Skip to main content
Free

Risk Sharing and Internal Migration

IOB, University of Antwerp, and LICOS, KU LeuvenInternational Food Policy Research Institute

Abstract

Over the past 2 decades, more than half the population in our sample of rural Tanzanians has migrated out of their home communities. We hypothesize that this powerful current of internal migrants is changing the nature of traditional institutions such as informal risk sharing. Mass internal migration has created geographically disperse networks, on which we collected detailed panel data. By quantifying how shocks and consumption covary across linked households, we show that, while both migrants and stayers insure negative shocks to stayers, there is no one in the network who insures the migrants’ negative shocks. While migrants do share some of their positive shocks, they ultimately end up nearly twice as rich as those at home by 2010, despite practically identical baseline positions in the early nineties before migration. Taken together, these findings point to migration as a risky but profitable endeavor, for which the migrant will bear the risk and also reap most of the benefit. We interpret these results within the existing literature on risk sharing and on the disincentive effects of redistributive norms.

I. Introduction

If, in the next decades, Africa catches up with the rest of the world, then that will almost certainly coincide with intergenerational mobility out of rural into urban areas and out of agriculture into nonagricultural activities (Lewis 1954; Harris and Todaro 1970). Historically, in both rich, developed countries and fast-growing, developing countries, this type of migration has moved in lockstep with development and poverty reduction (Collier and Dercon 2014). Recently, China’s urban population officially surpassed its rural one: of China’s 1.35 billion people, 51% lived in urban areas at the end of 2011, rising from less than 20% in 1980 (United Nations 2012). Furthermore, UNDP (2009) reports that of the 1 billion migrants worldwide, three-quarters are internal migrants. With international migration open to only very few Africans, we should expect massive internal migration to form a core part of the development process.

The scale of this demographic process is captured in the data that form the basis of this article, further motivating our focus on internal migration. These data are part of an exceptional panel data set from the Kagera region in Tanzania, spanning nearly 2 decades of migration and development. The 2004 and 2010 follow-up surveys attempted to trace all 6,353 individuals listed on the baseline 1991–94 household rosters and reinterview them irrespective of their location. Once we exclude the 1,275 individuals who had died by 2010, we are left with 4,996 baseline individuals whose 2010 locations are known.1 Of those, 45% were found residing in the baseline village, 53% had migrated within the country, 2% to another east African country (primarily Uganda), and 0.3% had moved outside of east Africa. This region—not atypical of remote rural Africa—is clearly on the move, with internal migration dwarfing international migration.

We attempt to understand how this powerful current of internal migration, which is part and parcel of the modernization process, interacts with a traditional institution like informal risk sharing to shape economic mobility and vulnerability. This is a key question because, as Munshi and Rosenzweig (2006, 1230) put it, “a complete understanding of the development process must not only take account of the initial conditions and the role of existing institutions in shaping the response to modernization and globalization, but must also consider how these traditional institutions are shaped in turn by the forces of change.” Our analysis departs from a number of other studies in the migration literature by focusing on consumption instead of transfers. This choice of the outcome variable is motivated by the fact that risk sharing and other economic exchange could happen through a multitude of different mechanisms, of which transfers is just one. Other mechanisms could include looking for a job for someone; employing someone directly; providing him or her with tips, advice, or a network link; or providing migration opportunities (Munshi 2003). By analyzing consumption, we focus on the joint and final effect of all such mechanisms.

Work using the 2004 follow-up round has shown that geographical mobility in rural Tanzania is associated with large income gains (Beegle, De Weerdt, and Dercon 2011). We reiterate that point by showing that despite only minor welfare differences during the 1991–94 baseline survey, those who moved out of the region to other parts of Tanzania have grown roughly twice as rich as those who did not by the time we interviewed them again nearly 2 decades later in 2010. As we are measuring consumption and not income, it is clear that the main beneficiaries of this migration-led growth were the migrants themselves and certainly not their relatives who remained at home.

But did these migrants simply leave and never look back, or did they maintain links with the home community? The empirical contribution of this article is to investigate this question by exploiting the fact that the 4,282 individuals interviewed in 2010 are grouped in 816 geographically disperse extended family networks. We quantify how household consumption responds to shocks experienced by other households in the extended family network. We find that while everyone suffers from own negative shocks, only the shocks to stayers negatively affect the consumption of other network members. There is no network reaction to migrants’ negative shocks, suggesting they are not insured within the network. Those who stay at home do not seem to bear any of the negative shocks of those who move, but neither do they fully share the migrant’s growth. Stayers do share some of the migrant’s positive shocks and also receive insurance from these migrants against their own negative shocks, but migrants still outgrow stayers by a factor of 3, realizing a growth of 120% over the survey period compared to 40% for the stayers. By 2010, migrants had become nearly twice as rich as those at home, whereas they were similar in observable wealth in 1991–94.

Because selection into migration is unlikely to be random, our analysis will remain inconclusive as to whether migration is causing these empirical facts. We cannot make any statements about what would have happened if migrants had stayed home or the stayers had migrated. It is possible that in this parallel universe roles would have switched (migration is causally responsible) or not (it is driven by the unobserved differences between migrants and nonmigrants). All indications are, however, that before migration the (future) migrants did not assume any different position in the network compared to stayers. The same holds for the position taken up postmigration by return migrants.

We discuss how we can understand the two important stylized facts that emerge from this article—migrants grow much richer and become unilaterally responsive to stayers’ shocks—within the existing literature. Our results cannot be easily explained within existing models of risk sharing (Altonji, Hayashi, and Kotlikoff 1992; Coate and Ravallion 1993; Townsend 1994; Fafchamps 1999; Attanasio and Ríos-Rull 2000; Ligon, Thomas, and Worrall 2002; Genicot 2006) or within models of exchange (Lucas and Stark 1985; Hoddinott 1994). The results are more consistent with the existence of obligations of migrants toward those who remain at home. This could be in the form of a debt being paid back state contingently or through redistributive norms and altruistic feelings toward the home community (Platteau 2000; Cox and Fafchamps 2007; Burke and Young 2011). Our analysis speaks further to an emerging literature that worries about the disincentive effects of such redistributive norms. Baland, Guirkinger, and Mali (2011) show how people take out costly loans in order to conceal their income, while Platteau (2014) sees migration as a means to escape the prying eyes and incessant demands of the kinship group. The kinship poverty trap model of Hoff and Sen (2006) predicts possible resistance from the home communities as they feel threatened by productive forces leaving and severing links with home to escape taxing demands for assistance. Anticipating this, the home community may set up subtle exit barriers, which could lead to below-optimal levels of migration.

Sections II and III describe the data and econometric model, respectively. Section IV presents the main empirical findings and contains further robustness checks. Section V interprets the results within the existing literature and provides a concluding discussion.

