DISCUSSION PAPER SERIES

IZA DP No. 3276

Why Remit? The Case of Nicaragua

George S. Naufal

January 2008

Forschungsinstitut
zur Zukunft der Arbeit
Institute for the Study
of Labor
 Why Remit?
The Case of Nicaragua

George S. Naufal
American University of Sharjah
and IZA

Discussion Paper No. 3276

January 2008

IZA

P.O. Box 7240

53072 Bonn
Germany

Phone: +49-228-3894-0
Fax: +49-228-3894-180
E-mail: iza@iza.org

Any opinions expressed here are those of the author(s) and not those of IZA. Research published in
this series may include views on policy, but the institute itself takes no institutional policy positions.

The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center
and a place of communication between science, politics and business. IZA is an independent nonprofit
organization supported by Deutsche Post World Net. The center is associated with the University of
Bonn and offers a stimulating research environment through its international network, workshops and
conferences, data service, project support, research visits and doctoral program. IZA engages in (i)
original and internationally competitive research in all fields of labor economics, (ii) development of
policy concepts, and (iii) dissemination of research results and concepts to the interested public.

IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion.
Citation of such a paper should account for its provisional character. A revised version may be
available directly from the author.
IZA Discussion Paper No. 3276
January 2008

ABSTRACT

Why Remit? The Case of Nicaragua

In the last two decades remittances have gained interest due to their large size. For several
developing countries remittances constitute a large portion of their GDP and sometimes
exceed FDI. While FDIs are usually profit driven, it is not clear what the driving force behind
remittances is. This paper presents a simple theoretical model of migrants' remitting
behavior. I consider two general motivations for remitting: altruism and self-interest. Using a
heteroskedastic Tobit with a known form of variance I test the findings of the theoretical
model with data from Nicaragua. Evidence suggests that migrants from Nicaragua remit for
altruistic reasons. Moreover some gender heterogeneity seems to exist in the remitting
behavior.

JEL Classification: J61, O15, D64

Keywords: remittances, censored regression, altruism, Central America, Nicaragua

Corresponding author:

George S. Naufal
Department of Economics
School of Business and Management
American University of Sharjah
P.O. Box 26666
Sharjah
United Arab Emirates
E-mail: gnaufal@aus.edu
1. Introduction

In the last two decades remittances have been on the rise. Official estimates show that

remittances averaged around 60 billion U.S. dollars per year in the 1990s (World Bank)

and reached 167 billion U.S. dollars in 2005 (World Bank’s Global Economic

Prospects). Several studies document that remittances already exceed foreign aid and

foreign direct investment (FDI) for some developing countries Connell and Brown

(2004), De Haas (2006), Heilmann (2006) and Chami et al. (2006). This fact raised

questions on whether remittances can be seen as a possible source of growth Durand et

al. (1996) and Widgren and Marin (2002).

Remittances differ from other types of capital flows in three main aspects. First,

remittances go directly into the hands of the households in the receiving countries rather

than indirectly through private or governmental institutions. Second, capital flows such

as FDIs are in general profit driven and therefore are positively related to GDP growth.

However, this is not always the case for remittances. Remittances are not always profit

driven and can be altruistically motivated. Finally, FDIs tend to be less stable relative to

remittances Orozco (2002).

Uncovering the reasons for remitting is crucial for policy implication for several

reasons. From the original household perspective, the forces behind remittances can shed

some light on households’ migration strategies De La Brière et al. (2002). In fact

Hoddinott (1994) stresses that remittances should be incorporated in the model of

household migration decisions. Hoddinott also notes that remittances can be part of a

2
long term contract between the head of the original household and the migrating

member.

From a macroeconomic look, remittances are thought to be intended to ease the

burden of poor economic performance on local recipients Chami et al. (2005). Therefore

altruistically motivated remittances are expected to be countercyclical with income

growth and consequently can decrease the scope of the government intervention in

recession times. In this particular case, policies built on predictions that remittances

behave in the same manner as other types of capital flows might have unanticipated

consequences.

The literature on remittances has mostly focused on finding the determinants of

remittances. In this paper I present a simple theoretical model of remittance behavior. I

consider remittances as unidirectional flows from the migrant in a host country to the

original household in the home country which I refer to in this paper as the receiving

household. This allows me to consider the reaction of remittances to a bad state outcome

on the receiving household. This is the first paper that looks at the response of

remittances to shocks that pertain to the receiving household. This is crucial in terms of

investigating the remittance behavior since most remittances consider the migrant as a

source and the receiving household as the end destination and therefore, they are

expected to react to any income shocks at the receiving end. This setup gives two broad

motivations for remitting: altruism where migrants simply care about the receiving

household members’ welfare and self-interest where migrants remit for investment

opportunities that are expected to yield a certain payoff in the future. I test the theoretical

3
predictions of this model using survey data from Nicaragua. I quantify the results of the

heteroskedastic Tobit for policy purposes.

Altruism seems to be the main motivation behind the remitting behavior to

Nicaragua. Moreover the remitting behavior is not identical across gender. Female

migrants seem to behave more altruistically toward the receiving household.

This paper proceeds as follows. Section 2 provides a brief summary of the

existing literature. Section 3 presents a simple theoretical model of remittance behavior.

Section 4 introduces the data and explains the estimation method. Section 5 includes the

results and section 6 represents the conclusion.

2. Literature Review

Lucas and Stark (1985) discuss several hypotheses for motivations to remit. Three

reasons for remitting are presented ranging from pure altruism to pure self-interest

spanning a more tempered point of view combining these two extremes. Under pure

altruism a migrant derives utility from the utility of those persons at home. A migrant

therefore enjoys remitting because this will subsequently increase his utility. Under pure

self-interest the migrant's satisfaction depends on self-interest goals that range from

inheritance, investments, and the intention of one day returning home. A third possible

motive is viewing remittances as part of an arrangement between the migrant and

persons at home. This arrangement is seen as a mutually beneficial contract between the

two parties.

Agarwal and Horowitz (2002) is one of the first papers that relate the remittance

behavior and the motivation behind remitting in a theoretical model. Agarwal and

4
Horowitz set up a two period model taking into consideration the possibility of multiple

migrants per household. They solve for the first order conditions of a migrant’s expected

utility function and define an implicit remittance function for two cases: pure altruism

and the insurance motive. The key result lies in the significant effect of the number of

other migrants on remittance under altruism. However the number of migrants does not

affect average remittance under the risk-sharing case. Agarwal and Horowitz use data for

Guyana to test their theoretical predictions. Their empirical findings show significant

differences in the remitting process of migrants from multiple and single migrants’

households. Their findings support altruism as a main motivation for remitting.

