Living Standard Survey (NLSS III) as the primary data source. It is a nationally representative survey of households and communities that is conducted by the Central Bureau of Statistics (CBS) Nepal, with assistance from the World Bank. It was administered between February 2010 and February 2011. It has a panel component of 1,128 households. Half of the households were followed from the rst round and the other half from the second round. The cross-sectional sample has 5,988 households, which were selected in three stages. 4 The survey collected detailed information on multiple topics related to household welfare. The survey provides rich information on household consumption, sociodemographic com- position of households, health and education attainment of the members, labor market outcomes of the household members, and the source of a wide range of household incomes. It contains detailed information on time-allocation for wage-employment and household production of every household member. Households were also asked to provide information on remittances received by the house- 4 For detailed description of the sample design and the methodology, see http://cbs.gov.np/ 6 holds in the previous 12 months and identied the age, gender, educational attainment, and the destination country of the remittance sender. This study uses 7,108 NLSS III households from both the cross-section and panel components that have complete information on the variables used in the analysis. Among the households in the sample, 33.1% (2,212) reported having at least a member abroad in 2010-11. Characteristics of migrants are reported in table (A.1). Migrants tend to be young (mean age of 28), dominated by males (84.4%) and mostly composed of the daughters/sons of the household head. They tend to have achieved some grade level of education, 53.7% have completed between grades 1 and 10, while 20.3% have completed grade 10 (School Leaving Certicate) and intermediate level (high school). India and the Middle East seem to be the favored destinations for most of the migrants, 44.4% and 24.7%, respectively. Although remittances coming from abroad tend to be relatively large, there are very little dierence in total and non-wage household incomes (Figure 2); households with a migrant have slightly higher total and non-wage incomes. A total of 16,879 working age adults (ages 18-60) are used for the analysis. For the analysis of time allocation in household activities, 4,985 boys and 5,195 girls (both ages 6-17) are added in the sample. Adults from households with and without migrants dier on demographics, household composition, their labor market outcomes, and communities where they reside. Table 1 reports the descriptive statistics of the 16,879 adults used in the analysis. Adults from migrating families, both male and female, are less likely to be employed in wage-employment but are more likely to be involved in self-employment activities compared to the adults from non-migrating families. Consistent with the theory (for the bottom 70% of the households), adults decrease wage-employment with household non-wage income (Figure 3). Migration seems to have a disincentive eect on wage-employment; adults from migrating families are less likely to have wage-work over all the non-wage household income distribution. 5 Adults from migrating families are older by a year and have achieved a year less education than those in households with no migrants. When comparing only female members, females from families with a migrant are more likely to be married and head the household when compared to females from families with no migrants. This is the opposite when comparing male members. Adults from 5 These results are suggestive, they do not account for the fact that the adult most likely to be the wage earner in the family might be more likely to migrate, leaving adults who are intrinsically less likely to have wage employment behind. 7 families with at least one member abroad have larger household size and are more likely to come from a family with a female head. Most of the agriculture in Nepal is subsistence farming, so households own small amounts of land. Migrating families tend to own more land compared to non-migrating families. At the community level, migrant-sending families live areas with sightly lower unemployment rates and literacy rates and higher poverty rates. 4 Empirical Specication and Identication Strategy The goal of the analysis is to evaluate the impact of migration on the labor market outcomes of the left-behind household members. The simplest strategy is to estimate the following equation. yij = β0 + β1 Mj + Xij β2 + ij (1) where yi,j is an outcome variable of individual i in household j. It is either employment status, or total hours spent in paid work, household production or other household activities. Xij is a vector of controls - individual and household characteristics that inuence individual i's productivity and local labor market conditions. Mj is an endogenous binary variable that takes the value of 1 if household j has at least one migrant - 0 otherwise, and ij is the unobserved error term. 4.1 Identication The decisions to migrate and work are selective processes which depend on observed and unob- served household and individual characteristics such as asset level, taste for work, human capital level, opportunities at home and abroad etc. Cross-sectional analysis of migration's eect on labor supply cannot identify a causal relationship because of the endogeneity of migration and household labor supply. Unobserved household and individual characteristics that inuence labor supply are likely to inuence the decision to migrate as well. Similarly, households may endogenize labor supply decisions and hence, earned income while making migration decisions. Thus, estimated eects of migration on labor supply will be biased using the OLS strategy. To address the potential endogene- ity bias, this paper exploits a very popular instrument in the literature: the local historical migrant network serves as an instrumental variable (IV) for the current migration decision. Specically, 8 using the 2001 Nepal Census I compute the percentage of international migrants from a Village Development Committee (VDC) 6 as an instrument for migration in 2010-2011 as in (3). Following is the equation for migration decisions. Mijc =α0 + Xijc α1 + α2 Zijc + νijc (2) where Mi,j is an identier that individual i lives in household j with or without a migrant, Xijc is a vector of controls as dened in equation (1). Zijc , dened in (3), is an exogenous instrumental variable that must satisfy as good as randomly assigned, the rst stage, and the exclusion criteria conditions. The 2001 migration decisions are likely to be random to the 2011 labor supply and mi- gration decisions as migrants from a community in 2001 would not have anticipated the community's labor market conditions in 2011, which satises as good as randomly assigned condition. However, migration networks provide information about the economic opportunities at the destination, po- tential costs, and might reveal migrations' impact on their family's wellbeings to the community, which might inuence other community members' migration decisions (rst stage). Similarly, as long as a decade lagged unobserved community characteristics do not inuence individual's labor supply in 2011, the 2001 community level migration shares are unlikely to feature in the 2011 labor supply equations. Hence, the instrument is likely to aect the outcome only through the endogenous variable, satisfying the exclusion criteria condition. Following is the instrument. M IG2001,c IV1 = (3) P OP2001,c where M IG2001,c is the total number of people living abroad in 2001 from a VDC c and P OP2001,c is the population of the VDC c in 2001. It is important to note that the IV estimates are likely to be greater than the OLS estimates as the instrumental variable approach identies the causal impact of treatment on outcome only on the compliers; that is, IV can recover only the local treatment eect (LATE) (Angrist, 1991). 