II. Data and Descriptive Analysis

Kagera is a region in the northwestern part of Tanzania. A large part of Lake Victoria is contained within this region, and it shares a border with Burundi, Rwanda, and Uganda. The region is overwhelmingly rural, and agricultural production is the most important source of income, with more than 80% of the region’s economically active population engaged in it (URT 2012). Bananas, beans, maize, and cassava comprise the main food crops, while coffee, tea, and cotton are important cash crops. Recent years have seen a rise in horticultural cash crops, such as tomatoes, pineapples, and vanilla. According to the 2012 census, the region has a population of roughly 2.5 million people (URT 2013).

The Kagera Health and Development Survey (KHDS) was originally designed and implemented by the World Bank and the Muhimbili University College of Health Sciences. It consisted of 915 households from 51 villages that were interviewed up to four times from autumn 1991 to January 1994 (see World Bank 2004). The KHDS-2004 survey aimed to reinterview all individuals who were ever interviewed in the baseline survey and were alive in 2004. This effectively meant that the original household panel survey turned into a panel of individuals. A full household questionnaire was administered where a panel respondent was found residing. Because of household dynamics, the sample size increased to more than 2,700 households (see Beegle, De Weerdt, and Dercon 2006). The second KHDS follow-up was administered in 2010, with this time more than 3,300 households interviewed.2

Although KHDS is a panel of individuals and the definition of a household loses meaning after 10–19 years, it is common in panel surveys to consider recontact rates in terms of households. Excluding households for which all previous members were deceased, the KHDS-2004 field team managed to recontact 93% of the baseline households. In 2010, 92% of the initial households were recontacted. Taking into account the long 12- or 18-year periods between surveys, the attrition rates in KHDS-2004 and KHDS-2010 are extremely low by the standards for such panels (Alderman et al. 2001).

This article exploits the fact that the survey includes all tracked split-offs from the original household and contains particularly rich information on the current links between them. The 2004 sample contains 4,430 individuals originating from 830 initial households. The 2010 sample has 4,282 individuals, originating from 816 initial households. In 2004, the average baseline household had spawned 3.3 households out of which 1.6 were nonmigrant and 1.7 were migrant households. In 2010, the average baseline household had spawned 4.1 households out of which 1.8 were nonmigrants and 2.3 were migrants. In what follows, we refer to these networks as extended family networks.

In this article we define a migrant as anyone who has moved out of the baseline village.3 By this definition 37% of the sample is considered migrant in 2004 and 48% in 2010. Details on where they were found in 2010 are given in figure 1.4

Figure 1.
Figure 1.

KHDS-2010: Recontacting after 16+ years

These internal migration flows described above are associated with structural transformation.5Table 1 shows that out of the 1,850 migrant households in 2010, only one-third reported agriculture as their main income-generating activity. For the 1,460 nonmigrant households, this is 65%. More than 25% of the migrant households engage in informal or formal wage employment, and 11% are self-employed in the nonagricultural sector. Furthermore, migrants who move farther from the baseline village are less likely to engage in agriculture and more likely to be in wage employment.

Table 1.

Main Income-Generating Activity by Migrant Status in 2010 (%)

 Nonmigrant HouseholdMigrant Household
 AllNearby VillageElsewhere in KageraOutside Kagera
Agriculture64.933.051.341.98.5
Wage employment6.226.812.020.145.8
Self-employed8.811.210.59.514.4
Trading11.717.217.215.120.7
Casual labor5.57.66.79.55.3
Fishing1.81.7.62.6.9
Transfers and savings1.22.51.81.44.6
Number of households1,4601,850343917590

Note. Agriculture category includes farming and livestock keeping; trading includes agricultural and nonagricultural trading. Wage employment can be either formal or informal. Transfers include pensions, remittances, and rental income. Self-employed category only considers self-employment outside agriculture. The information is missing for two nonmigrant and five migrant households.

View Table Image

Table 2 provides an overview of the reasons for leaving the baseline village. More than one-third of the female respondents but none of the male respondents cited marriage as the reason for migrating, which is what one would expect in a culture with patrilocal marriages. Less than 15% of the female respondents reported that they left because of work. In contrast, almost 45% of the male migrants reported to have moved because they had found work or went looking for work.6

Table 2.

Reasons for Leaving the Baseline Village (%)

ReasonMaleFemale
To look for work29.87.5
Own schooling16.010.3
Found work15.16.7
To live in a healthier environment10.411.7
Marriage.038.9
Other reason28.824.9
 Total100.0100.0

The consumption data originate from extensive food and nonfood consumption modules in the survey, carefully designed to maintain comparability across survey rounds and to control for seasonality. The aggregates are temporally and spatially deflated using data from a price questionnaire included in the survey. Consumption is expressed in annual per capita terms using 2010 Tanzanian shillings.7

Using the 1991–2004 panel, Beegle et al. (2011) document how migrants grow much richer than their family members who did not migrate. Table 3 provides the summary of the consumption and poverty developments of the panel respondents with respect to their 2010 location. On average, consumption levels in the sample almost doubled over 19 years. Individuals who stayed in their community saw their consumption increase by more than 40%. Consumption growth for migrants was much higher: those who left Kagera saw their consumption nearly triple over the same 2 decades. The poverty statistics tell the same story: nearly all respondents who left the region managed to escape poverty, while poverty reduction among nonmigrants was more modest. These descriptive statistics, which reinforce the results reported in Beegle et al. (2011), form the first stylized fact documented in this article: migrants grow much richer than those who stay.8

Table 3.

Consumption and Poverty Movements of the Panel Respondents in 1991–2010 by 2010 Location

 Mean 1991Mean 2010Difference in MeansN
Consumption per capita (TZS) by 2010 location:    
 Within community343,718492,398148,680***2,224
 Migrant locations369,190805,702436,511***2,047
 Nearby community364,099569,438205,339***382
 Elsewhere in Kagera357,930695,951338,021***1,007
 Out of Kagera389,3791,110,827721,449***658
 Full sample355,926642,558286,632***4,271
Consumption poverty head count (%) by 2010 location:    
 Within community3119−13***2,224
 Migrant locations2813−16***2,047
 Nearby community3020−10***382
 Elsewhere in Kagera3116−15***1,007
 Out of Kagera233−21***658
 Full sample3016−14***4,271

Note. All consumption values are in annual per capita terms and expressed in 2010 Tanzanian shillings. Significance of the difference in means uses a t-test.

***. p < .01.

View Table Image

After moving, migrants remain linked to extended family members at home: 90% of the migrants in the 2010 round report that they communicated with a nonmigrant network member in the 12 months preceding the survey. Migrants who maintained some form of communication experienced an average consumption growth of 110%, while those who did not grew by 88%.9 This difference is statistically significant at the 1% level. The severing of the most basic links does not seem to be associated with higher consumption growth; if anything, the reverse is true.