Brown and Poirine (2005) make use of the theory of intergenerational transfers to

sketch a two-period informal, intrafamilial loan arrangement to analyze migrants’

remittances of Pacific Island migrants in Sydney, Australia. They develop an alternative

theory based on parental behavior that lies between strong altruism and self-interest that

they refer to as “weak altruism”. Their results imply that neither strong altruism nor pure

self-interest needs to be used to explain intergenerational transfers in low-income

countries. They suggest linking the theory of private intergenerational transfers, the

theory of human capital investment to the theory of migrants’ remittances when

investigating remittance behavior.

In a more recent paper Amuedo-Dorantes and Pozo (2006) stress upon the part of

remittances transferred to buy two types of insurance: family-provided and self-provided

insurance. The authors use data on Mexican immigrants to measure income risk and find

5
that increases in the latter raise both the likelihood and the percentage of migrants’

earnings remitted for insurance purposes.

All the papers listed above focus on the risk sharing aspect of remitting by

investigating the effects of a bad state outcome in the host country on the migrants’

remitting behavior. While an income shock in the host country is important in

determining the remitting ability of the migrant, remittances are consequences of

migration and they are expected to react to shocks in the receiving country. In the

following section I present a theoretical model of migrant remitting behavior that allows

for a bad state shock on the receiving household.

3. Theoretical Model

The goal of this paper is to derive a hypothesis on the migrant’s remitting behavior. In

this section I present a variant of the model presented in Agarwal and Horowitz (2002).

The model presented in Agarwal and Horowitz (2002) defines the bad state

shock to be migrant specific and therefore originates in the destination country of the

migrant. In this paper I include a bad state shock on the receiving household and

investigate the remitting behavior of migrants towards that shock. The main reason

behind the placement of the bad state shock is that migration and remittances are to a

certain extent related Hoddinott (1994). In this regard, exploring the reaction of

remittances to an income shock in the receiving household might be crucial for

determining the remitting behavior. Moreover, in the theoretical model presented in

Agarwal and Horowitz (2002) migrants expect monetary transfers from the receiving

household in case of a bad state outcome in the host country. The authors model the flow

6
of remittances as a two way stream. In this paper I model remittances as unidirectional

monetary flows with the origin being the migrants and the final destination being the

receiving households.

In effect, the Nicaraguan dataset analyzed in this paper includes 505 families that

have migrants living abroad of which only 16 families send monetary transfers to these

migrants. Out of these 16 families, six families also receive remittances from migrants.

This last number of families is around 1.1% of the number of the families that have

migrants living abroad in the Nicaraguan 2001 survey sample. Table 1 presents the

characteristics of households and migrants by the level of monetary engagement of the

receiving households in the remitting process conditional on having one migrant living

abroad. Comparing households that receive remittances in column (B) to households that

send remittances in column (C), the main difference is in the location of residence.

Households that send remittances tend to reside in urban areas. In addition, differences

include the gender composition and labor force status of the head of the household, the

destination of the migrant, and the relationship of this migrant to the head of the

receiving household. Male and working head of households tend to form the bulk of the

receiving households that send remittances abroad. Moreover, it seems that a migrant’s

move to a developed country requires households in Nicaragua to share the cost of the

move. In fact, receiving households that send remittances represented in columns (C)

and (D) show larger percentages of migrants living in developed countries relative to

those households that receive remittances and those that do not send or receive. For

7
those households with dual remittances flows, column (D), the striking difference is the

location of the residence and the destination of the migrant.

To summarize, the receiving households that participate in sending remittances

have on average notably higher percentages of working head of households, male head

of households and younger head of households. Now focusing on only columns (A) and

(B) I note that there might be a threshold level of households’ characteristics that define

receiving households which do not send or receive versus those that do send monetary

transfers to migrants living abroad. The percentages of working head of the household,

residing in urban areas and male head of household are indeed higher under column (A)

than those in column (B) but still lower than the percentages in column (C). Also the

migrants who belong to households in column (C) tend to be living in developed

countries2.

Additionally the small number of families who engage in two direction

remittances seem to be consistent across low income countries. Agarwal and Horowitz

(2002) report a very similar finding for Guyana (1.4%). For the purpose of this paper I

ignore remittances from receiving households because it seems that across developing

countries the frequency of two-way remittances is relatively small. In the following

subsection I present the theoretical model.

3.1. Pure Altruism

2
The subset of developed countries as a destination for Nicaraguan migrants includes Canada, Greece,
Sweden and United States. The countries that did not make it in this sample are Algeria, Argentina, Brazil,
China, Colombia, Costa Rica, Cuba, El Salvador, Guatemala, Guinea, Haiti, Honduras, Mexico, Panama
and Tunisia. Both samples cover the destination of all the migrants in the Nicaraguan 2001 survey sample.

8
 Based on the previous section, I assume that migrants do not receive monetary

transfers from their original household. This assumption leaves out the specific case of

risk-sharing that the literature has extensively modeled Agarwal and Horowitz (2002)

and Amuedo-Dorantes and Pozo (2006) but it does follow the empirical evidence more

closely. I build a two period model where a migrant who cares about the welfare of the

receiving household has the following utility:

U i = α log C i1 + β log C i 2 + δ log C H (1)

where α is the weight on migrant i ’s consumption in period 1 given by C i1 , β is the

weight on migrant i ’s consumption in period 2 given by C i 2 and δ is the weight on C H ,

the recipient household consumption. The weights on consumption are positive such as

0 < α and 0 < β and 0 ≤ δ . The receiving household consumption depends on high

income YH with probability of π and low income YL with probability of 1 − π , with

YH − YL > 0. The receiving household consumption also depends on the total remittances

received by the household R . The total remittances R can be written as ri + kr−i where

ri is migrant i ’s own remittances and, k is the number of other migrants belonging to

the same receiving household who remit on average r−i . The altruistic migrant chooses ri

to maximize utility subject to

C i1 = Yi1 − ri (2)

C i 2 = Yi 2 (3)

and

9
 C H = πYH + (1 − π )YL + ri + kr−i (4)

where Yi1 is the migrant’s income in the first period and ri is the migrant’s remittances.