7 6 Village Development Committee (VDC) is the lowest level of administrative unit. Similar to a municipality, it is responsible for the proper use and distribution of state funds and local level service delivery. Depending on the size, it may represent a single community or multiple communities. It is divided into 9 subdivisions called wards and currently there are 3,276 VDCs in Nepal. 7 OLS assumes treatment to have homogeneous eect. It estimates an average slope, which, in reality, might not be true as there are likely to be heterogeneous responses to the treatment. Let y be a potential outcome, d be a 9 The historical migrant networks are extensively used in the literature to estimate current levels of migration. Migrant networks, which are ties between migrants, former migrants, and non-migrants at the origin through bonds of kinship, friendship, and shared community origins, might be the most important mechanism for international migration (Massey, 1988). Sociologists and anthro- pologists have been studying the role of networks on migration for a long-time (Tilly and Brown, 1967; Mitchell, 1969; Choldin, 1973; Hugo, 1981) and economists have also found that networks play an important role in migration decisions (Hägerstrand, 1957; Greenwood, 1969; McKenzie and Rapoport, 2007; Woodru and Zenteno, 2007; Foged and Peri, 2013). This is because migrant net- works reduce the potential hazards at both the destination and the origin and decrease the cost of relocation (Massey, 1988; McKenzie and Rapoport, 2007). Historically, international migration networks are region specic in Nepal. For example, most of the people joining the Indian and British armies in the 1930's (and up to the present day) were Gorkhas from the Pokhara region when the recruitment started. Similarly, people from the southern plains migrated after the beginning of the extensive cultivation of tea in northern India (Seddon et al., 2001). People from Far-Western villages in Nepal tend to migrate to the same destination in India as their co-villagers (Thieme, 2006). Figure (A.3) shows the top 10 origin districts for labor- employment migration at present. These top districts, which are located mostly in South-Eastern Terai, account for 36.5% of the total labor-related migration from Nepal between 2008 and 2014. In order to investigate the role of past community level migration on current levels of migration in the community, I calculate the shares of migrants in 2001 and 2011 to a particular destination country from a VDC using the 2001 and 2011 Nepal Censuses. Then, for each destination country, I regress the share of migrants to the destination from a VDC in 2011 on the share of migrants from potential treatment and z be an instrument, then under the assumptions of independence, exclusion and monotonicity IV is a LATE. Independence assumption states that dierence in outcome (y) and dierence in treatment (d) between zi = 0 and zi = 1 should capture the causal eect of the instrument on outcome and treatment. This is satised when as good as randomly assigned condition in IV is fullled. The exclusion assumption, y(di , zi = 0) = y(di , zi = 1) for di = 0, 1, is the same as the exclusion criteria condition in IV. Whereas, the monotonicity assumption states that the instrument, if it has any eect, should aect everyone in the same direction i.e. d1i ≥ d0i or d1i ≤ d0i . Given these assumptions, IV = LAT E = E E (yi |zi =1)−E (yi |zi =0) (di |zi =1)−E (di |zi =0) = E (y1i − y0i | d1i > d0i ). Notice that the denominator is just the shares of compliers i.e. the percentage of the sample that participate in the treatment only because of the instrument. So, when there is treatment eect heterogeneity, IV estimates the causal eect of treatment on outcome only among compliers. In our case, compliers are those who decide to migrate due to the higher share of migrants in the community in the past. Since all three assumptions are likely to be satised, the estimated IVs in the paper are LATEs. Contrarily, under the homogeneity assumption or perfect compliance, the denominator would be 1 and hence, LATE = IV = ITT (intention to treatment eect), which is what OLS assumes. Since the share of compliance is always less than or equal to 1, the LATE is always greater than the ITT. 10 the same VDC to the same destination in 2001. 8 Estimated correlations for the top four destination countries are presented in Figure (4). There is a strong correlation between historical and current level migration shares. Coecients are either closer to 1 or greater than 1 and are highly signicant. 9 Additionally, I regress the number of migrants from a VDC to a particular destination country in 2011 on shares of migrants from the VDC that went to the same destination in 2001. Estimations are presented in Table (2). Again, the 2011 migration levels are highly correlated with the 2001 destination specic propensity scores to migrate. A potential complication with the instrument is that, although lagged by a decade, lagged un- observed VDC characteristics can inuence labor supply decisions. Historically, migrant sending communities might be systematically dierent to those less migrant sending communities on eco- nomic and labor market characteristics. High historical migration might be linked to bad economic conditions at the origin or remittances might have improved the local economic conditions over time. Depending on these conditions, IV estimates might be biased downward or upward. To ad- dress this problem, I control for a host of community level economic characteristics such as poverty rate, illiteracy rate, unemployment, and inequality within a VDC. 10 Even so, I cannot claim with certainty that the instrument captures no unobserved VDC characteristics that triggered the past migration and inuenced the present labor supply. However, the identication is at least as valid as those used in previous studies. Furthermore, I create an additional instrument, average GDP growth between 2001 and 2011 of the top 8 destination countries in 2011 and interact it with the share of migrants to those destinations from a VDC in 2001 as in (4). 8 M IG2001,i,c i=1 (4) IV2 = X GDP Growth01−11 P OP2001,c where M IG2001,i,c is international migrants in 2001 from a VDC c to destination i and P OP2001,c is the population of a VDC c in 2001. The top eight destination countries in 2011 are India, Malaysia, Saudi Arabia, Qatar, Kuwait, the United Arab Emirates, the United Kingdom, and the United States. GDP Growth01−11 is the average GDP growth of the top 8 countries between 2001 and 2011. 8 For detailed calculations of shares of migrants and regression equations see the notes of the gures and tables. 9 These results hold for other destination countries as well. 10 It is more appropriate to control the community level characteristics at the baseline, but due to the data limitation I use the 2011 community conditions as the controls. 11 Since the GDP growth of the destination countries is exogenous to the local labor market conditions the interacted term is likely to be exogenous as well. I report results from both IV1 and IV2 in the main specication. Despite the binary endogenous variable, I use the linear 2SLS estimation strategy as suggested by Angrist and Pischke (2009). Because of the binary endogenous variable, the conditional expectation function (CEF) associated with the rst-stage might be nonlinear. One can use a nonlinear rst-stage and use the predicted probabilities as an instrument in a garden- variety 2SLS as suggested by Angrist and Pischke (2009) and Wooldridge (2010) to avoid `'forbidden regression in the second step. However, this requires making distributional assumption of the rst- stage CEF. In contrast, with the linear 2SLS, one need not worry whether the rst-stage is linear (Angrist and Pischke, 2009). 