We use data from shock modules administered in 2004 and 2010. During both of these rounds, the panel respondents were asked to consider each year between the survey rounds and indicate whether a particular year was, in economic terms, very good, good, normal, bad, or very bad. For each “very bad” response, the respondents were asked to provide the main reason for the hardship. We consider each “very bad” response as a negative economic shock and each “very good” response as a positive economic shock. More than 60% of the panel respondents reported experiencing at least one negative shock between 1994 and 2009. The positive shocks were less frequent, with 37% of the respondents reporting experiencing one or more.

Table 4 provides an overview of the shocks experienced. The most frequently reported negative shocks were death of a family member, serious illness, and poor harvest due to bad weather. Good harvest and high income from wage employment and crop prices were the most frequently reported positive shocks.

Table 4.

Shocks Reported by the Panel Respondents, 1994–2009

 Frequency%
Negative shock:  
 Death of family member79726
 Poor harvest due to adverse weather63821
 Serious illness57719
 Loss in wage employment2197
 Loss of assets2057
 Eviction/resettlement993
 Poor harvest due to pests or crop diseases983
 Low crop prices853
 Loss in off-farm employment783
 Low income due to lower remittances431
 Loss of livestock6.2
 Loss of gifts and support by organizations4.1
 Other reasons1726
  Total3,021100
Positive shock:  
 Good harvest19825
 High income from wage employment16120
 High crop prices15319
 High income from off-farm employment719
 New assets547
 High income from remittances233
 High income from support by development organizations162
 High returns from assets172
 Extra income from livestock71
 Other reasons10413
  Total804100

The shock data were collected at the individual level—in particular for each person on the 2004 and 2010 roster who also appears on the original 1991–94 rosters. Since our focus is to examine the role of shocks on consumption that is defined at the household level, the data had to be reformatted from the individual to the household level. If at least one individual in the current household reported to have experienced a shock, we interpret it as a household-level shock. We should also exclude shocks that occurred before the households split. Fortunately, we know the year in which the respondents out-migrated, allowing us to include only shocks that occurred at least 1 year after this move.

Furthermore, some of the shock categories are problematic to our network analysis. Mortality shocks may trigger inheritance flows within extended families. A negative shock in one household may then actually be a positive income shock in another household. A similar problem arises with the (positive or negative) remittance shocks, if these capture the loss of transfers from a household within the same extended family. We therefore exclude these two shock categories from our final shock variables.

Another worry is that we are only measuring a subset of relevant shocks. First, if shocks are self-reported, then respondents may fail to mention those that were effectively insured. Second, the extended family network in the home community may extend beyond the networks as defined in our data. Fortunately, the survey provides an alternative shock measure, which is community wide and not self-reported. We have historical rainfall data from the Tanzanian Meteorological Agency for gauges in 212 weather stations in Kagera and at the migration destinations in our sample. The drawback is that this shock measure does not allow us to quantify positive shocks: too much rain is not good for yields, especially when it falls in the wrong season (e.g., when the beans are drying in the field). We therefore treat this exercise as a robustness check for the self-reported negative shocks.

In a first step, each household is linked to all rainfall stations within a 100 kilometer radius. Next, a monthly rainfall figure is calculated, for each household, by weighing each monthly rainfall reading with the inverse of the distance of the rainfall station where it was recorded to the household in question. The mean distance to the nearest rainfall station is 17 kilometers (median 9 kilometers) among the 2004 households and 30 kilometers (median 10 kilometers) among the 2010 households. For each household we can calculate average monthly z-score deviations of rainfall during the two rainy seasons, in relation to the 30-year average (1980–2010) for that village. Rainfall shocks are then constructed by truncating the positive yearly average rainfall deviations to zero. We calculate a nonmigrant household’s own shock as the most negative shock in the 5 years before the interview round.

Table A1 presents the summary statistics for the final sample of 4,782 individuals (resulting in a total of 8,430 observations) by migration status. Migration is not random, and table A1 shows how migrants are more likely to be female, are younger, and have more years of formal education.10Section IV discusses how the endogeneity of migration comes to bear on the interpretation of our results.

III. Econometric Model

Table 3 illustrates the basic result of Beegle et al. (2011) and is the departure point of this article: despite small differences at baseline, migrants grow much richer than those who remain at home. The migrant’s growth, therefore, does not seem to be shared with family at home. Our contribution is to measure the extent to which own consumption of migrants and stayers is affected by negative shocks to others in the network. In particular, we ask whether stayers and migrants are differentially insured within the network and find that indeed they are: while both migrants and stayers remain responsive to the shocks of stayers, neither is responsive to the shocks of migrants. We will provide further evidence that before the move stayers and (future) migrants were responsive to each other’s shocks, showing that the special status of the migrant in the relationship coincided with the physical move.

The outcome variable in our econometric analysis is logged per capita consumption in period t{2004,2010}, for individual i in extended family j (lncijt). The vector of own shock variables is sijt—one for negative and one for positive shocks. The shock variables obtain value 1 if the individual experienced a shock in the previous 5 years and 0 otherwise.11The vector of network shock variables, zijt, measures the number of households in the network affected by an income shock. As before, this vector contains both positive and negative network shock variables. The shocks that occurred in an individual’s own household are excluded from these variables.

All individuals were living in the same household j at baseline. Some also shared a household in 2004 and 2010. We remain agnostic about how to treat this continued grouping of individuals into households in the follow-up surveys. In the main analysis, we think in terms of a network—and a panel—of individuals. It is, however, important to consider that both our outcome variable and our shock variable are measured at the household level. Therefore, every individual gets assigned the logged consumption value lncijt and own shock value sijt of the household in which he or she lives. Consequently, the network shock variable, zijt, is a count of the number of households containing at least one network member that have received a shock. Our analysis clusters the standard errors at the network (j) level, and the Robustness section will redo the whole analysis at the household level.

The use of consumption as the outcome variable has the advantage that it incorporates all forms of assistance, including more subtle forms that could hurt one’s own position or have an opportunity cost in terms of time (employing a relative, helping with a job search, house sharing, etc.). Furthermore, other forms of exchange, the outcomes of which are consumed within the survey period, are captured in a final consumption figure. Consumption is attractive because it is the bottom-line sacrifice someone has made, after all is said and done.