The second period migrant’s consumption C i 2 depends on the migrant’s second period

income Yi 2 . The migrant chooses the level of remittances to maximize utility subject to

(2), (3) and (4). The first order conditions (FOC) are:

∂U −α δ
= + =0 (5)
∂r Yi1 − ri πYH + (1 − π )YL + ri + kr−i

Solving for ri from equation (5) I define a remittance function given by:

ri = r (Yi1 ; YH ; YL ; k ; π )
*
(6)

Equation (6) states that remittances sent by migrant i depends on the migrant’s

first period income, the receiving household income, the number of other migrants

belonging to the same receiving household, and the probability of a good state in the

receiving country. Using the implicit function theorem, I derive two hypotheses on

migrants’ remitting behavior3:

∂r − δC H−2 r−i
=− <0 (7)
∂k − αC i−12 − δC H− 2

∂r − δC H−2 (YH − YL )
=− <0 (8)
∂π − αC i−12 − δC H− 2

Both derivations represented in equations (7) and (8) have a negative sign. This

suggests that altruistic migrants’ remittances respond negatively to both the number of

other migrants belonging to the same receiving household and the probability of a good

3
The derivations are in appendix I.

10
state in their original country. As the number of migrants from the same household

increases, the amount of remittances sent by migrant i decreases. Also, as the likelihood

of a good state increases it is more likely for an altruistic migrant to decrease remittances

sent home. This is consistent with the belief that remittances are often thought to be

intended to mitigate the burden of poor economic performance on the receiving

household.

These two hypotheses follow from the altruistic migrant’s utility function where

the consumption of the receiving household directly enters the migrant utility. For self-

interest motivated remitters the utility of the receiving household does not enter the

migrant’s utility function as explained in more detail in the next subsection.

3.2. Self-Interest

In the following I consider the opposite case of pure altruism. For a pure self-interest

migrant the receiving household’s welfare does not enter the utility function and this is

given by δ = 0 . Therefore the utility function of a self-interest motivated remitter is:

U i = α log C i1 + β log C i 2 (9)

This migrant maximizes utility subject to:

C i1 = Yi1 − ri (10)

and

C i 2 = Yi 2 + g (ri ) (11)

where for each dollar remitted migrants receive a return on their investment g (1) where

g ' (ri ) > 0 and g '' (ri ) > 0 . Migrant i again chooses ri to maximize the following utility:

11
 U i = α log(Yi1 − ri ) + β log(Yi 2 + g (ri )) (12)

subject to constraints (9) and (10). The FOC is the following:

∂U −α β g ' (ri )
= + =0 (13)
∂r Yi1 − ri Yi 2 − g (ri )

∂r
From equation (13) and the implicit function theorem it is clear that = 0 , and
∂k

∂r
= 0 which suggests that the number of other migrants in the receiving household and
∂π

the likelihood of a good state have no effect on the amount remitted by a self-interest

motivated migrant. These findings follow from the self-interest migrant utility function

which does not account for the welfare of any member of the receiving household.

Both cases of remittance behavior discussed above give distinct theoretical

predictions that can be empirically tested. In the next section I describe the data and the

estimation method.

4. Data and Estimation Method

4.1. Data

The data set is a national living standards measurement survey (LSMS) administrated in

2001 in Nicaragua. The LSMS was established by the World Bank. This nationally

representative survey includes data on several aspects of the household and includes

4191 families in 4001 households4. The survey comprises a remittances module where a

knowledgeable member of the receiving household in Nicaragua was asked about other

household members living abroad. The remittances module includes a total of 897

4
In some cases more than one family live in one household. For the migrants sample the number of
families is the same as the number of households.

12
migrants who belong to 505 families residing in Nicaragua. I have information on the

migrants’ destination, labor force status, age, gender, education, and years of migration.

I also have information on the receiving household. I know the number of migrants who

belong to the same household, the labor force status, gender, age and education of the

head of the receiving household, as well as the residence of the receiving household.

4.2. Estimation Method

In order to investigate the migrant’s remitting behavior I need to determine the signs of

two relationships: remittances ri and the number of other migrants k and also

remittances ri and the likelihood of a good state π or a bad state 1 − π .

The dependent variable ri is never negative. The level of remittances is zero for a

large number of observations which means that the data on remittances are truncated

since remittances are unobserved for the migrants that do not participate in the remitting

process. In a censored regression model, equation (6) determines both the probability of

remitting and the level of remittances. I consider a remittance equation which has

remittances by Nicaraguan migrants as a function of individual and household

characteristics:

ri = β 0 + β 1 X i + β 2 Z + u i (14)

where X i includes migrants’ individual characteristics, Z refers to the household

( )
characteristics and u i ~ N 0, σ 2 . The migrants and households characteristics enter the

remittances implicit function in equation (6) through the migrants’ and the receiving

households’ income levels. In the Nicaraguan survey data, I do not observe migrants’

13
income. However I know the migrants’ characteristics (age, gender, education,

destination, years living abroad and labor force status) and I use those as a proxy for

income. In equation (6) the migrant’s first period income Yi1 is therefore a function of

migrants’ characteristics given X i by Yi ( X ) . For the receiving household I do observe

the income but for endogeneity reasons I follow the same approach and use the receiving

heads of households’ characteristics Z to proxy for their income level.

Ordinary least squares (OLS) give biased estimates because of the nature of the

dependent variable. The Tobit model uses the same set of covariates to model both the

decision to remit and the amount of remittances. However the coefficients on the

likelihood of remitting and the amount remitted from a Tobit have the same sign.

Following Wooldridge (2003), comparing the results of a standard probit to the Tobit

can be an assessment of the suitability of the Tobit model. For comparison reasons I

show the results of a standard Probit and compare the signs of the statistically significant

coefficients with the signs of the significant coefficients from the Tobit equation.

The Nicaraguan survey data identifies migrants who are remitters but does not

identify the exact amount remitted by those migrants. I know the total supply of

remittances received by a particular receiving household, the number of migrants living

abroad and which of these migrants are remitters and which are not. It seems that this

type of data problem is not uncommon. In fact the same problem exists in the Guyanese

data explored by Agarwal and Horowitz (2002). To overcome this data limitation I

proceed with two different approaches. The first approach is to define what I will refer to

hereafter as the average model. I re-write equation (14) as follows:

14
 rij = β 0 + β 1 X ij + β 2 Z j + u ij (15)

where i refers to a specific migrant belonging to the receiving household j . I

take the average of equation (15) by summing over remitters in household j and

dividing by the number of remitters s j . This leads to the following equation:

sj sj sj
1 1 1 1
sj
∑
i =1
rij = R j = β 0 + β1
sj sj
∑
i =1
X ij + β 2 Z j +
sj
∑u
i =1
ij (16)

where R j is the total supply of remittances to household j . If the number of remitting

migrants s j is either zero or one then the model follows equation (15). Otherwise the

model is defined by equation (16). Note that the coefficients in equations (14), (15) and

(16) are the same which insures the same interpretation of the results. Note that since

sj
1
uij ~ N (0,σ 2 ) then the new error term
sj
∑u
i =1
ij is not homoskedastic with

sj
1
sj
∑u
i =1
ij ~ N (0, σ 2j ) . Therefore, equation (16) defines a heteroskedastic Tobit with a

known form of heteroskdeasticity. In fact:


 1 
)
sj
Var 
 s ∑u ij  s
(
 = Var  1 u1 j + u 2 j + u 3 j + K + u s j
j
(17)
 j i =1   j 

Equation (17) can be rewritten as:

sj
1 1 2
Var (
sj
∑u
i =1
ij )=
sj
σ (1 + ( s j − 1) ρ ) = σ 2j (18)

15
where s is a migrant other than migrant i in household j , cov(u ij ; u ij ) = Var (u ij ) = σ 2 ,

cov(u ij ; u sj )
cov(u ij ; u sj ) = σ j and corr (u ij ; u sj ) = = ρ . The variance of the new
std (u ij ) * std (u sj )

error term is a function of the variance of the original model in equation (8), the number

of remitting migrants within the receiving household and the correlation of the error

terms of different remitting migrants who belong to the same receiving household5.