11 5 Results A linear estimation of equation (2) using IV1 (equation 3), and IV2 (equation 4), are presented in Table (A.2) columns (1) and (2) respectively. Coecients on both the instruments are highly correlated with the household's migration decisions. Adults in VDCs with higher proportions of migrants in 2001 were more likely to live in a household with at least one international migrant in 2011. Interestingly, there is a positive relationship between local unemployment rate and illiteracy rate. The negative correlation between the migrant outows and the local market conditions results in downward biased OLS estimates of the impact on the labor market outcomes (Foged and Peri, 2013). The OLS estimates of specication (1) are presented in columns (3) and (4), the 2SLS using IV1 are presented in columns (5) and (6) and IV2 are presented in columns (7) and (8) of table (A.2). The dependent variables are weekly hours supplied for wage-employment and self-employment. I use all the adults, both employed and non-employed in wage and self-employment. Since 0 hours are observed for non-employees, there is no problem of sample selection. The estimated eects are a combination of intensive and extensive margins. Columns (3), (5) and (7) are estimated wage labor supply equations while (4), (6) and (8) are self-employment labor supply equations. Overall, the estimates of β2 s, coecients on X's in equation (1), are comparable across the 11 With a linear estimation, one would be worried about predicted probabilities from the rst-stage not being within 0 and 1. Assuredly, all the predicted values in the estimated models are within the range. 12 estimation strategies. ˆ2 's β direction corresponds well with the economic intuitions. Individual characteristics, age, and household head status strongly determine the level of labor supply for both the wage and self-employment. It is not surprising that women, in Nepal, are less likely to work outside their homes, and the supply of wage-hours increases with years of schooling. Similarly, household characteristics, and land ownership decrease wage-hours and increases time spent in self- employment. Owning a home and being from a higher social caste, Brahman/Chhetri, discourage adults from working. The coecient of interest, ˆ1 , β which is the coecient on migration decision in equation (1) is statistically signicant across all the econometric techniques used for time spent in wage employ- ment. While comparable with each other, IV estimates are signicantly greater in magnitude than the OLS estimate. As discussed earlier, OLS suers from selection bias and the negative correlation between the local market conditions and the migration ows is likely to bias the eects downward. Furthermore, the 2SLS can recover the impact only on the compliers i.e. local average treatment eect (LATE), which is always less than the intention to treat eect (ITT). 7 Both the instruments are reasonably strong, with a high correlation between the endogenous variable and the instruments. F -statistics of the rst stage are always above 60, which are greater than the threshold value of 10 researchers usually consider below which one might run into the problem of weak instrument (Stock and Yogo, 2005). Results from the IV regressions suggest that having a migrant in the household discourages the left-behind members from working in wage-employment. Adults from migrant-families decrease their weekly hours of labor supply for wage-employment by about 8 hours when compared to the adults from the non-migrant households. This is a decrease of almost one ocial work day. The direction of the eect is consistent with the prediction of the theoretical model. The income trans- fers through remittances increase the reservation wages and since most of the home production is subsistence farming, members' time-inputs are likely to be substitutes, both of which reduce non- migrating household members' hours of work in market employment. Contrary to the prediction of the theoretical model, migration has a negative eect on self-employment. However, both the IV results are small in magnitude and statistically not dierent from 0. As discussed earlier, these IV results are combinations of intensive and extensive margins. Separating these eects provides better insight into the economic behavior of the left-behind members and may assist better in policy 13 design. I apply the following strategy to separate out these eects. I use a binary employment status, 1 for employed - 0 otherwise, as the dependent variable instead of hours supplied to estimate the impact on extensive margins. To examine the impact on intensive margins, I analyze the hours supplied for a particular employment by limiting the sample to those who are engaged in that employment. That is, depending on the left-hand side outcome variable, the analyzed sample is conditioned on being employed in that particular sector. Tables (3) and (4) reports 2SLS estimates of the impact of migration on wage employment and self-employment outcomes, respectively, by sex using IV1 as the instrument. Panel A of the tables shows the impacts on extensive margins while panel B shows the eects on intensive margins. Again, the instrument is reasonably strong with all F -statistics from the rst stage above the threshold value of 10 except for the hours supplied for non-agricultural wage employment, which is at 7.1. Both male and female adults from the migrant sending families have smaller involvement in wage-employment, both on extensive and intensive margins when compared to adults from families without migrants. Meanwhile, migration has greater negative eect on left-behind men's propensity to participate in market employment than women's, 31.1% vs 20.5%, (Table 3, Panel A). However, among working adults, left-behind women supply fewer hours in wage-employment than the left-behind men (Table 3, Panel B), which could be a result of Nepali women, if employed outside their homes, having mostly part-time jobs. At the same time, having migrants in the family aects left-behind women's self- employment decisions on both extensive and intensive margins, but not men's. Compared to women from households without migrants, women in households with migrants are 28.0% more likely to be engaged in self-employment, mostly working in their own farms. However, among those employed, women decrease weekly hours worked by 13 hours (Table 4). Overall, migration has negative eect on all the left-behind household adults' involvement in wage-employment but it is only women who increase their involvement in self-employment. In a country with traditional household norms and social culture, where women tend to be subordinate to men, it is not surprising that only women redistribute their time allocation in response to sending some family members abroad. Table (5) presents the 2SLS estimates of the impacts of migration on the time spent in household production and other activities by the left-behind adults and children using IV1 as the instrument. 12 12 Fetching water, collecting rewood and dung, taking care of animals, making mats, knitting, weaving, tailoring 14 Sending some members abroad has 0 eect on time allocation in household activities of left-behind male adults and boys. Women and girls, on the contrary, signicantly decreased their weekly hours, 7.8% and 4.1% respectively, in less productive activities and increase their weekly hours in more productive activities more than proportionally, 8.2% and 5.5% respectively. When a household sends some of its members abroad, there are several pathways through which non-migrating member's labor supply is aected. First, the income transfers through remittances increase the valuation of leisure of left-behind members making leisure more appealing. Second, because of the decrease in the household's stock of labor hours, left-behind members might over- burden themselves by adding the role vacated by the migrants onto their own workload. Third, households might realign their priorities and redistribute the remaining stock of labor time once they send members abroad. In order to distinguish the latter two pathways, I add weekly hours supplied by the left-behind household members across all activities. The results are presented in table (A.3). The second scenario is true neither for the left-behind women nor men. As a matter of fact, non-migrating