We model logged per capita consumption in period t{2004,2010}, for individual i in extended family j as

(1)lncijt=sijtβ+zijtδ+xijγ+hijtν+wjtϑ+αj+ϵijt,
where xij is a vector of individual time-invariant characteristics, such as sex, age, and a number of baseline characteristics such as relation to head, marital status, and education relative to age-specific peers (and its quadratic term).12 These characteristics are likely to influence the current level of consumption but also the role taken by the individual regarding insuring others in the network. To control for the life cycle effects associated with consumption and risk sharing, age is modeled through age interval dummies (see table A1 for details on how they are defined). The variable hijt includes time-variant individual characteristics, such as relation to the individual’s current household head, which may correlate with consumption and the level of insurance provision in the network. The variable wjt captures the time-variant network characteristics comprising the number of migrant and nonmigrant households in each period. The term aj represents the network fixed effect, and εijt is the error term. The standard errors are clustered by network.13 If negative shocks matter—and they are not completely smoothed within the network—we expect β < 0.14 Finding δ < 0 implies that individual consumption is negatively affected by negative income shocks to others in the network (some of the individual shock gets absorbed by the extended family).

After running equation (1) on the pooled sample, we run it separately for migrants and nonmigrants to establish whether there is any differential responsiveness to network shocks between these two groups. In the final version of equation (1), we also split the network shock variable into shocks to migrants in the network and shocks to stayers in the network, to explore heterogeneity in that dimension.

The ability to include network fixed effects (NFEs) makes this specification particularly powerful. First, the inclusion of NFEs means that we compare the impact of shocks between the individuals originating from the same baseline household. The NFEs control for all time-invariant observable and unobservable network characteristics. In particular, through NFEs we control for aggregate resources (e.g., income, assets) in the network. Moreover, they also capture the level of inequality within the network. It may well be that the decision to split or to migrate will be related to the level of risk sharing provided in the baseline household. In particular, household division or out-migration could be related to high inequality within the household (Foster 1993; Foster and Rosenzweig 2002), which may then be correlated with the risk-sharing arrangement after the baseline household splits—or produces a migrant.15 Fortunately, the level of inequality within the network is also captured in the NFEs.

We remain concerned about unobserved heterogeneity in who within the network decides to migrate. In Section II, we discussed how migrants are more likely to be female, are younger, and have more years of formal education. This begs the question whether, perhaps, they were already the unilaterally insuring family member, even before they moved.

We can investigate this by restricting the sample to 2,547 individuals who were identified as stayers in either 2004 or 2010. In other words, this new restricted sample of 4,397 observations drops individuals who were migrants in both 2004 and 2010. Out of these 2,547 individuals, 547 were stayers in 2004 and move by 2010, while 202 were migrants by 2004 but have returned by 2010. The essence of our test is to look at whether these individuals had already taken on a different role in the risk-sharing networks at home (i.e., with other stayers) before their move (for the 547 future migrants) or whether they continued to do so after their return home (for the 202 return migrants). Interacting the shock variable with future or past migration status (mijt) allows us to quantify the insurance relation that exists between stayers, differentiated by their future or past mobility. Building on equation (1), we now estimate

(2)lncijt=mijt+sijtβ1+sijtβ2×mijt+zijtδ1+zijtδ2×mijt+xijtγ+hijtν+wjtϑ+αj+ϵijt.
We exploit these migration dynamics by studying whether the risk-sharing role taken by these mobile individuals differs from that taken on by individuals who never migrated. If the roles are the same, then we expect δ2=0.

IV. Results

Main Results

Table 5 estimates equation (1) for the pooled sample (col. 1) and separately for the migrant (col. 2) and nonmigrant (col. 3) samples. The coefficient on the own negative shock variable appears significant in all columns, implying that the shocks we are considering are meaningful for both migrants and nonmigrants. The same is true for the own positive shocks, with the exception of the migrant column where the coefficient is not statistically significant at conventional levels (p = .149). The individuals in these networks are also responsive to negative shocks occurring to others in the same network, implying that some level of risk sharing takes place in these networks.16 The coefficient on the positive network shock variables appears insignificant in the pooled model.

Table 5.

Effect of Network Shocks on Consumption

 PooledMigrantNonmigrant
 (1)(2)(3)
Own negative shock−.138***−.078**−.153***
 (.019)(.031)(.024)
Own positive shock.080**.083.113***
 (.035)(.057)(.041)
Number of households that experienced a negative shock in the network−.039***−.038***−.027*
 (.011)(.015)(.015)
Number of households that experienced a positive shock in the network.010−.011.036
 (.021)(.032)(.027)
Number of split-off households moved.082***.060***.054***
 (.010)(.014)(.013)
Number of split-off households stayed.016.003.041**
 (.014)(.017)(.019)
Network fixed effectsYesYesYes
Other controlsYesYesYes
Number of observations8,4303,5384,892
R2.124.125.094
Adjusted R2.121.120.089

Note. Cluster-robust standard errors by network are in parentheses. Dependent variable is log household per capita consumption. Unit of observation is panel respondent.

*. p < .1.

**. p < .05.

***. p < .01.

View Table Image

Table 5 columns 2 and 3 show that for both migrants and nonmigrants, the negative network shock coefficient is negative, whereas the positive network shock coefficient is insignificant. These negative network shocks have a sizable impact on migrants’ consumption: on average, a shock in one household in the network results in a drop of 3.8% in a migrant’s household per capita consumption. This point estimate is significant at the 1% level. Also, the nonmigrants are affected by these shocks, with each network shock resulting in a 2.7% drop in stayer’s household per capita consumption. However, the coefficient is significant only at 10% level.

In order to investigate this further, we decompose the network shock variables into shocks in nonmigrant and migrant households. The first network-shock variable measures the number of nonmigrant households that experienced a shock in the extended family. The second network-shock variable measures the number of migrant households affected by shocks. As before, the individual’s own shocks have been excluded from these variables. Table 6 presents the regression results. We see that both migrants and nonmigrants are susceptible to negative shocks affecting nonmigrant households within their extended family network, while negative shocks in migrant households exert no impact for either group: the coefficient is nearly zero and insignificant in both columns. On average, a negative shock in one nonmigrant household in the network leads to a drop of 8.8% in migrant households’ consumption. This point estimate is significant at the 1% level. Similarly, a shock in one nonmigrant household in the same network results in a fall of 4.8% in nonmigrant households’ consumption. This coefficient is significant at the 5% level.

Table 6.

Effect of Network Shocks by Migrant Status

 MigrantNonmigrant
Own negative shock−.069**−.160***
 (.031)(.025)
Own positive shock.116*.104**
 (.063)(.042)
Number of nonmigrant households that experienced a negative shock in the network−.088***−.048**
 (.023)(.024)
Number of nonmigrant households that experienced a positive shock in the network−.061−.007
 (.056)(.035)
Number of migrant households that experienced a negative shock in the network.003−.003
 (.021)(.021)
Number of migrant households that experienced a positive shock in the network.058.097**
 (.049)(.048)
Number of split-off households moved.051***.050***
 (.015)(.014)
Number of split-off households stayed.013.048***
 (.017)(.019)
Network fixed effectsYesYes
Other controlsYesYes
Number of observations3,5384,892
R2.129.097
Adjusted R2.123.092

Note. Cluster-robust standard errors by network are in parentheses. Dependent variable is log household per capita consumption. Unit of observation is panel respondent.