Finally I estimate the average model using maximum likelihood estimation6.

The second approach is to limit the sample to those migrants belonging to

households with at most one remitting migrant. For each of those migrants I can exactly

identify the amount remitted. I count 387 households in that category which constitutes

around 78% of the 494 receiving households. The new migrant sample is 555 which

represent around 62% of the original 897 migrants. However, there is some concern

regarding selectivity bias. Households with at most one remitting migrant probably share

unobserved characteristics that make them form a non random sample. The selection

issue comes into play in forming the limited sample: households with at most one

remitting migrant. In order to overcome this issue I follow Heckman (1979). The next

section discusses the selection bias problem in more details. In addition, section 5

elaborates more on the data and presents the results of these two approaches.

−1
5
This condition ρ> is necessary when s j ≥ 2 to guarantee a positive variance.
s j −1
6
More details on the likelihood function of the average model are presented in appendix II.

16
5. Results

To explore the remittance behavior of Nicaraguan migrants I need to investigate the

relationship between ri and k , and between ri and π . However before going into the

results I examine the data in more detail. Table 2 examines the characteristics of the

receiving households by number of other migrants. Table 2 searches for any possible

relationship between the number of other migrants and receiving household

characteristics that might play a role in the sign of the coefficient on k . There is no clear

pattern that can be inferred from Table 2. The percentage of head of household working

seems to be decreasing with the number of other migrants but with 3 other migrants in

the household this number picks up again and then with more than 4 other migrants it

decreases again. Note that the larger the number of other migrants is, the smaller is the

sample of households. The other household characteristics do not show any specific

pattern.

In order to capture the probability of a good state versus the probability of a bad

state I define two different measures. The first proxy is a dummy variable that is one if

the head of the receiving household left the last job for a particular set of reasons. In

total, fifteen different answers are listed. The question in the Nicaragua survey is not

very clear about when the head of the receiving household left their last job. Table 3 lists

the reasons and the distribution of households by reason. The list does not follow any

particular order and the reasons are listed as they appear in the survey. The reasons that

the heads of household mention include liquidation of the enterprise, being fired,

retirement plans, end of contract, seasonal work, lack of work, personal duties, school

17
duties, lack of safety at work, harassment in the work place and illness. I presume that

leaving for all of the reasons in Table 3 except for the following reasons: retirement

plan, end of contract and studies (reasons numbered 3, 5 and 12 in Table 3) is a measure

of bad outcome. I exclude these latter reasons from the construction of the bad outcome

measure because they define reasons that could have been expected and therefore the

receiving household could have acted upon ahead of time.

A second measure of the likelihood of a bad state is the length of time that the

head of household has been without work. Out of 494 heads of household 128 have been

looking for a job for at least one day. From Table 4, 101 heads of household out of 128

have been looking for a job for at least one year. I construct a dummy variable for those

households that have been looking for a job for more than one year. I chose the longest

search time (the other choices are days, weeks and months) since a long period of time

better tests the remitting behavior of migrants. It also signals a worse financial situation

for the households relative to the other search periods.

Note that both proxies define two different income levels for the receiving

household. If the head of the household is unemployed or has been looking for a job for

more than a year, then, in either case, the total income level of the receiving household

must be different from the total household income in the opposite situation.

Table 5 presents the characteristics of households by measure of bad state and the

characteristics of those households not affected by a bad state shock. For both measures

the majorities of households are located in urban areas and have a female head of

household. The mean age of the head of the household is around 60 years old. Those

18
households not affected reside in relatively more rural areas than those affected and also

have a majority of male head of households.

Table 6 shows the characteristics of the pool of migrants who originated from

non-affected head of households, from head of households who left their job for one of

the 15 reasons in Table 3 and those head of households who have been looking for a job

for at least one year. Table 6 investigates any differences in migrants’ characteristics that

determine migrants’ income Yi . The only striking difference is the gender composition of

the migrants’ population. More than 50% of the migrant population from unaffected

households is male whereas more than 50% of migrants from affected households are

females.

From the theoretical model in section 3 the characteristics of the head of the

receiving household and of the migrants determine their respective income levels.

Equation (14) includes migrants and household characteristics. Migrant’s individual

characteristics play a major role in the remitting decision. I control for age, level of

schooling, gender, destination, years since migration and employment status of the

migrant. These characteristics affect the migrant’s ability to remit. Moreover, I control

for the head of the household education level, age, gender, the receiving household area

of residence and the number of household nonmigrating members. The main two

covariates in the theoretical model, the number of other migrants and the measure of bad

state are also considered household characteristics.

Before going into the results I investigate the selection bias problem in more

details. Table 7 compares the households and migrants’ characteristics across two

19
different samples: the limited sample, which includes migrants who belong to

households with at most one remitting migrant, and the total migrant sample. All

characteristics between these two samples seem to match suggesting that the limited

sample is a reliable representation of the total migrant population. The only significant

discrepancy is the percentage of migrants living in developed countries. For the limited

sample, the percentage of migrant living in developed countries is 25% while for the

total sample it is around 31%. However, since unobservable factors can affect the

membership to the limited sample I investigate what variables can help determine the

association with this sample.

Table 8 compares the relationship of the migrant to the head of the receiving

household for three samples: limited sample, the remaining migrants not belonging to

the limited sample and total migrant sample. The first column in Table 8 is notably

different from both columns 2 and 3. It seems that migrants forming the limited sample

are more likely to be spouses and parents to the head of the receiving household than the

migrants belonging to the other two samples. The migrants forming the limited sample

are less likely to be the child of the head of the receiving household relative to the other

two migrant samples. I proceed with spouse and parent as the variables defining

membership to the limited sample to correct for selection bias. I do that partly because of

the differences of the percentages in Table 8 and partly because I expect that in the case

of being the spouse or the parent of the head of the receiving household chances are that

there would be at most one remitting migrant. I also include in the selection equation the

labor status, education level, age, gender, destination, years since migration of the

20
migrant and the residence location, education level, age and gender of the head of the

receiving household because these characteristics have an effect the ability to remit7.