male members increase their weekly leisure consumption by 18 hours by cutting their involvement in all forms of employment (wage + self-employment). This ts well with the rst scenario rather than the latter two. To isolate the rst path, of impacts through income transfers, one has to model migrants' decision to send back remittances, which requires a new identication strategy and potentially a new instrument. Due to limitations in the data, I believe the analysis is beyond this papers' scope. Meanwhile, the aggregate eect is statistically 0 for the left-behind women members, which corresponds well with the third scenario. In Nepal, while left-behind males increase their leisure consumption in response to sending members abroad, women realign their priorities and assume the roles in home production and self-employment - roles likely to have been vacated by the migrants. The dierential impacts between the left-behind male and female members are most likely to be a result of dierential bargaining power male and female members have within a household. Bar- gaining heterogeneity, however, is not limited to gender dierences. Even within a gender group, members might have dierent levels of bargaining, creating heterogeneous responses to migration and processing preserved food are classied as household production activities. Whereas, minor house repair, cooking food, cleaning, laundry, dishes, babysitting, and taking care of the elderly are categorized as other household activities. They are classied as such because the CBS takes into account only the time spent on the former activities while calculating unemployment. Males and females are ages between 18 and 60 while boys and girls are ages between 6 and 17. 15 within the group. Among female members, older females and household heads are likely to have a higher level of bargaining than other female members in the family. Table (A.4) presents 2SLS results of the impact of migration on total labor supply by women's age (Panel A) and women's house- hold head status (Panel B). Consistent with the hypothesis, older women from migrant-household supply substantially less overall labor hours (16) mostly by reducing their hours in wage and self employment (13). In contrast, migration has no eect on younger women. At the same time, when a household sends its members abroad, it negatively aects overall labor supply of women who are head of the household but has no eect on other female members' labor supply (Table A.4, Panel B). It is plausible that depending on each member's bargaining power within a household, the de- cision to send members abroad has dierential impacts on each left-behind members' labor supply decision. 5.1 Heterogeneous Eects As with the sex of the left-behind members, households' migration decisions are likely to aect dierent groups within the household dierently. I divide the sample by skill level, age, and the household head status to analyze the potential treatment eect heterogeneity. Table (6) presents the 2SLS estimates by skill level. Adults with school leaving certicate (SLC) or more are dened as high skilled. 13 It is only the low skilled left-behind members that are aected by the migration. There is statistically zero eect on labor supply of high skilled adults. While less likely to participate in market jobs, low skilled adults are more likely to be involved in self employment activities. At the same time, low skilled adults, if employed, supply fewer weekly hours for both wage-employment and self-employment. The dierential impact by skill level is reasonable as high skilled workers are likely to be already involved in more formal and permanent jobs that have higher opportunity costs of switching. Tables (7), and (8) present the eect of migration on labor supply from the 2SLS estimation using IV1 as an instrument by age and household head status respectively. Adults with ages between 18 to 40 are dened as young adults. Irrespective of age, migration aects the left-behind members 13 All the students whether in private or public schools that follow the Nepali education system, have to take the school leaving certicate (SLC) exam at the end of tenth grade. All schools in the country follow the system with very few exceptions, which follow the Indian or American system. It is mandatory that students pass the exam to continue their studies further within the country and qualify for most of the government and private jobs. 16 similarly, both younger and older adults decrease their participation in wage employment while increasing participation in self-employment. Likewise, both types of adults have similar responses in their intensive margins - lower weekly labor supply (Table 7). Although outmigration of some family members has similar eect on the intensive margins for wage-work and own work, there is a signicantly dierent eect on the participation rate by household head status. The left-behind household heads are 41.4% less likely to participate in paid work compared to 23.1% of the other household members. Concurrently, outmigration of family members does not aect the household head's decision to participate in self-employment activities but signicantly increases other members' involvement, 29.8% (Table 8). This corresponds well with the intuition that the household heads are likely to have higher bargaining power within the household and are likely to receive remittances sent by the migrants, which in turn, may further increase their bargaining power. 6 Discussion and Conclusions In this paper I use a unique source of nationally representative data during the period that Nepal experienced a boom in outmigration. The data set contains detailed information on time allocation of every individual in the household, which allows me to extend the analysis further and answer the question that previous studies on the topic could not answer in Nepal. Using the NLSS III data set, this paper explores the impact of migration on labor supply of the left-behind household members both on extensive and intensive margins for wage-employment, self-employment, and household activities. The paper also provides an answer to the question, what are the remaining members in the households engaged in instead of work employment? I nd that having international migrants in the family discourage members staying behind from participation in wage-employment. This is true for both male and female members. However, female members increase participation in self-employment, almost entirely through subsistence farming. The paper also nds that both self-employed and wage-employed adults decrease weekly hours of labor supply and only women in the migrant-sending household increase time in household activities. Findings presented in the paper suggest that male-dominated migration forces women to realign priorities and reallocate time from market-work to farming and household activities. In contrast, 17 because of the income transfers, men now value their leisure more and decrease their overall labor supply. These are reasonable ndings in a country where the traditional household norms and social culture see women as subordinate to men. The question of the impact of outmigration on the well-being of the left-behind members is of importance for Nepal, which already has high levels of outmigration and the trend for outmigration is on the rise. The neoclassical micro theory identies wage and employment opportunity dier- entials between the place of origin and the place of destination as the main cause of migration. Therefore, women switching from the formal labor market to self-employment should speed-up the process of equalization of wages and opportunities between the two places (Lokshin and Glinskaya, 2009). However, with the opening of new destinations, economic incentives abroad, and increasingly simpler migration processes, outmigration from Nepal will not decline anytime soon. International migration for labor-employment from Nepal, however, is risky. 