*. p < .1.

**. p < .05.

***. p < .01.

View Table Image

Positive network shocks that take in place in nonmigrant households do not exert any impact on either group’s consumption. Interestingly, however, positive shocks taking place in migrant households appear with a positive and significant sign in the nonmigrant column. On average, a positive shock in one migrant household in the network leads to a gain of 9.7% in nonmigrant households’ consumption.

These econometric results comprise the second stylized fact: negative shocks to stayers are insured through their migrant network and their home network, while negative shocks to migrants are uninsured within these networks. Stayers also benefit from positive shocks to migrants, but not vice versa.

We next turn to the question whether migrants had this peculiar position in the network before becoming migrants—or after they returned home. To investigate this, we estimate equation (2), with results reported in table 7. The results from table 7 show that these future or past migrants are not more (or less) responsive to their own and other (stayer) network shocks compared to their sedentary network members. Put differently, while living at home these mobile individuals do not take on any special role in the network: they are equally responsive to their own and other stayers’ shocks as everyone else in the baseline community.

Table 7.

Interactions with Future or Past Migration Status

 Nonmigrant
Future or past migrant−.060
 (.044)
Own negative shock−.196***
 (.025)
Own negative shock × future or past migrant.018
 (.044)
Own positive shock.129***
 (.043)
Own positive shock × future or past migrant−.034
 (.061)
Number of nonmigrant households that experienced a negative shock in the network−.055**
 (.025)
Number of nonmigrant households that experienced a negative shock in the network × future or past migrant.002
 (.033)
Number of nonmigrant households that experienced a positive shock in the network−.006
 (.035)
Number of nonmigrant households that experienced a positive shock in the network × future or past migrant.108
 (.084)
Number of split-off households stayed.045**
 (.018)
Network fixed effectsYes
Other controlsYes
Number of observations4,397
R2.086
Adjusted R2.081

Note. Cluster-robust standard errors by network are in parentheses. Dependent variable is log household per capita consumption. Unit of observation is panel respondent. Sample is restricted to individuals observed as nonmigrants in t = {2004, 2010}.

**. p < .05.

***. p < .01.

View Table Image

Robustness

We conducted an array of robustness checks to validate our second stylized fact. First, similar results hold using the rainfall shock variable. However, as discussed in Section II, we can only verify our results regarding the negative shocks. The first row in table A2 shows that rainfall shocks are important in determining consumption growth, with every standard deviation decrease in (negative) rainfall deviation causing consumption growth to decline by 7% for migrants and 15% for nonmigrants.17

Knowing that rainfall shocks drive the incomes of both stayer and migrant households, we can use them as an alternative network shock indicator. We replace the network shock variable with the baseline village rainfall shock variable in equation (1). For migrant households, this rainfall shock is constructed as the most negative rainfall deviation in the baseline village after the migrant left. For stayer households, we take the most negative rainfall deviation among the migrant household locations, after the migrant left. Table A2 columns 1 and 2 report the results for the migrant households. We see that, after the migrants leave, their consumption remains responsive to rainfall shocks at the baseline village. Each standard deviation decrease in (negative) rainfall deviation in the baseline village leads, on average, to a 7.5% fall in consumption in the migrant households. Columns 3 and 4 report the corresponding results for the nonmigrant households. Consistent with the results presented earlier, we see that nonmigrants are not affected by rainfall shocks that take place in migrant households.

Second, the results are not driven by other important life events such as changes in marital status. Table A3 shows that the results presented in table 6 hold if we add dummies for the current marital status to the specification.

Third, the shock variables consider the last 5 years prior t. Using shocks that happened in the previous year (i.e., t − 1) does not alter our findings. Table A4 shows that considering a shorter shock window yields similar results as in table 6. The coefficients on the own shock variables turn insignificant in this specification, possibly due to the small cell size in this variable (less than 2% of the full sample report a positive shock in t − 1).

Fourth, the results are not driven by the configuration of the data. We conducted the analysis at the individual level to facilitate better modeling of the within-network relationships and differences in individual-level characteristics. Conducting the empirical analysis at the household level, however, does not affect our main findings. Table A5 reruns table 6 using household-level data.

Fifth, the demographic composition may systematically differ between the migrant and nonmigrant households. Therefore, the use of per capita consumption as the dependent variable may not be entirely appropriate. To address this issue, we defined household consumption per adult equivalent instead of per household member. Table A6 provides the results. The shock coefficients and their standard errors are of similar magnitude. The difference is that the positive migrant network shock coefficient turns insignificant in the nonmigrant column.

Finally, equation (1) exploits panel data but treats repeated individual-level observations as independent. We addressed this concern by replacing the NFEs in table 6 with individual-level fixed effects. However, we cannot use the full sample for this exercise. The within transformation requires that we have two observations for each individual. Therefore, in the subsample regressions we can only consider those individuals who appear either as migrants or nonmigrants in both rounds. Columns 1 and 3 in table A7 replicate table 6 using these reduced subsamples. The magnitude of the stayer network-shock coefficient in the migrant column reduces by a third but remains significant at the 5% level. In table A7 columns 2 and 4, we replace the NFEs with individual-level fixed effects. As expected, the use of individual-level fixed effects takes a toll on the efficiency of these estimates, but still the stayer-network shock coefficient appears negative and significant at the 10% level in the migrant column. Of note is that the coefficient is of similar magnitude as in column 1, suggesting that the individual-level heterogeneity is not driving the results in table 6.

V. Interpretation and Concluding Discussion

We find that consumption of both migrants and stayers co-moves with own shocks. This empirical result holds after controlling for aggregate network resources, which indicates that these networks are not fully insuring their members, in line with a lot of the literature on this topic. Still, some insurance takes place. Interestingly it is only the stayers who have their negative shocks insured: migrants and stayers alike cut back consumption when a stayer in their network is hit by a negative shock. The negative shocks of migrants, however, are not insured: neither migrants nor stayers cut back their consumption when a migrant is hit by a shock. Migrants share their positive shocks with stayers, but not vice versa. Further analysis reveals that before their move (future) migrants did not share risk differently with other stayers in the networks—any differences in how they participate in the insurance network seem to coincide with the physical move the migrant makes. Even though migrants lack such insurance from their network, they are nevertheless much more successful than those at home when it comes to consumption growth. While migrants more than double their consumption from 1991–94 to 2010, those who have remained at the baseline location grow by 40% over the same time period. Taken together, these findings point to migration as a risky but profitable endeavor, for which the migrant will bear the risk but also reap most of the benefit. This can be interpreted within a number of strands of the literature.