Table 9 presents the results of a standard Probit on equation (14). As mentioned

in section 4, I can exactly identify the remitters from the non-remitters and this fact will

identify the dependent variable in the Probit equation. I compare the signs of the

statistically significant coefficients in the Probit equation to the signs of the coefficients

in the main results presented in Tables 10a and 10b. All the statistically significant

coefficients from the Probit equation and from tables 10a and 10b have the same signs. I

turn now to the main results.

Table 10a presents the results of two proxies of good state following the average

model explained in section 4. Table 10b limits the sample to those receiving households

with at most one remitting migrant. In Tables 10a and 10b column (1) refers to a dummy

variable for households where the head had lost the last job for one of the reasons

discussed above and column (2) refers to a dummy variable for those head of households

who have been looking for a job for at least one year. I control for the budget constraint

of the migrant by including age, gender, level of education, labor force status and

destination of migrants which implicitly determine migrants’ income. I also control for

household characteristics as the level of education, the age and gender of the head of the

receiving household and the location of the household.

In the average model the variables of interest for this paper have the sign of the

altruistic migrant model. However the coefficient on k is also significant at the 1%

7
The results of the selection equation (first stage Probit) are in Table A in appendix III.

21
significance level. Nicaraguan migrants decrease the amount remitted with the increase

of migration in the original household that they belong to. The coefficients on 1 − π

match the theoretical predictions of the altruistic model but are not statistically

significant under both proxies. Having a job, being a female and living in a developed

country increase remittances. Being older than 30 seems to positively affect the remitting

decision. The location of the residence of the receiving household also matters.

Table 10b presents the results of a sample selection corrected estimation on

equation (14) limiting the sample to migrants belonging to receiving households with at

most one remitting migrant. Similar results to the average model are found in this sample

of 555 migrants. The signs on k and 1 − π match the theoretical predictions of the

altruistic migrant. Again, only the coefficient on k is statistically significant. The other

covariates also follow the same pattern as the variables in the average model except now

the gender of the head of the household significantly affects remittances.

To summarize, there is some empirical evidence that points to some extent to the

theoretical predictions of the altruistic migrant model developed in section 3. Controlling

for the migrants’ budget constraint and some head of household characteristics, migrants

remit less when the number of other migrants increase and they also remit more in case

of negative income shock in the receiving household. However, Nicaraguan migrants

seem to react more to the number of migrants in their original household in Nicaragua.

In both approaches the coefficient on k is negative and significant. The coefficient on

1−π is positive in all these cases but again not statistically significant.

The labor status, destination and gender of the migrant affect the remitting decision and

22
seem to be robust across all three approaches. The receiving household income level also

seems to affect the remitting decision since the household income level is determined by

the education of the head of the household, the gender of the head of the household and

the location of the residence. All these characteristics affect the remitting decision.

Note that the average model computes the correlation coefficient between the

error terms of the remitting migrants belonging to the same receiving households. The

correlation coefficient ρ is positive, statistically significant and close to 0.63 in value.

This positive value suggests that the remitting decision of migrants belonging to the

same receiving household is positively correlated. Also, from table 10b I calculate the

sample selection parameter λ to be around -0.48 and statistically significant suggesting

that a sample selection bias does exist in building the limited sample.

For policy purposes, Table 11 separates the Tobit coefficients of both variables

of interests from the average approach into two effects: a change in the probability of a

remitting and a percentage change in the amount remitted. One additional migrant

decreases the probability of remittances by no more than 13%. Migrants are 6% more

likely to remit in case of a bad state shock. For the amount percentage changes, migrants

remit 28% less with one additional migrant and they remit between 13% more in

response to a bad income shock.

This finding raises questions concerning the consequences of the trade-off

between migration and per migrant remittances in developing countries. One additional

migrant leaving the labor exporting country decreases per migrant remittances by a

number close to 13%. This negative relationship might have unanticipated effects on the

23
overall impact of migration and remittance on the original country. For instance, the

finding in Adams and Page (2005) that an increase in both international migration and

remittances decrease poverty in developing countries might not hold anymore.

One interesting finding across both approaches is the robustness of the migrant

gender variable. In all equations (including the Probit equations) female migrants seem

to remit more than male migrants. In the Nicaraguan sample female migrants constitute

more than 47% of the total migrants’ population. This gender neutrality makes the

remitting behavior across gender an interesting topic. Following Vanwey (2004) I

further investigate the gender heterogeneity in the migrant behavior. Table 12 repeats the

same estimation approaches while limiting the sample to male and then female migrants.

In all cases the coefficient on the number of other migrants k is negative and significant.

However the coefficient on the bad state measure 1 − π is only positive and significant

for female migrants. The results seem to point out that male migrant do not really

respond to an income shock at the receiving household. However, female migrants

respond to the same income shock and their response falls under the altruistic model

predictions. Table 12 suggests that female migrants have a different remitting behavior.

6. Conclusion

This paper presents a theoretical model of migrants' remitting behavior. I consider two

main motivations towards remitting: altruism and self-interest. This paper contributes to

the remittances literature by investigating the reaction of remittances to a bad state

outcome on the receiving household rather than on the migrant. The remittance literature

has focused on studying the remittance behavior in regards to a bad outcome shock to

24
the migrant which leads to an ex-ante risk-sharing behavior. In this paper migrants do

not expect monetary transfers from the original households. This assumption is

consistent with the data evidence from poor developing countries.

In the theoretical predictions of the model a pure altruistic migrant receives direct

satisfaction from the welfare of the original household. The total supply of remittances

enters the receiving household consumption function and therefore the migrant’s utility

function. On the contrary pure self-interest motivated migrants do not receive

satisfaction from the welfare of the receiving household. The theoretical predictions

suggest that the number of other migrants who belong to the same receiving household

has a negative effect on remittances in the case of altruistically motivated migrants and

no effect at all on the self-interest driven migrants. Also the probability of a good state in

the receiving country which affects the level of income in the receiving household has a

negative effect on remittances for an altruistic migrant and again no effect for a self-

interest motivated migrant.

I test the findings of the theoretical model with data from Nicaragua. I use a 2001

LSMS data and define two proxies for the bad state outcome and find some empirical

evidence supporting altruism as a main motivation behind remittances in Nicaragua.

The results here are in accord with Agarwal and Horowitz (2002). The number of other

migrants belonging to the same household seems to play a crucial role in determining the

remittance behavior. I also test the gender heterogeneity of the remitting behavior and

find supporting evidence that female migrants seem to behave more altruistically than

their male counterparts.