14 Many Nepali emigrant laborers nd themselves working in hazardous conditions, work long hours, face delays in getting paid, and some even lose their lives (Kaphle, 2014). It is imperative that Nepal put in place a broad set of policies that protect the welfare of migrants, many of which are breadwinners in their families, and safeguard the wellbeing of the left-behind members. 15 The results presented in the paper may play an important role in designing some of these policies, especially in protecting the wellbeing of the remaining household members. They highlight the need for tailored policy initiatives that target specic subpopulations. Male-dominated outmigration pushes left-behind women to withdraw from wage-employment and increase the labor supply in their own farms. Policy initiatives should be focused on improving the rural wage labor markets, which would allow households to hire workers to replace those who migrate. Additionally, these policies may help in insuring households against the negative migrant-related shocks if they can encourage the left-behind members to remain in formal wage-employment, which may have higher returns than self-employment. 14 International migration from Nepal, especially for labor, is costly and requires extensive planning ahead. It requires obtaining passport and visa, purchasing ticket and saving up or borrowing for the associated costs. Similarly, most of the international work-related migration involves migration brokers who charge high fees for their services and there are contractual agreements in place between migrants and the hiring agency ahead of migration and reversing the decision once made can be very costly (Bhattarai, 2005). 15 See McKenzie and Yang (2015) for reviews of dierent policies about migration. 18 References Adams, R. H. and Page, J. (2005). Do international migration and remittances reduce poverty in developing countries? World development, 33(10):16451669. Amuedo-Dorantes, C., Georges, A., and Pozo, S. (2010). Migration, remittances, and children's schooling in haiti. The Annals of the American Academy of Political and Social Science, 630(1):224244. Angrist, J. D. (1991). Instrumental variables estimation of average treatment eects in econometrics and epidemiology. Working Paper 115, National Bureau of Economic Research. Angrist, J. D. and Pischke, J.-S. (2009). Mostly harmless econometrics: An empiricist's companion. Princeton university press. Antman, F. M. (2010). Adult child migration and the health of elderly parents left behind in mexico. American Economic Review, 100(2):20508. Antman, F. M. (2011). The intergenerational eects of paternal migration on schooling and work: What can we learn from children's time allocations? Journal of Development Economics, 96(2):200208. Bhattarai, P. (2005). Migration of Nepalese Youth for Foreign Employment: Problems and Prospects : (a Review of Existing Government Policies and Programmes). YOAC. Binzel, C. and Assaad, R. (2011). Egyptian men working abroad: Labour supply responses by the women left behind. Labour Economics, 18:S98S114. Brown, R. P. and Conneil, J. (1993). The global ea market: migration, remittances and the informal economy in tonga. Development and Change, 24(4):611647. CEDAW (2003). Consideration of reports submitted by states parties under article 18 of the con- vention on the elimination of all forms of discrimination against women. Technical report, United Nations. Choldin, H. M. (1973). Kinship networks in the migration process. International Migration Review, 7(2):163175. 19 Department of Foreign Employment (2014). Labour migration for employment a status report nepal: 2013/2014. Technical report, Government of Nepal, Ministry of Labour and Employment, Department of Foreign Employment. Foged, M. and Peri, G. (2013). Immigrants' and native workers: New analysis on longitudinal data. Working Paper 19315, National Bureau of Economic Research. Greenwood, M. J. (1969). An analysis of the determinants of geographic labor mobility in the united states. The review of Economics and Statistics, 51(2):189194. Hägerstrand, T. (1957). Migration and area. In Hannerberg, D., Hägerstrand, T., and Odeving, B., editors, Migration in Sweden: A Symposium, pages 27158. Gleerup: Lund Studies in Geography. Hugo, G. J. (1981). Village-community ties village norms and ethnic and social networks: a review of evidence from the third world. In De Jong, G. F. and W, G. R., editors, Migration Deci- sion Making, Multidisciplinary Approaches to Microlevel Studies in Developed and Developing Countires, pages 186224. New York NY Pergamon Press 1981. Kaphle, A. (March 24, 2014). Nepal, once known for farming, now exports people; migrants earn big but face risks. The Washington Post. Lokshin, M., Bontch-Osmolovski, M., and Glinskaya, E. (2010). Work-related migration and poverty reduction in nepal. Review of Development Economics, 14(2):323332. Lokshin, M. and Glinskaya, E. (2009). The eect of male migration on employment patterns of women in nepal. The World Bank Economic Review, 23(3):481507. Massey, D. S. (1988). Economic development and international migration in comparative perspec- tive. The Population and Development Review, 14(3):383413. McKenzie, D. and Rapoport, H. (2007). Network eects and the dynamics of migration and in- equality: theory and evidence from mexico. Journal of development Economics, 84(1):124. McKenzie, D. and Rapoport, H. (2011). Can migration reduce educational attainment? evidence from mexico. Journal of Population Economics, 24(4):13311358. 20 McKenzie, D. and Yang, D. (2015). Evidence on policies to increase the development impacts of international migration. The World Bank Research Observer, pages 155192. Mendola, M. and Carletto, C. (2012). Migration and gender dierences in the home labour market: Evidence from Albania. Labour Economics, 19(6):870880. Ministry of Finance (2014). Economic survey: Fiscal year 2013/14. Technical report, Government of Nepal Ministry of Finance. Mitchell, J. C. (1969). The concept and use of social networks. Bobbs-Merrill. Mu, R. and van de Walle, D. (2011). Left behind to farm? women's labor re-allocation in rural china. Labour Economics, 18:S83S97. Murard, E. (2016). Consumption and leisure: The welfare impact of migration on family left behind. Nandini, A. (1999). Engendered mobilization-the key to livelihood security: Ifad's experience in south asia. Rome: IFAD. Seddon, D., Adhikari, J., and Gurunga, G. (2001). The New Lahures: Foreign Employment and Remittance Economy of Nepal. Nepal Institute of Development Studies. Stark, O. and Lucas, R. E. (1988). Migration, remittances, and the family. Economic development and cultural change, 36(3):465481. Stock, J. H. and Yogo, M. (2005). Testing for weak instruments in linear iv regression. In Andrews, D. and Stock, J. H., editors, Identication and inference for econometric models: Essays in honor of Thomas Rothenberg, pages 80108. New York: Cambridge University Press. Taylor, J. E., Rozelle, S., and De Brauw, A. (2003). Migration and incomes in source communities: A new economics of migration perspective from china. Economic Development and Cultural Change, 52(1):75101. Thieme, S. (2006). Social networks and migration: Far West Nepalese labour migrants in Delhi, volume 7. LIT Verlag Münster. Tilly, C. and Brown, C. H. (1967). On uprooting, kinship, and the auspices of migration. Interna- tional Journal of Comparative Sociology, 8(2):139164. 21 Woodru, C. and Zenteno, R. (2007). Migration networks and microenterprises in mexico. Journal of development economics, 82(2):509528. Wooldridge, J. M. (2010). Econometric analysis of cross section and panel data. MIT press. World Bank (2006). Nepal - resilience amidst conict: an assessment of poverty in nepal, 1995-96 and 2003-04. Technical report, World Bank, Data. World Bank (2014a). Migration and development brief 23. Technical report, World Bank, Migration and Remittances Team, Development Prospects Group. World Bank (2014b). World development indicators. Technical report, World Bank, Data. Yang, D. (2008). International migration, remittances and household investment: Evidence from philippine migrants' exchange rate shocks. The Economic Journal, 118(528):591630. Yang, D. and Choi, H. (2007). Are remittances insurance? evidence from rainfall shocks in the philippines. The World Bank Economic Review, 21(2):219248. 