With respect to the risk-sharing literature, this observed unilateral insurance relationship is difficult to explain within general models of risk sharing (Altonji et al. 1992; Coate and Ravallion 1993; Townsend 1994; Fafchamps 1999; Attanasio and Ríos-Rull 2000; Ligon et al. 2002; Genicot 2006). In particular, there should be no subgroups of households—delineated along exogenous or endogenous characteristics—that are completely unresponsive to the shocks of others. It is on this basis that we reject these basic risk-sharing models.

Recent work in the risk-sharing literature presents a more specialized version of the risk-sharing model that explicitly incorporates income inequality across agents. Indeed, with heterogeneous risk preferences, a Pareto-efficient contract allocates more aggregate risk to less risk-averse households (Schulhofer-Wohl 2011; Mazzocco and Saini 2012; Chiappori et al. 2014). Our empirical results could be consistent with an extreme version of this phenomenon in which the poorer, more risk-averse stayer pays an insurance premium to the richer, less risk-averse migrant. In this model, migrants, in effect, sell insurance to the stayers, and regressive transfers result (Fafchamps 1999; Genicot 2006).

An alternative explanation to the observed lack of reciprocity could be that migrants insure nonmigrants in exchange for other benefits. Some of these benefits may even accrue to the migrant in the more distant future. Lucas and Stark (1985) mention that there could be exchange motives for insurance provision relating to the desire for nonmigrants to look after local assets, the intention to return home, and the aspiration to inherit. In a context that lacks technology to allow future income to be consumed now, we could confuse unilateral insurance with postponed reciprocity. De Weerdt and Hirvonen (2013) explore these explanations but find no support for any of the three exchange motives mentioned above.

Of particular interest in this context is the issue of return migration. Indeed, even if migrants do not have some of the main shocks insured, some of them do return home and, as table 7 suggests, are reinserted into the risk-sharing system. Hirvonen and Lilleør (2015) discuss return migration in more detail and find that return in this context is associated with an unsuccessful migration experience. Returning can then be viewed as a final fallback option for the migrants when everything else fails. Still, the evidence does not support the notion that migrants engage in strategic remittance behavior to keep their return options open.

We think that the unilateral insurance provision documented in this article is more consistent with risk sharing motivated by social norms. Such redistributive values may have been instilled since childhood and carefully nurtured through oral transmission, rituals, and ceremonies in which the importance of the kinship group is strongly emphasized (Lévi-Strauss 1969). Remittances and other forms of assistance may buy social prestige or political power or serve to perpetuate subordination (Platteau and Sekeris 2010; Platteau 2014). In the risk-sharing literature, social norms have been seen as the glue that keeps the risk-sharing contract from breaking apart by alleviating enforcement and information problems (Stark and Lucas 1988; Fafchamps 1999; Foster and Rosenzweig 2001). Theoretically this can be modeled as subjective satisfaction that individuals receive from participation (Fafchamps 1999; Foster and Rosenzweig 2001; De Weerdt and Fafchamps 2011). The satisfaction can stem from the fulfilment of obligations and the avoidance of social sanctions, such as guilt, shame or ridicule, or fear of witchcraft. It can also include altruism, which we do not attempt to distinguish from social norms. Social norms could weaken the constraints to risk sharing to the extent that they never bind and allow for the existence of sustained, unreciprocated transfers, as documented, for example, for Paraguay by Schechter and Yuskavage (2011) and for Tanzania by De Weerdt and Fafchamps (2011). Finally, there may be obligations the migrant has at home, for example, related to investments in the migrant’s education or the financing of the move. The empirical patterns we describe could occur if migrants are repaying these loans state contingently postmigration.

We believe that our results are indicative of redistributive norms and can provide further interpretation with regard to possible disincentive effects that may result. Platteau (2014) discusses how redistributive pressures can discourage effort, entrepreneurship and risk taking. Regarding the latter, he notes that “these pressures operate in an asymmetrical manner: if the investment project fails, the risk taker will be the only one to bear the burden of the ensuing loss, while, if the project is successful, the risk taker will have to share the benefits with his or her kith and kin. Given a certain degree of risk aversion, a dynamic individual will therefore refuse to embark on a risky project that he (she) would have attempted in the absence of redistributive norms” (168–69).18 Platteau continues by outlining three possible strategic reactions for dynamic individuals to undertake. First, they could engage in the strategic hiding of income and assets. An excellent example of this is Baland et al. (2011), who show how people take out costly loans in order to conceal their income. Second, religious conversion is one strategy that could serve as a respectable way to distance oneself from some of the traditional obligations and to be, instead, subject to a new set of obligations. The final avenue would be physical separation through migration.

With respect to this latter interpretation, it is important to note that migrants are allowed to grow, albeit without any insurance from the home community but also with relatively little tax on their wealth. Migrants do share their positive shocks with nonmigrants but, after all is said and done, end up almost twice as rich as stayers in 2010, while they had started from similar baseline positions in the early nineties. This would fit well with the idea of migration as an escape from the traditional kin systems. In that respect it is interesting to calculate the cost the migrant incurs for providing the kind of unilateral insurance we have documented above. From table 6 we observe that for each negative shock in the extended family network at home there is a drop of 8.8% in the migrant’s consumption. The average migrant has 0.45 nonmigrant households that experienced a negative shock in their network (table A1), resulting in an implied consumption penalty of 4.0% (the 95% confidence interval ranges between 1.9% and 6.0%).19 We conclude that migrants share 4% of their consumption with home communities through insurance provision.20

To many readers, this number will seem relatively low and suggestive that migrants’ growth is not stifled in any significant way by the kinds of demands from the home communities discussed in this article. By way of conclusion, we note that also that the experimental literature on income hiding has come up with similar single-digit tax rates. Jakiela and Ozier (2016) find that women in a laboratory setting in Kenya purposefully reduced their income in order to keep it hidden. They acted as if they were expecting any observable winnings to be taxed at around 4%–8%. Ambler (2015) reports that El Salvadorian migrants living around Washington, DC, remit 5% more of a windfall income if they are told that the organizers of the experiment will inform potential recipients at home about it. One important difference between these experiments and our observational data is that they look at the short-run reactions to windfall incomes, while we study the long-run consequences of reactions to actual income shocks. Another difference is that they look at how people change remittance behavior when going from actual belief sets to full information or how much they would be willing to sacrifice to avoid a full information state of the world. We look at the effect of shocks within real-world belief sets and in the context of migration.

Appendix
Table A1.