25
 Remittances can be motivated by pure altruism without any economic aspirations

but they can also be self motivated in terms of an implicit contract between the original

household and the migrant which includes for example inheritance plans. In the former

case migrants belonging to the same original household together insure that the original

household is not in financial need and therefore an increase in the number of migrants is

expected to decrease remittances per migrant. In the latter case there is no clear

connection between the number of migrants and remittances since migrants act by self-

interest. From policy perspective and in the case of altruistically motivated remittance, to

maximize remittances per migrant, labor exporting countries can work on incentives for

keeping potential migrants from joining other household members. Therefore sending

countries’ governments can affect remittances per migrant by targeting potential

migrants. These governments need to be aware of the existing trade-off between the

number of migrants belonging to the same receiving household and remittances per

migrant. One potential policy interest is to find the optimal k that maximizes

remittances per migrant.

Finally, researchers such as Hoddinott (1994) model remittances and migration

as a family decision. From that point of view there is some concern regarding the

endogeneity of the number of other migrants. This concern raises questions pertaining to

the choice of instruments and their validity. This forms the next step in research.

26
References

Adams Jr., R., and Page, J. (2005) Do International Migration and Remittances Reduce
Poverty in Developing Countries? World Development, 33(10), 1645-1669

Agarwal, R., and A, Horowitz. (2002) Are International Remittances Altruism or

Insurance? Evidence from Guyana Using Multiple-Migrant Households, World
Development 30(11), 2033-2044

Amuedo-Dorantes, C., and S, Pozo (2006) Remittances as insurance: evidence from

Mexican immigrants, Journal of Population Economics, 19, 227-254

Brown, R., and B, Poirine. (2005) A Model of Migrants’ Remittances with Human
Capital Investment and Intrafamilial Transfers, International Migration Review, 2, 407-
438

Chami, R., T, Cosimano and M, Gapen. (2006) Beware of Emigrants Bearing Gifts:
Optimal Fiscal and Monetary Policy in the Presence of Remittances. IMF- Working
Paper, WP/06/61

Chami, R., C, Fullenkamp and S, Jahjah. (2005) Are Immigrant Remittance Flows a
Source of Capital for Development? IMF-Working Paper, 52(1), 55-81

Connell, J., and R, Brown. (2004) The Remittances of Migrant Tongan and Samoan
Nurses from Australia, Human Resources for Health, 2(2), 1-45

De Haas, H. (2006) Migration, Remittances and Regional Development in Southern

Morocco, Geoforum, 37(4), 565-580

De La Brière, B., E, Sadoulet, A., and S, Lambert. (2002) The Roles of Destination,
Gender, and Household Composition in Explaining Remittances: An Analysis for the
Dominican Sierra, Journal of Development Economics, 68, 309-328

Durand, J., W, Kandel, E, Parrado and D, Massey. (1996) International Migration and
Development in Mexican Communities, Demography, 33(2), 249-264

Heckman, J. (1979) Sample Selection Bias as a Specification Error, Econometrica,

47(1), 153-162

Heilmann, C. (2006) Remittances and the migration-development nexus-Challenges for

the sustainable governance of migration, Ecological Economics, 59(2), 231-236

Hoddinott, J. (1994) A Model of Migration and Remittances Applied to Western Kenya,

Oxford Economic Papers, 46(3), 459-476

27
Lucas, R., and O, Stark. (1985) Motivations to Remit: Evidence from Botswana, Journal
of Political Economy, 93(5), 901-918

Orozco, M. (2002) Globalization and Migration: The Impact of Family Remittances in

Latin America 44(2), 41-66

Vanwey, L. (2004) Altruistic and Contractual Remittances between Male and Female
Migrants and Households in Rural Thailand, Demography 41(4), 739-756

Widgren, J., and P, Martin. (2002) Managing Migration: The Role of Economic
Instruments, International Migration, 40(5), 213-229

Wooldridge, J. (2003) Introductory Econometrics: A Modern Approach. 2nd edition,

Thomson South Western

World Bank www.worldbank.org

World Bank’s Global Economic Prospects Accessed August 2007

www.worldbank.org/prospects/gep2005

28
Table 1. Characteristics of Receiving Households and Migrants by Remitting
Process
Households Households Households Households
that Do Not that Receive that Send that Send
Receive Nor Remittances Remittances and Receive
Send Remittances
Remittances
(A) (B) (C) (D)
Receiving Households
Percent Residing in Urban
73.3 71.8 81.2 100.0
Areas
Percentage Head of
58.3 49.1 60.0 66.6
Household Male
Percent Head of Household
75.0 57.6 86.6 100.0
Working
Mean Age Head of
51.6 54.5 50.6 48.5
Household
Mean Years of Education
3.2 2.8 2.8 3.3
of Head of Household

Sample 180 309 16 6

Migrants
Mean Migrant Age 28.0 30.3 29.5 33.1
Mean Migrant Education 6.9 4.5 4.5 4.8
Mean Years of Migration 5.7 6.7 7.4 9.0
Percent Residing in
20.0 36.3 48.6 66.6
Developed Countries
Percent Working 62.3 78.5 75.6 94.4
Percent Male 54.2 52.8 51.3 50.0

Sample 260 600 37 18

Note: 1- All the households in this table have at least one migrant living abroad. 2- Developed Countries
include Canada, Greece, Sweden and United States.

29
Table 2. Characteristics of Receiving Households by Number of Other Migrants k
Percentage Percentage Percentage Mean Years
Mean Age
Residing in Working Head of Education
Head of
k Urban Head of Household Head of
Household
Areas Household Male Household Sample
0 75.3 68.3 49.6 52.7 2.9
300
1 64.7 62.8 58.1 52.4 2.7
105
2 63.4 56.1 58.5 57.8 2.4
41
3 90.9 59.0 59.0 56.8 2.9
22
4 or more 69.2 46.1 50.0 54.9 2.2
26
All 72.4 64.5 52.6 53.4 2.8
494

30
Table 3. Distribution of Households by Reason of Head of the Household Leaving
the Last Job
Reasons Percentage Count
1- The enterprise was liquidated 1.8 9
2- You were dismissed 0.6 3
3- Retirement Plan 0.2 1
4- By age 3.6 18
5- End of the contract 1.6 8
6- Agricultural cycle/seasonal work ended 0.2 1
7- You are pensioned off 2.4 12
8- You earned not much money 2.0 10
9- You did not like your job 0.6 3
10- Not much work 0.0 0
11- Family/home duties 4.6 23
12- Studies 0.0 0
13- Insufficient industrial safety 0.4 2
14- Improper treatment or psychological
6.6 33
pressures
15- Illness 1.0 5
Sample 25.6 128