22 7 Tables and Figures Figure 1: International Migration Trend from Nepal 30 .5 personal remittances received (% of GDP) 25 international labor migrants (1,000,000) .4 20 .3 15 .2 10 .1 5 0 0 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 year migrants remittances 23 Figure 2: Household Income- Kernel Density 24 Figure 3: Rate of Labor Market Participation - All Adults Note: Sample is limited to working-age (18 to 60) population. Figures are created using fractional polynomial regression. Non-wage income is monthly per-capita. 25 Figure 4: Correlation Between 2001 and 2011 Proportions of Migrants by Major Destination Coun- tries. Data Source : 2001 and 2011 Nepal Census. Linear t is Yj,d,11 = α0 + α1 Yj,d,01 , where Yj,d,11 , and Yj,d,01 are share of migrants to destination d from VDC j in 2011 and 2001 respectively. Share of migrants is calculated as Mj,d,t Yj,d,t = P OP j,t , where Mj,d,t is number of migrants from VDC j to destination d in year t. P OPj,t is the population of VDC j in year t. 26 Table 1: Descriptive Statistics All Adults Female Male Nonmigrant Migrant-Nonmigrant Nonmigrant Migrant-Nonmigrant Nonmigrant Migrant-Nonmigrant (Mean) (Dierence) (Mean) (Dierence) (Mean) (Dierence) Wage employment 0.370 -0.079 *** 0.239 -0.021 ** 0.517 -0.084 *** Self employment 0.670 0.044 *** 0.644 0.055 *** 0.699 0.044 *** Individual Characteristics Age 35.699 0.915 *** 35.046 0.454 * 36.436 2.337 *** Years of education 5.303 -0.912 *** 4.025 -0.495 *** 6.744 -0.686 *** Married 0.793 0.005 0.798 0.040 *** 0.786 -0.067 *** Household head 0.340 0.010 0.113 0.154 *** 0.596 -0.085 *** Household Characteristics Female household head 0.131 0.163 *** 0.170 0.201 *** 0.087 0.059 *** Household size (AE) 5.217 0.251 *** 5.175 0.139 ** 5.265 0.502 *** Share of children 0-7 0.122 0.020 *** 0.125 0.028 *** 0.118 0.002 27 Share of children 8-15 0.186 -0.005 0.189 0.006 0.184 -0.028 *** Share of female adult 0.310 0.048 *** 0.341 0.038 *** 0.274 0.041 *** Share of male adult 0.286 -0.073 *** 0.243 -0.080 *** 0.334 -0.024 *** Share of elderly 0.055 0.010 *** 0.060 0.008 *** 0.049 0.009 *** Landless household 0.298 -0.092 *** 0.285 -0.083 *** 0.313 -0.099 *** Own less than 1 acres 0.316 0.011 0.319 0.019 * 0.313 -0.008 Own 1-2 acres 0.184 0.024 *** 0.192 0.021 ** 0.174 0.024 ** Own 2-5 acres 0.152 0.057 *** 0.154 0.047 *** 0.150 0.075 *** Own 5 or more acres 0.049 0.000 0.049 -0.004 0.050 0.009 Own a house 0.854 0.072 *** 0.868 0.057 *** 0.839 0.091 *** VDC and Region Characteristics Unemployment rate 0.083 -0.005 *** 0.082 -0.005 *** 0.084 -0.005 ** Illiteracy rate 0.395 0.039 *** 0.400 0.039 *** 0.389 0.036 *** VDC inequality (Gini) 0.469 0.001 0.471 0.002 0.466 -0.002 District poverty rate 0.212 0.011 *** 0.215 0.011 *** 0.209 0.008 ** Rural 0.615 0.105 *** 0.629 0.101 *** 0.599 0.100 *** Observations 16879 9597 7282 Note: *** (**) (*) indicates signicance at the 1% (5%) (10%) level. Sample is working age (18 to 60) adults. Table 2: Migration Network: Dependent Variable - Number of Migrants in 2011 (1,000) All VDCs in Nepal NLSS VDCs (1) (2) (3) (4) Share of migrants 2001 3.758*** 3.779*** 6.606*** 6.714*** (0.101) (0.101) (0.676) (0.678) Constant 0.013*** 0.001 0.037*** -0.005*** (0.001) (0.001) (0.006) (0.002) District Fixed Eect Yes Yes Observations 100880 100880 11648 11648 R2 0.145 0.161 0.060 0.097 Data Source : 2001 and 2011 Nepal Census. Standard errors in parentheses.∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001. Standard errors are clustered at VDC level. Estimated model is Mj,d,11 = β0 + β1 Yj,d,01 , where Mj,d,11 is number of migrants, in 1,000, to destination country d from a VDC j in 2011. Yj,d,01 is share of migrants to destination d from a VDC j in 2001. Share of migrants is calculated as Yi,d,t = P OP , where Mj,d,t Mj,d,t j,t is number of migrants from VDC j to destination d in year t. P OPj,t is the population of VDC j in year t. 28 Table 3: 2SLS Estimation of Labor Supply by Gender- Wage Employment Female Male (1) (2) (3) (4) (5) (6) Any Wage Emp. Agri. Non-Agri. Any Wage Emp. Agri. Non-Agri. Panel A: Labor market participation Household with migrant -0.205* -0.049 -0.221** -0.311* 0.027 -0.371** (0.104) (0.083) (0.088) (0.175) (0.114) (0.185) Observations 9597 9597 9597 7282 7282 7282 Wald χ 2 1236.455 576.071 1744.580 941.219 476.251 858.504 F -test 1stage 70.938 70.938 70.938 41.195 41.195 41.195 Panel B: Hours supplied Household with migrant -15.219** 7.292 -26.255* -13.564* 2.955 -13.505* 29 (6.826) (6.134) (13.613) (7.066) (4.996) (7.730) Observations 2222 1069 1264 3623 773 3189 Wald χ2 2004.359 205.487 920.637 1538.916 314.929 1475.447 F -test 1stage 18.791 16.894 7.140 37.343 20.071 31.814 Controls Individual characteristics Yes Yes Yes Yes Yes Yes Household characteristics Yes Yes Yes Yes Yes Yes VDC characteristics Yes Yes Yes Yes Yes Yes Region Fixed Eect Yes Yes Yes Yes Yes Yes Note: 2SLS estimates are reported in the table. Share of international migrants in a VDC in 2001 is used as an instrument for the estimations. Sample is working age (18 to 60) adults and standard errors are clustered at VDC level. Standard errors in parentheses.∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Individual controls are age age2 , years of education, marital status, and household head identier. Household controls are female HH head, share of male and female adults, share of elderly, share of children, amount of land-owned, house ownership and social caste. Similarly VDC level controls are, unemployment rate, poverty rate, illiteracy rate, inequality (Gini), and urban/rural location. Regions are Kathmandu, other urban areas, Western hills, Eastern hills, Western Terai, and Eastern Terai. Sample in Panel B is conditioned on being employed in that particular sector. Log of hourly wages is added as an extra individual control. Table 4: 2SLS Estimation of Labor Supply by Sex- Self Employment Female Male (1) (2) (3) (4) (5) (6) Any Self Emp. Agri. Non-Agri. Any Self Emp. Agri. Non-Agri. Panel A: Labor market participation Household with migrant 0.280*** 0.308*** -0.098 0.141 0.205 -0.179 (0.089) (0.113) (0.087) (0.120) (0.148) (0.166) Observations 9597 9597 9597 7282 7282 7282 Wald χ 2 8883.913 15715.418 703.258 7294.166 18469.293 841.418 F -test 1stage 70.938 70.938 70.938 41.195 41.195 41.195 Panel B: Hours supplied Household with migrant -13.030*** -6.460* -24.345 -10.252 -5.819 4.337 30 (4.559) (3.787) (16.457) (6.781) (4.098) (15.669) Observations 6358 5550 1557 5160 4006 2109 Wald χ2 525.033 434.614 504.186 2193.401 662.555 863.127 F -test 1stage 61.724 58.076 10.609 42.797 35.274 11.702 Controls Individual characteristics Yes Yes Yes Yes Yes Yes Household characteristics Yes Yes Yes Yes Yes Yes VDC characteristics Yes Yes Yes Yes Yes Yes Region Fixed Eect Yes Yes Yes Yes Yes Yes Note: 2SLS estimates are reported in the table. Share of international migrants in a VDC in 2001 is used as an instrument for the estimations. Sample is working age (18 to 60) adults and standard errors are clustered at VDC level. Standard errors in parentheses.∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Individual controls are age age2 , years of education, marital status, and household head identier. Household controls are female HH head, share of male and female adults, share of elderly, share of children, amount of land-owned, house ownership and social caste. Similarly VDC level controls are, unemployment rate, poverty rate, illiteracy rate, inequality (Gini), and urban/rural location. Regions are Kathmandu, other urban areas, Western hills, Eastern hills, Western Terai, and Eastern Terai. Sample in Panel B is conditioned on being employed in that particular sector. Table 5: 2SLS Estimation of Labor Supply for Household Activities Household production Other household activities (1) (2) (3) (4) (5) (6) (7) (8) Adult-Male Adult-Female Boys Girls Adult-Male Adult-Female Boys Girls Household with migrant 1.034 8.184** 2.006 5.541** -2.409 -7.776** 0.366 -4.107* (4.456) (3.810) (2.326) (2.354) (2.361) (3.218) (0.970) (2.151) Observations 7282 9597 4985 5195 7282 9597 4985 5195 Wald χ 2 1805.517 3126.243 755.167 1678.898 1055.533 1854.033 583.091 1950.024 F -test 1stage 41.195 70.938 55.647 47.598 41.195 70.938 55.647 47.598 Coecient 1stage 1.611*** 1.913*** 1.976*** 1.815*** 1.611*** 1.913*** 1.976*** 1.815*** 31 Controls Individual characteristics Yes Yes Yes Yes Yes Yes Yes Yes Household characteristics Yes Yes Yes Yes Yes Yes Yes Yes VDC characteristics Yes Yes Yes Yes Yes Yes Yes Yes Region Fixed Eect Yes Yes Yes Yes Yes Yes Yes Yes Note: Fetching water, collecting rewood and dung, taking care of animals, making mats, knitting, weaving, tailoring and processing preserved food are classied as household production activities. Whereas, minor house repair, cooking food, cleaning, laundry dishes, babysitting and taking care of elderly are categorized as other household activities. 2SLS estimates are reported in the table. Share of international migrants in a VDC in 2001 is used as an instrument for the estimations. Male and female samples are working age,18 to 60, adults while boys and girls samples are restricted to ages between 6 and 17. Standard errors are clustered at VDC level. Standard errors in parentheses.∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001. Individual controls are age age2 , years of education, marital status, and household head identier. Household controls are female HH head, share of male and female adults, share of elderly, share of children, amount of land-owned, house ownership and social caste. Similarly VDC level controls are, unemployment rate, poverty rate, illiteracy rate, inequality (Gini), and urban/rural location. Regions are Kathmandu, other urban areas, Western hills, Eastern hills, Western Terai, and Eastern Terai. Coecient 1stage is the estimated coecient on the exogenous instrument in the rst-stage. Table 6: 2SLS Estimation of Labor Supply by Skill Level Wage Employment Self Employment (1) (2) (3) (4) Low Skilled High Skilled Low Skilled High Skilled Panel A: Labor market participation Household with migrant -0.205** -0.486 0.199*** 0.324 (0.098) (0.305) (0.069) (0.346) Observations 14054 2825 14054 2825 Wald χ2 3577.578 997.890 3864.717 2175.469 F -test 1stage 68.765 8.179 68.765 8.179 Coecient 1stage 1.925*** 1.460*** 1.925*** 1.460*** Panel B: Hours supply Household with migrant -15.028*** 1.032 -13.064*** 5.048 (5.754) (11.830) (4.454) (14.563) Observations 4714 1131 10319 1199 Wald χ2 2170.635 505.787 1130.111 1815.099 F -test 1stage 33.235 7.061 63.906 6.082 Coecient 1stage 1.820*** 1.665*** 1.921*** 1.503** Controls Individual characteristics Yes Yes Yes Yes Household characteristics Yes Yes Yes Yes VDC characteristics Yes Yes Yes Yes Region Fixed Eect Yes Yes Yes Yes Note: High skilled adults are dened as adults with 11 years or more of education. Sample in Panel B is conditioned on being employed in that particular sector. Log of hourly wage is added as an extra individual control for column (1) and (2). 2SLS estimates are reported in the table. Instrument used for 2SLS is share of international migrants in a VDC in 2001 (IV1). Sample is working age (18 to 60) adults and standard errors are clustered at VDC level. Standard errors in parentheses.∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Individual controls are age age2 , years of education, gender, household head identier, and marital status. Household controls are female HH head, share of male and female adults, share of elderly, share of children, amount of land-owned, house ownership and social caste. Similarly VDC level controls are, unemployment rate, poverty rate, illiteracy rate, inequality (Gini), and urban/rural location. Regions are Kathmandu, other urban areas, Western hills, Eastern hills, Western Terai, and Eastern Terai. Coecient 1stage is the estimated coecient on the exogenous instrument in the rst- stage. 32 Table 7: 2SLS Estimation of Labor Supply by Age Wage Employment Self Employment (1) (2) (3) (4) Young Adults Old Adult Young Adults Old Adults Panel A: Labor market participation Household with migrant -0.292*** -0.260** 0.252** 0.236*** (0.113) (0.122) (0.103) (0.085) Observations 10861 6018 10861 6018 Wald χ 2 2028.588 1552.717 15778.586 3651.974 F -test 1stage 58.207 50.061 58.207 50.061 Coecient 1stage 1.836*** 1.937*** 1.836*** 1.937*** Panel B: Hours supply Household with migrant -13.546* -13.516** -8.632* -14.457** (7.380) (5.618) (5.037) (5.925) Observations 3913 1932 6901 4617 Wald χ 2 2379.303 1482.415 2401.545 545.837 F -test 1stage 31.766 24.765 52.713 43.511 Coecient 1stage 1.645*** 2.078*** 1.804*** 1.957*** Controls Individual characteristics Yes Yes Yes Yes Household characteristics Yes Yes Yes Yes VDC characteristics Yes Yes Yes Yes Region Fixed Eect Yes Yes Yes Yes Note: Young adults are ages between 18 and 40 while old adults are ages between 41 and 60. Sample in Panel B is conditioned on being employed in that particular sector. Log of hourly wage is added as an extra individual control for column (1) and (2). 2SLS estimates are reported in the table. Instrument used for 2SLS is share of international migrants in a VDC in 2001 (IV1). Sample is working age (18 to 60) adults and standard errors are clustered at VDC level. Standard errors in parentheses.∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Individual controls are age age2 , years of education, gender, household head identier, and marital status. Household controls are female HH head, share of male and female adults, share of elderly, share of children, amount of land-owned, house ownership and social caste. Similarly VDC level controls are, unemployment rate, poverty rate, illiteracy rate, inequality (Gini), and urban/rural location. Regions are Kathmandu, other urban areas, Western hills, Eastern hills, Western Terai, and Eastern Terai. Coecient 1stage is the estimated coecient on the exogenous instrument in the rst-stage. 33 Table 8: 2SLS Estimates of Labor Supply by Household Head Status Wage Employment Self Employment (1) (2) (3) (4) Household Head Other Members Household Head Other Members Panel A: Labor market participation Household with migrant -0.414*** -0.231** 0.125 0.298*** (0.147) (0.100) (0.098) (0.091) Observations 5791 11088 5791 11088 Wald χ 2 906.861 1882.340 6353.722 8786.879 F -test 1stage 53.309 62.508 53.309 62.508 Coecient 1stage 1.633*** 2.030*** 1.633*** 2.030*** Panel B: Hours supply Household with migrant -17.642** -11.559* -11.369* -10.493** (7.434) (6.610) (6.662) (4.285) Observations 2772 3073 4450 7068 Wald χ 2 2009.570 2551.071 943.402 1107.469 F -test 1stage 24.419 30.862 50.477 57.718 Coecient 1stage 1.515*** 2.032*** 1.647*** 2.036*** Controls Individual characteristics Yes Yes Yes Yes Household characteristics Yes Yes Yes Yes VDC characteristics Yes Yes Yes Yes Region Fixed Eect Yes Yes Yes Yes Note: 2SLS estimates are reported in the table. Instrument used for 2SLS is share of international migrants in a VDC in 2001 (IV1). Sample is working age (18 to 60) adults and standard errors are clustered at VDC level. Standard errors in parentheses.∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Individual controls are age age2 , years of education, gender, household head identier, and marital status. Household controls are female HH head, share of male and female adults, share of elderly, share of children, amount of land-owned, house ownership and social caste. Similarly VDC level controls are, unemployment rate, poverty rate, illiteracy rate, inequality (Gini), and urban/rural location. Regions are Kathmandu, other urban areas, Western hills, Eastern hills, Western Terai, and Eastern Terai. Coecient 1stage is the estimated coecient on the exogenous instrument in the rst-stage. Sample in Panel B is conditioned on being employed in that particular sector. Log of hourly wage is added as an extra individual control for column (1) and (2). 