Summary Statistics

 MigrantNonmigrant
Male.396.529
 (.489)(.499)
Log per capita household consumption13.1612.81
 (.737)(.563)
Own negative shock.269.484
 (.444)(.500)
Own positive shock.0571.0748
 (.232)(.263)
Number of households that experienced a negative shock in the network.832.789
 (1.072)(1.011)
Number of households that experienced a positive shock in the network.179.197
 (.504)(.538)
Number of nonmigrant households that experienced a negative shock in the network.448.410
 (.700)(.677)
Number of nonmigrant households that experienced a positive shock in the network.0842.108
 (.332)(.370)
Number of migrant households that experienced a negative shock in the network.384.380
 (.721)(.709)
Number of migrant households that experienced a positive shock in the network.0947.0895
 (.327)(.330)
Number of split-off households moved3.5431.997
 (2.031)(1.839)
Number of split-off households stayed1.6882.480
 (1.394)(1.510)
Head of the current household.354.431
 (.478)(.495)
Spouse of the current household head.371.176
 (.483)(.381)
Child of the current household head.114.262
 (.317)(.440)
Baseline characteristic:  
 Head or spouse.0825.299
 (.275)(.458)
 Biological child of head.488.474
 (.500)(.499)
 Grandchild of the head.191.0981
 (.393)(.298)
 Unmarried.897.717
 (.305)(.450)
 Unmarried male.361.409
 (.480)(.492)
 Baseline age 0–15 (reference category).666.506
 (.472)(.500)
 Baseline age 16–25.239.185
 (.426)(.388)
 Baseline age 26–35.0404.0983
 (.197)(.298)
 Baseline age 36–45.0263.0828
 (.160)(.276)
 Baseline age 46–55.0130.0664
 (.113)(.249)
 Baseline age 56–65.0150.0615
 (.121)(.240)
 Baseline age 66+.00565.0166
 (.0750)(.128)
 Deviation from median school years of peer group−.117−.789
 (1.943)(2.376)
 Deviation from median school years of peer group23.7876.265
 (10.09)(12.31)
Observations3,5384,892

Note. Standard deviations are in parentheses.

View Table Image: 1 | 2
Table A2.

Recalculating Insurance Provision through Rainfall Shocks

 MigrantNonmigrant
 Mean
(1)
Coefficient
(2)
Mean
(3)
Coefficient
(4)
Max rain shock in own location in the past 5 yearsa−.84.072**−1.07.146***
 [.52](.033)[.44](.039)
Max rain shock in deviation in baseline villagea−.62.075**  
 [.58](.035)  
Max rain shock in deviation in migrant locationsb  −.94.018
   [.60](.035)
Network fixed effectsNAYesNAYes
Other controlsNAYesNAYes
Number of observations3,5384,892
R2NA.128NA.075
Adjusted R2NA.123NA.071

Note. Standard deviations in brackets. Cluster-robust standard errors by network are in parentheses. Dependent variable is log household per capita consumption. Unit of observation is panel respondent.

a . For migrants, this is after they migrated.

b . After the migrant moved to their 2004 or 2010 location.

**. p < .05.

***. p < .01.

View Table Image
Table A3.

Replicating Table 6 with Additional Controls

 MigrantNonmigrant
Own negative shock−.083***−.167***
 (.030)(.025)
Own positive shock.105*.106**
 (.061)(.042)
Number of nonmigrant households that experienced a negative shock in the network−.082***−.048**
 (.023)(.024)
Number of nonmigrant households that experienced a positive shock in the network−.060−.005
 (.058)(.035)
Number of migrant households that experienced a negative shock in the network−.001−.003
 (.021)(.021)
Number of migrant households that experienced a positive shock in the network.053.095**
 (.048)(.048)
Number of split-off households moved.049***.051***
 (.014)(.014)
Number of split-off households stayed.007.048***
 (.017)(.018)
Network fixed effectsYesYes
Current marital status dummiesYesYes
Other controlsYesYes
Number of observations3,5384,892
R2.169.107
Adjusted R2.162.102

Note. Cluster-robust standard errors by network are in parentheses. Dependent variable is log household per capita consumption. Unit of observation is panel respondent.

*. p < .1.

**. p < .05.

***. p < .01.

View Table Image
Table A4.

Replicating Table 6 Using a Shorter Shock Window

 MigrantNonmigrant
Own negative shock (t − 1)−.143***−.172***
 (.032)(.025)
Own positive shock (t − 1).014.027
 (.098)(.061)
Number of nonmigrant households that experienced a negative shock in the network (t − 1)−.076***−.059**
 (.024)(.027)
Number of nonmigrant households that experienced a positive shock in the network (t − 1).008−.062
 (.076)(.047)
Number of migrant households that experienced a negative shock in the network (t − 1).009−.029
 (.022)(.023)
Number of migrant households that experienced a positive shock in the network (t − 1).051.152**
 (.074)(.060)
Number of split-off households moved.044***.040***
 (.015)(.015)
Number of split-off households stayed−.002.035*
 (.017)(.020)
Network fixed effectsYesYes
Other controlsYesYes
Number of observations3,5384,892
R2.129.104
Adjusted R2.123.099

Note. Cluster-robust standard errors by network are in parentheses. Dependent variable is log household per capita consumption. Unit of observation is panel respondent.

*. p < .1.

**. p < .05.

***. p < .01.

View Table Image: 1 | 2
Table A5.

Replicating Table 6 Using Household-Level Data

 MigrantNonmigrant
Own negative shock−.047−.140***
 (.030)(.024)
Own positive shock.096.063
 (.064)(.040)
Number of nonmigrant households that experienced a negative shock in the network−.083***−.045**
 (.022)(.021)
Number of nonmigrant households that experienced a positive shock in the network−.072−.010
 (.049)(.039)
Number of migrant households that experienced a negative shock in the network.013−.011
 (.022)(.020)
Number of migrant households that experienced a positive shock in the network.058.096**
 (.050)(.038)
Number of split-off households moved.051***.050***
 (.015)(.013)
Number of split-off households stayed.014.050***
 (.018)(.016)
Network fixed effectsYesYes
Other controlsYesYes
Number of observations3,0752,651
R2.136.083
Adjusted R2.129.074

Note. Cluster-robust standard errors by network are in parentheses. Dependent variable is log household per capita consumption. Unit of observation is household observed in 2004 or 2010.

**. p < .05.

***. p < .01.

View Table Image
Table A6.

Replicating Table 6 Using (log) Consumption per Adult Equivalent as a Dependent Variable

 MigrantNonmigrant
Own negative shock−.082***−.181***
 (.029)(.025)
Own positive shock.105*.087**
 (.058)(.042)
Number of nonmigrant households that experienced a negative shock in the network−.074***−.041*
 (.022)(.024)
Number of nonmigrant households that experienced a positive shock in the network−.055−.001
 (.053)(.034)
Number of migrant households that experienced a negative shock in the network.008−.002
 (.020)(.021)
Number of migrant households that experienced a positive shock in the network.065.074
 (.047)(.046)
Number of split-off households moved.050***.052***
 (.014)(.014)
Number of split-off households stayed.009.047**
 (.017)(.019)
Network fixed effectsYesYes
Other controlsYesYes
Number of observations3,5384,892
R2.126.133
Adjusted R2.119.129

Note. Cluster-robust standard errors by network are in parentheses. Dependent variable is log household per adult equivalent consumption. Unit of observation is panel respondent.