31
Table 4. Distribution of Households by Length of Job Search
Time Spent looking for a Job Percentage Count

Days 0.7 1
Weeks 0.7 1
Months 19.5 25
Years 78.9 101

Sample 25.6 128

32
Table 5. Characteristics of Receiving Households by Measures of Bad State versus Unaffected
Households
Percentage Percentage Head Mean Years of
Measure of Bad State Mean Age Head
Residing in of Household Education Head
of Household
Urban Areas Male of Household
Left Last Job 82.8 38.2 60.1 2.6
(Sample: 128)

More than 1 Year looking

for a Job 84.1 31.6 61.8 2.5
(Sample: 101)

Not Affected 68.8 57.6 51.0 2.8

(Sample: 366)

33
Table 6. Migrants’ Characteristics by Measures of Bad State versus Unaffected
Households
Head of
Head of
Head of Household
Household Not
Household Left More than 1
Affected
Last Job Year looking
Characteristics for a Job
Male 56.3 44.58 44.1

Working 74.6 73.9 76.9

Residing in a Developed Country 30.9 33.7 35.3

Mean Age 28.9 30.1 31.3

Mean Education 3.6 3.9 3.9

Sample 623 249 195
Note: 1- Male, Working and Residing in a Developed Country are percentages. 2- Developed Country
destination includes the United States, Canada, Greece and Sweden.

34
Table 7. Characteristics of Receiving Households and Migrants for Households
with at most One Remitting Migrant (Limited Sample) versus Full Migrant Sample

Households with at Most Full Migrant Sample

One Remitting Migrant

Receiving Households
Percent Residing in Urban
0.74 0.72
Areas
Percentage Head of
0.51 0.52
Household Male
Percent Head of Household
0.67 0.64
Working
Mean Age Head of
52.8 53.4
Household
Mean Years of Education
2.8 2.6
of Head of Household

Sample 387 494

Migrants
Mean Migrant Age 28.5 29.3
Mean Migrant Education 3.4 3.7
Mean Years of Migration 5.7 6.0
Percent Residing in
0.25 0.31
Developed Countries
Percent Working 0.70 0.74
Percent Male 0.52 0.53

Sample 555 872

Note: 1- All the households in this table have at least one migrant living abroad. 2- Developed Countries
include Canada, Greece, Sweden and United States.

35
Table 8. Relationship of the Migrant to the Head of the Receiving Household for
Households with at most One Remitting Migrant, Full Migrant Sample and the
Remaining Sample
Not in Limited
Limited Sample Full Sample
Sample
Relationship of the Migrant
to the Head of the
Receiving Household
Percentage if Spouse 5.9 2.5 4.7
Percentage if Parent 3.4 1.8 2.8
Percentage if Child 55.6 65.2 59.1

Sample 555 317 872

Note: 1- All the households in this table have at least one migrant living abroad. 2- Developed Countries
include Canada, Greece, Sweden and United States.

36
Table 9. Probit Estimates for Equation (14): All Migrants
Amount Remitted
Variables (1) (2)
Intercept -0.5137** -0.4993**
(0.2199) (0.2188)
Number of other Migrants = k -0.0443* -0.0434*
(0.0248) (0.0247)
Bad State Measure = 1 − π 0.1726 0.1041
(0.1120) (0.1194)
1 if Working 1.0687*** 1.0648***
(0.1149) (0.1151)
1 if Education less than 4 Years -0.206** -0.2062**
(0.1022) (0.1020)
1 if Male -0.1807* -0.1834*
(0.0953) (0.0954)
1 if Age greater than 29 0.2805*** 0.2805***
(0.1085) (0.0954)
1 if Destination is Developed Country 0.4598*** 0.4572***
(0.1160) (0.1157)
1 if Years since Migration greater than 5 -0.0456 -0.0469
(0.1108) (0.1106)
1 if Urban Residence -0.2863** -0.2806**
(0.1123) (0.1122)
1 if Education of HHH less than 4 -0.3084** -0.3011**
(0.1216) (0.1212)
1 if HHH Male -0.0221 -0.03164
(0.0957) (0.0954)
1 if HHH age is greater than 64 0.0696 0.0910
(0.1060) (0.1067)
Number of Nonmigrants 0.0436** 0.0434**
(0.0172) (0.0171)
Log Likelihood -506.38 -507.17
Sample 872 872
Note: 1- Columns refer to three different measures for the good state probability: column (1) refers to a
dummy variable for households where the head had lost the last job for one of the reasons discussed in
table 3. Column (2) refers to a dummy variable for those head of households who have been looking for a
job for at least one year. 2- HHH refers to head of the receiving household. 3-*** means significant at the
1 percent level; ** at the 5 percent level; * at the 10 percent level.

37
Table 10a: Tobit Estimates for Equation (14) following the Average Model: All
Migrants
Amount Remitted
Variables (1) (2)
Intercept -0.9169* -0.8821
(0.5610) (0.5586)
Number of other Migrants = k -0.8700*** -0.8713***
(0.1071) (0.1078)
Bad State Measure = 1 − π 0.3896 0.3695
(0.2887) (0.3143)
1 if Working 2.5565*** 2.5498***
(0.3339) (0.3344)
1 if Education less than 4 Years -0.2452 -0.2519
(0.2706) (0.2699)
1 if Male -0.8370*** -0.8404***
(0.2448) (0.2497)
1 if Age greater than 29 0.8398*** 0.8379***
(0.2778) (0.2780)
1 if Destination is Developed Country 1.1550*** 1.1574***
(0.2871) (0.2876)
1 if Years since Migration greater than 5 -0.1876 -0.1898
(0.2703) (0.2702)
1 if Urban Residence -0.4500* -0.4528*
(0.2801) (0.2806)
1 if Education of HHH less than 4 -0.2080 -0.1915
(0.2687) (0.2673)
1 if HHH Male -0.3634 -0.3650
(0.2376) (0.2399)
1 if HHH age is greater than 64 -0.1087 -0.1030
(0.2939) (0.2960)
Number of Nonmigrants 0.0508 0.0482
(0.0407) (0.0407)

Log Likelihood -641.29 -641.49

Theta = θ 0.3899*** 0.3898***
(0.0151) (0.0151)
Rho = ρ 0.6398*** 0.6373***
(0.2109) (0.2116)
Sample 708 708
Note: 1- Columns refer to three different measures for the good state probability: column (1) refers to a
dummy variable for households where the head had lost the last job for one of the reasons discussed in
table 3. Column (2) refers to a dummy variable for those head of households who have been looking for a
job for at least one year. 2-*** means significant at the 1 percent level; ** at the 5 percent level; * at the
10 percent level. 3- Robust standard errors are in parentheses.