34 A Appendix Figure A.1: Migration Trend for Labor-Employment to Top 5 Destination Countries Note: Migrants for labor-employment to foreign countries are required to obtain labor permits from the Department of Foreign Employment. Migrants can apply on their own or through a recruitment agency. Number reported in the gures are total labor permits issued to migrants who apply through the services of recruitment agencies. 35 Figure A.2: Migration Trend by gender of Migrants Note: Source is Department of Foreign Employment (2014). Migrants for labor-employment to foreign countries are required to obtain labor permits from the Department of Foreign Employment. Migrants can apply on their own or through a recruitment agency. Number reported in the gures are total labor permits issued to migrants who apply through the services of recruitment agencies. 36 Figure A.3: Top-Ten Districts of Origin for Labor Employment Migration Note: Source is Department of Foreign Employment (2014). Migrants for labor-employment to foreign countries are required to obtain labor permits from the Department of Foreign Employment. Migrants can apply on their own or through a recruitment agency. Number reported in the gures are total labor permits issued to migrants who apply through the services of recruitment agencies. These top ten districts account for 36.5% of the all labor-permits issued between 2008 and 2014. 37 Table A.1: Characteristics of Migrants Mean Standard Deviation Age 28.109 11.370 Male 0.844 0.363 Relation to HH head 0.546 0.498 (Son/Daughter) Education level Illiterate 0.172 0.377 1 to 10 grade 0.537 0.499 SLC/Intermediate 0.205 0.404 College or more 0.086 0.280 Migration to India 0.440 0.496 Malaysia 0.095 0.294 Middle east 0.247 0.432 Households with migrant 2212 ( 31.1%) Total households in the sample 7108 38 Table A.2: Migration and Labor Supply - Full Model Migration (First-Stage) OLS IV1 IV2 (1) (2) (3) (4) (5) (6) (7) (8) Migrant network 2001 (IV1) 2.088*** (0.253) Migrant network 2001 * GDP growth (IV2) 1.042*** (0.132) Household with migrant -1.604*** -0.833* -8.106*** -0.542 -7.625*** -0.717 (0.282) (0.426) (2.521) (3.422) (2.438) (3.495) Age -0.0250*** -0.0250*** 1.152*** 1.279*** 0.988*** 1.286*** 1.000*** 1.282*** (0.00240) (0.00240) (0.119) (0.110) (0.139) (0.137) (0.137) (0.139) Age squared 0.000335*** 0.000334*** -0.0158*** -0.0148*** -0.0136*** -0.0149*** -0.0137*** -0.0148*** (2.95e-05) (2.95e-05) (0.00159) (0.00141) (0.00182) (0.00177) (0.00180) (0.00181) Sex: Female 0.134*** 0.134*** -9.444*** -4.697*** -8.536*** -4.737*** -8.603*** -4.713*** (0.00809) (0.00812) (0.409) (0.598) (0.518) (0.781) (0.516) (0.801) Years of education -1.93e-05 4.40e-05 0.180*** -0.0878 0.180*** -0.0878 0.180*** -0.0878 (0.00162) (0.00162) (0.0459) (0.0642) (0.0458) (0.0644) (0.0457) (0.0642) Married -0.000824 -0.00157 -0.0914 4.051*** -0.114 4.052*** -0.112 4.051*** (0.0112) (0.0112) (0.390) (0.384) (0.415) (0.383) (0.413) (0.383) HH head 0.0865*** 0.0863*** 1.835*** 1.658*** 2.422*** 1.632*** 2.379*** 1.648*** (0.00944) (0.00944) (0.430) (0.381) (0.489) (0.474) (0.486) (0.487) Share of children 0 to 6 0.141*** 0.143*** -1.010 0.460 -0.0579 0.417 -0.128 0.443 (0.0443) (0.0443) (1.190) (1.220) (1.278) (1.320) (1.264) (1.319) Share of children 7 to 15 -0.0341 -0.0337 -0.284 0.838 -0.472 0.846 -0.458 0.841 39 (0.0309) (0.0309) (1.061) (1.100) (1.090) (1.102) (1.086) (1.104) Land own (Acres) 0.00676** 0.00669** -0.519*** 0.346*** -0.476*** 0.344*** -0.479*** 0.345*** (0.00279) (0.00279) (0.0909) (0.0770) (0.0911) (0.0819) (0.0909) (0.0822) Owns a house 0.0878*** 0.0838*** -3.403*** -2.119** -2.844*** -2.144** -2.885*** -2.129** (0.0163) (0.0166) (0.520) (0.928) (0.589) (1.001) (0.579) (0.999) Social caste: Brahmin/Chhetri -0.0334** -0.0345** -0.876** -1.352** -1.050** -1.344** -1.037** -1.348** (0.0160) (0.0160) (0.424) (0.639) (0.414) (0.635) (0.416) (0.636) Unemployment rate (VDC) 0.133 0.143 1.932 -26.93*** 2.574 -26.96*** 2.526 -26.94*** (0.103) (0.105) (3.197) (3.688) (3.214) (3.722) (3.210) (3.724) Illiteracy rate (VDC) 0.201*** 0.196*** 1.525 -5.073** 2.410 -5.113** 2.344 -5.089** (0.0556) (0.0555) (1.444) (2.034) (1.537) (2.068) (1.527) (2.065) Inequality: Income Gini (VDC) -0.114** -0.126** -1.409 2.480 -2.099 2.511 -2.048 2.492 (0.0542) (0.0550) (1.515) (1.934) (1.534) (1.947) (1.530) (1.958) Poverty rate (District) -0.168** -0.200** -4.378** -2.990 -5.422** -2.943 -5.345** -2.972 (0.0837) (0.0849) (2.066) (2.659) (2.121) (2.747) (2.109) (2.750) Constant 0.405*** 0.431*** -1.005 1.396 2.159 1.254 1.925 1.339 (0.0637) (0.0636) (2.482) (2.514) (2.875) (2.936) (2.846) (3.011) Observations 16,879 16,879 16,879 16,879 16,879 16,879 16,879 16,879 R2 0.087 0.086 0.157 0.099 0.132 0.099 0.136 0.099 Wald χ2 1996 1038 2014 1037 F -test 1stage 68.21 68.21 62.58 62.58 Region Fixed Eect Yes Yes Yes Yes Yes Yes Yes Yes ∗ Note: Standard errors clustered at VDC level. Robust standard errors in parentheses. p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Sample is working age (18 to 60) adults. Columns (1) and (2) are rst stage for IV1 and IV2 respectively. Columns (3), (5), and (7) are hours in wage employment while Columns (4), (6), and (8) are hours in self employment. Table A.3: 2SLS Estimation of Labor Supply by Gender- Full Model Female Male (1) (2) (3) (4) (5) (6) Total Hours Work Hours Total Household Total Hours Work Hours Total Household (Wage + Self ) Activity Hours (Wage + Self ) Activity Hours Household with migrant -5.702 -4.538 -1.164 -17.65** -18.34** 0.689 (6.312) (3.552) (4.960) (8.652) (7.142) (5.980) Observations 9,597 9,597 9,597 7,282 7,282 7,282 Wald χ 2 2436 960.9 2287 1528 1085 1127 40 F -test 1stage 74.90 74.90 74.90 35.55 35.55 35.55 Controls Individual characteristics Yes Yes Yes Yes Yes Yes Household characteristics Yes Yes Yes Yes Yes Yes VDC characteristics Yes Yes Yes Yes Yes Yes Region Fixed Eect Yes Yes Yes Yes Yes Yes Note: 2SLS estimates are reported in the table. Instrument used for 2SLS is share of international migrants in a VDC in 2001 (IV1). Sample is working age (18 to 60) adults and standard errors are clustered at VDC level. Robust standard errors in parentheses.∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Individual controls are age age2 , years of education, household head identier, and marital status. Household controls are share of children, amount of land-owned, house ownership and social caste. Similarly VDC level controls are, unemployment rate, poverty rate, illiteracy rate, and inequality (Gini). Table A.4: 2SLS Estimation of Labor Supply by Women's Age and Household Head Status- Full Model (1) (2) (3) (4) (5) (6) Total Hours Work Hours Total Household Total Hours Work Hours Total Householdl (Wage + Self ) Activity Hours (Wage + Self ) Activity Hours Panel A Old Young Household with migrant -16.17** -12.73*** -3.440 -0.0365 0.293 -0.330 (7.662) (4.652) (5.720) (7.072) (4.345) (5.395) Observations 3,144 3,144 3,144 6,453 6,453 6,453 Wald χ2 776.8 360.9 885.7 2287 958.8 2627 F -test 1stage 60.28 60.28 60.28 62.12 62.12 62.12 Panel B HH Head Other Members 41 Household with migrant -12.86 -12.72** -0.146 -3.036 -1.939 -1.097 (9.121) (5.703) (6.672) (6.589) (3.970) (5.135) Observations 1,593 1,593 1,593 8,004 8,004 8,004 Wald χ 2 428.1 204.5 1042 2298 921.2 2163 F -test 1stage 31.33 31.33 31.33 62.19 62.19 62.19 Controls Individual characteristics Yes Yes Yes Yes Yes Yes Household characteristics Yes Yes Yes Yes Yes Yes VDC characteristics Yes Yes Yes Yes Yes Yes Region Fixed Eect Yes Yes Yes Yes Yes Yes Note: Young adults are ages between 18 and 40 while old adults are ages between 41 and 60. 2SLS estimates are reported in the table. Instrument used for 2SLS is share of international migrants in a VDC in 2001 (IV1). Sample is working age (18 to 60) adults and standard errors are clustered at VDC level. Robust standard errors in parentheses.∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Individual controls are age age2 , years of education, household head identier, and marital status. Household controls are share of children, amount of land-owned, house ownership and social caste. Similarly VDC level controls are, unemployment rate, poverty rate, illiteracy rate, and inequality (Gini).