*. p < .1.

**. p < .05.

***. p < .01.

View Table Image: 1 | 2
Table A7.

Replicating Table 6 Using Individual-Level Fixed Effects

 MigrantNonmigrant
 (1)(2)(3)(4)
Own negative shock−.064*−.089***−.172***−.203***
 (.033)(.031)(.026)(.028)
Own positive shock.128**.089.115***.132***
 (.063)(.063)(.043)(.048)
Number of nonmigrant households that experienced a negative shock in the network−.059**−.045*−.049**−.029
 (.023)(.024)(.025)(.029)
Number of nonmigrant households that experienced a positive shock in the network−.051−.054−.020−.040
 (.058)(.061)(.033)(.034)
Number of migrant households that experienced a negative shock in the network.017.037−.004−.003
 (.023)(.027)(.021)(.022)
Number of migrant households that experienced a positive shock in the network.055.042.093*.092*
 (.055)(.067)(.049)(.051)
Number of split-off households moved.034**.029*.049***.040***
 (.015)(.015)(.015)(.015)
Number of split-off households stayed−.004−.008.046**.032
 (.020)(.019)(.019)(.020)
Network fixed effectsYesNoYesNo
Individual-level fixed effectsNoYesNoYes
Other controlsYesYesYesYes
Number of observations2,1504,143
R2.102.081.106.153
Adjusted R2.091.076.101.151

Note. Cluster-robust standard errors by network are in parentheses. Dependent variable is log household per capita consumption. Unit of observation is panel respondent. Migrants (nonmigrants) sample is formed of individuals who appear as migrants (nonmigrants) in 2004 and 2010.

*. p < .1.

**. p < .05.

***. p < .01.

View Table Image

Notes

The fieldwork was primarily funded by the Rockwool Foundation and the World Bank, with additional funds provided by AFD, IRD, and AIRD through the Health Risks and Migration grant of the William and Flora Hewlett Foundation. Kalle Hirvonen gratefully acknowledges the financial support from the Economic and Social Research Council (grant ES/I900934/1), the Finnish Cultural Foundation, and the Yrjö Jahnsson Foundation. Stefan Dercon was instrumental in conceptualizing this article. We further thank Kathleen Beegle, Marcel Fafchamps, Garance Genicot, Markus Goldstein, Flore Gubert, Cynthia Kinnan, Andy McKay, Imran Rasul, Martin Ravallion, Barry Reilly, and seminar and conference participants at Aix-Marseille School of Economics, BREAD, CSAE (Oxford), FUNDP, Georgetown, Leuven, LICOS, NEUDC (Dartmouth), Paris School of Economics, Sussex, and UNU-WIDER for useful comments. The usual disclaimer applies. For more information, contact Joachim De Weerdt () or Kalle Hirvonen ().

1 . We lack location information on 82 individuals. Because this is after multiple attempts through various sources, it is unlikely that these individuals have moved outside of east Africa. Information on such an important, low-occurrence event is unlikely to be hidden.

2 . Whereas the 2004 round was conducted on paper, the 2010 round was conducted on handheld devices (Caeyers, Chalmers, and De Weerdt 2012). De Weerdt et al. (2012) provide a full overview of the survey.

3 . Our results are robust to alternative migrant definitions, such as also defining households that moved to a nearby village as nonmigrant households.

4 . A similar figure for the 2004 round is presented in Beegle et al. (2011).

5 . This is also documented by Christiaensen, De Weerdt, and Todo (2013), who use the same data to study the role of urbanization and diversification in poverty reduction.

6 . Despite these differences in migration motives across the two gender groups, we do not find any statistically significant differences in risk-sharing provision between male and female migrants. Results are not reported but available on request.

7 . Using adult equivalent units as the denominator instead of household size produces almost identical results across all specifications.

8 . This finding is not driven by the fact that migrants are the younger generation. The divergence between migrants and stayers observed in table 3 remains even if we net out the age effects. Results available on request from the authors.

9 . The mean consumption growth among those who maintained contact was TZS 394,679, and among those who severed links, TZS 286,991.

10 . These reported differences are statistically significant.

11 . If t = 2004, we consider shocks that took place in 1999–2003. If t = 2010, the shock window is 2005–9. Our results are robust to considering t − 1 shocks only; i.e., 2003 if t = 2004 and 2009 if t = 2010 (see Sec. IV).

12 . A number of individuals in our data had not yet completed schooling in 1991–94. A raw measure of education would consequently be highly correlated with age. To circumvent this problem, we follow Beegle et al. (2011) in computing the years of schooling relative to peers and use that variable in our empirical analysis.

13 . The total sample of 8,430 individuals groups into 779 networks. Our results hold if we cluster the standard errors at the baseline village level.

14 . We do not attempt to test a full risk-sharing model (e.g., Altonji et al. 1992; Townsend 1994). Recent literature notes that the rejection of the full risk-sharing model in this type of specification may stem from the violation of the assumption that risk preferences are identical within the network (Schulhofer-Wohl 2011; Mazzocco and Saini 2012; Chiappori et al. 2014). In a context of heterogeneous risk preferences, a Pareto-efficient contract allocates more aggregate risk to less risk-averse households. As demonstrated by Schulhofer-Wohl (2011), Mazzocco and Saini (2012), and Chiappori et al. (2014), this would lead to a upward bias (in absolute terms) in β in eq. (1). The standard full risk-sharing test is then biased against the null hypothesis of full risk sharing.

15 . By “household division,” we refer to an event in which a household splits into two or more households. Migration is then one, special, form of household division.

16 . Note that shocks are only weakly correlated within these extended family networks: the intraclass correlation coefficient for the own negative shock variable equals .076 with a standard error of .008.

17 . Although only one-third of the migrant households report agriculture as their main income-generating activity (table 1), nearly two-thirds cultivate land. This explains why also the migrants are susceptible to rainfall shocks.

18 . These predictions remain to be empirically verified. D’Exelle and Verschoor (2015), e.g., find the opposite is true in a lab-in-the-field experiment in Uganda. They find that investments increase when profits can be shared or when losses cannot be shared.

19 . The same calculations based on the rainfall shock regressions in table A2 show that average migrants sacrifice 4.7% out of their consumption to insure their network members back home.

20 . Migrants also share part of the gains from their positive income shocks (table 6). Each positive shock in a migrant household results in a 9.7% increase in nonmigrants’ consumption. Since an average stayer has only 0.09 positive network shocks compared to migrants, the magnitude of the implied penalty here is very small: 0.09%—expressed in terms of stayer households’ consumption.

References