38
Table 10b: Sample Selection Estimates for Equation (14): Households with at Most
One Remitting Migrant
Amount Remitted
Variables (1) (2)
Intercept 1.1499*** 1.1617***
(0.3101) (0.3109)
Number of other Migrants = k -0.2568*** -0.2540***
(0.0378) (0.0388)
Bad State Measure = 1 − π 0.2153 0.1435
(0.1721) (0.1912)
1 if Working 0.7901*** 0.7926***
(0.1419) (0.1415)
1 if Education less than 4 Years -0.1472 -0.1446
(0.1484) (0.1485)
1 if Male -0.2828** -0.2866**
(0.1241) (-0.1246)
1 if Age greater than 29 0.3843** 0.3808**
(0.1533) (0.1537)
1 if Destination is Developed Country 0.6309*** 0.6273***
(0.1849) (0.1858)
1 if Years since Migration greater than 5 -0.0510 -0.0522
(0.1522) (0.1526)
1 if Urban Residence -0.0918 -0.0776
(0.1617) (0.1605)
1 if Education of HHH less than 4 -0.1198 -0.1009
(0.1866) (0.1868)
1 if HHH Male -0.2720** -0.2898**
(0.1345) (0.1347)
1 if HHH age is greater than 64 0.1323 0.1458
(0.1784) (0.1861)
Number of Nonmigrants 0.0173 0.0154
(0.0244) (0.0243)

Log Likelihood -1516.47 -1517.04

Sigma = σ 1.4529 1.4544

Lambda = λ -0.4866*** -0.4871***

Sample 555 555
Note: 1- Columns refer to three different measures for the good state probability: column (1) refers to a
dummy variable for households where the head had lost the last job for one of the reasons discussed in
table 3. Column (2) refers to a dummy variable for those head of households who have been looking for a
job for at least one year. 2-*** means significant at the 1 percent level; ** at the 5 percent level; * at the
10 percent level. 3- Robust standard errors are in parentheses.

39
Table 11. Summary of The Change in Amount of Remittances and Change in Probability of
Remitting Results for columns (1) in Table 10a
Percentage Change in Probability Percentage Change in Amount
Variables
Number of other
-13.39 -28.95
Migrants = k
Bad State Measure =
5.99 12.97
1−π

40
Table 12. Estimates for Equation (14) with Different Specifications: Male versus Female
Average Model Limited Sample
Male Female Male Female

Number of other -1.8871*** -1.1680*** -0.2868*** -0.2178***

Migrants = k (0.3846) (0.2200) (0.0462) (0.0589)

Bad State 0.2501 0.7544** -0.1093 0.6097**

Measure = 1 − π (0.3844) (0.3708) (0.2160) (0.2611)

Likelihood -376.39 -363.94 -787.17 -713.53

Sample 400 370 290 265
Note: 1- The bad state measure is the first proxy used under column (1) in Tables 10. The same results are found using the
second measure of the bad state but they are not reported here. 2-*** means significant at the 1 percent level; ** at the 5
percent level; * at the 10 percent level. 3- Robust standard errors are in parentheses. 4- All the equations in this table include
the same set of covariates in tables 8a through 8c.

41
Appendix I:

Derivations of equations (6) and (7):

From the FOC equation (5) in the altruism model

∂U −α δ
= + =0
∂r Yi1 − ri πYH + (1 − π )YL + ri + kr−i

and the implicit function theorem I can write:

∂FOC
∂r − δC H− 2 r−i
= − ∂k = − p0
∂k ∂FOC − αC i−12 − δC H− 2
∂r

∂FOC
∂r − δC H− 2 (YH − YL )
= − ∂π = − p0
∂π ∂FOC − αC i−12 − δC H− 2
∂r

where YH − YL f 0 .

∂r ∂r
For the self-interest model δ = 0 which leads to = 0 and = 0.
∂k ∂π

Solving for ri* in equation (5):

−α δ
Equation (5) gives + = 0 which leads to
Yi1 − ri πYH + (1 − π )YL + ri + kr−i

απYH + α (1 − π )YL + αri + αkr−i = δYi1 − δri

and therefore I can write (α + δ )ri = δYi1 − απYH − α (1 − π )YL − αkr−i

and then after rearranging some terms I get to the following:

δ α α α
*
ri = Yi1 − π (YH − YL ) − YL − kr−i = r (Yi1 ; YH ; YL ; k ; π )
(α + δ ) (α + δ ) (α + δ ) (α + δ )

42
The utility function is strictly quasi-concave which insures the uniqueness of the solution

ri* .

Appendix II:

kj

The likelihood function L j = ∑ ln Lij for the average model is the following where s j is
i =1

the number of remitting migrants in household j :

[ ( )] [
ln Lij = ln 1 − Φ X 'γ = ln 1 − Φ X ' β * θ ( )] if s j = 0 (12)

[ ( ) (
ln Lij = 0.5 * ln θ 2 − θRij − X 'γ )]
2
if s j = 1 (13)

  θ2  1 2
ln Lij = 0.5 * ln − ( )
θRij − X ' γ  if s j > 1 (14)
  h j + ρ (1 − h j )  h j + ρ (1 − h j )
  

β 1
where Φ(.) is the standard normal cumulative distribution function and γ = ;θ = .
σ σ

The likelihood function for the third case ( s j > 1 ) is derived from the likelihood function

β 1
; σ j = σ (h j + ρ (1 − h j )) and
0.5
of the second case ( s j = 1 ) with γ j = ; θj =
σj σj

1
hj = . I maximize L j with respect to γ ;θ and ρ .
sj

43
Appendix III:

Table A: First Stage Probit Estimates for the Sample Selection Estimates on
Equation (14) in Table 10b
Amount Remitted
Variables
Intercept
0.6395***
(0.1864)
1 if Parent 0.4331
(0.3023)
1 if Spouse 0.6224
(0.2434)
1 if Working -0.2575
(0.1119)
1 if Education less than 4 Years 0.1249
(0.1032)
1 if Male -0.0177
(0.0926)
1 if Age greater than 29 -0.1553
(0.1126)
1 if Destination is Developed Country -0.4801***
(0.1095)
1 if Years since Migration greater than 5 0.0372
(0.1070)
1 if Urban Residence 0.2512
(0.1072)
1 if Education of HHH less than 4 0.0101
(0.1149)
1 if HHH Male -0.1876
(0.0931)
1 if HHH age is greater than 64 -0.1286
(0.1066)
-540.00
Log Likelihood
872
Sample
Note: 1- Columns refer to three different measures for the good state probability: column (1) refers to a
dummy variable for households where the head had lost the last job for one of the reasons discussed in
table 3. Column (2) refers to a dummy variable for those head of households who have been looking for a
job for at least one year. 2-*** means significant at the 1 percent level; ** at the 5 percent level; * at the
10 percent level. 3- Standard errors are in parentheses.

44