Content uploaded by Gabriel Montes-Rojas

Author content

All content in this area was uploaded by Gabriel Montes-Rojas on Jul 21, 2015

Content may be subject to copyright.

Migration, Remittances and Capital Accumulation:

Evidence from Rural Mexico

Vera Chiodi

Esteban Jaimovich

Gabriel Montes-Rojas

No. 140

February 2010

www.carloalberto.org/working_papers

© 2010 by Vera Chiodi, Esteban Jaimovich and Gabriel Montes-Rojas. Any opinions expressed here

are those of the authors and not those of the Collegio Carlo Alberto.

Migration, Remittances and Capital Accumulation:

Evidence from Rural Mexico

Vera Chiodi∗Esteban Jaimovich†Gabriel Montes-Rojas‡

This version: February 2010

Abstract

This paper studies the link between migration, remittances and productive assets

accumulation for a panel of poor rural households in Mexico over the period 1997-

2006. In a context of ﬁnancial markets imperfections, migration may act as a

substitute for imperfect credit and insurance provision (through remittances from

migrants) and, thus, exert a positive eﬀect on investment. However, it may well be

the case that remittances are channelled towards increasing consumption and leisure

goods. Exploiting within family variation and an instrumental variable strategy, we

show that migration indeed accelerates productive assets accumulation. Moreover,

when we look at the eﬀect of migration on consumption of non-productive assets

(durable goods), we ﬁnd instead a negative eﬀect. Our results then suggest that

poor rural families resort to migration as a way to mitigate constraints that prevent

them from investing in productive assets.

JEL Classiﬁcation: O15, D31, J24, R23, F22.

Key Words: Migration; Remittances; Capital Accumulation; Rural Poverty.

∗Paris School of Economics (PSE) and Poverty Action Lab (J-PAL Europe), 48 Boulevard Jourdan,

75014 Paris, France. chiodi@pse.ens.fr

†Collegio Carlo Alberto, Via Real Collegio 30, Moncalieri (TO) 10024, Italy. este-

ban.jaimovich@carloalberto.org

‡Corresponding Author: City University London, Department of Economics, D306 Social Sci-

ences Building, Northampton Square, London EC1V 0HB, United Kingdom. Email: Gabriel.Montes-

Rojas.1@city.ac.uk Tel: +44 (0)20 7040 8919

1

1 Introduction

The migration of labor out of agriculture has represented a fundamental issue in the early

models of development economics (Lewis, 1954; Sen 1966; Harris and Todaro, 1970; see

Ghatak, Levine and Wheatly Price, 1996, for an excellent survey). In these models, the

agricultural sector is typically characterized by stagnation and under-productive use of

labor, while the urban industrial sector is viewed as the one that contributes most to

economic development and modernization. The above literature has then seen migration

from the rural to the urban sector basically as a road out of backwardness and poverty,

which are intrinsically linked to agricultural production.1

However, recent work has argued that rural migration may also exert a positive eﬀect

on the rural sector itself. The general argument is that migration and remittances may

contribute to alleviate ﬁnancial and productive constraints in the rural sector.2More

speciﬁcally, Stark (1991) sustains that migrants may play the role of ﬁnancial interme-

diaries, enabling rural households to overcome credit constraints and missing insurance

markets. Furthermore, migration may mitigate the impact of agricultural income shocks

by allowing families to relocate labor to the cities when that is needed (Lucas and Stark,

1985). In fact, it is often the case that individuals in a household commonly pool resources

to ﬁnance migration of one of their members who later on repays by remitting a part of

his/her income back to the family. Thus, households tend to optimally spread their labor

force over diﬀerent geographic markets in order to better pool risks. Adams (1998) studies

the eﬀect of remittances in rural Pakistan and found that they help to increase investment

in rural assets by raising the marginal propensity to invest for migrant households.

Our paper contributes to this latter stand of literature by assessing the eﬀect of mi-

gration on the process of asset accumulation using household data from poor rural areas

1For some more recent models in this vein see Banerjee and Newman (1998) and Lloyd-Ellis and

Bernhardt (2000).

2See, for example, Stark and Levhari (1982), and Rozelle, Taylor and DeBrauw (1999, 2003).

2

in Mexico. We propose a diﬀerentiation by the types of assets that are accumulated by

households, i.e. productive vs. non-productive assets, and we especially focus on the role

played by remittances in helping the accumulation of productive assets.

In that regard, rural Mexico represents a promising setup where to study the link be-

tween migration, remittances and assets accumulation, owing to the very high incidence

of deprivation in terms of access to formal ﬁnancial markets and perceived well-being. In

such a context, migration may exert a positive eﬀect on asset accumulation and, thus,

help lift families permanently out of poverty. Additionally, asset accumulation may also

represent an important component for consumption smoothing as shown by Rosenzweig

and Wolpin (1993). Using a unique panel database for Mexican rural households, the

econometric results presented in this paper show that migration and remittances indeed

open up a possibility for poor households to accelerate assets accumulation, particularly

in productive assets. The fact that rural households use remittances to increase the accu-

mulation of productive assets represents an important and, at the same time, not obvious

result. More precisely, it may well be the case that remittances are channelled instead to-

wards increasing consumption and leisure goods, which may increase households’ current

well-being, but will not help to improve their dynamic prospects. These conclusions are

similar to those of Adams (1998) for Pakistan.

Several endogeneity issues need to be addressed in order to avoid potential biases.

First, households may respond to adverse or positive shocks by changing the number of

migrants or the nature of migration (temporal vs. permanent). Second, selection bias may

occur if migrant households are intrinsically diﬀerent from non-migrant ones. Following

previous work on this subject (see Acosta, 2006, and McKenzie and Sasin, 2007) we

deploy an instrumental variable strategy in order to cope with endogeneity issues based on

migration networks. Because we have a panel data, we can include this variable together

with the ﬁxed eﬀects at the household level. Therefore, our identiﬁcation strategy relies

3

on variation in aggregate migration across time and space.

We frame the empirical results within a two-period model of investment and migration

decisions of credit constrained rural households. The model shows that migration aﬀects

investment only for moderately poor households, while it leads to increasing consumption

for the very poor and relatively rich households. Moreover, the model makes it explicit

that household characteristics need to be properly controlled for in the empirical setting

in order to obtain unbiased results of the eﬀect of migration on investment. More pre-

cisely, the model shows that migration decisions correlate with certain household-speciﬁc

characteristics that also inﬂuence migration and investment choices. Controlling for these

household-speciﬁc characteristics is then crucial for insulating the eﬀect of sending out a

migrant on the investment behavior of those who remain at the rural village. Interest-

ingly, the predicted bias in the theoretical model goes in the same direction as that in our

empirical results.

Migration and remittances have been largely studied in the microeconometric liter-

ature with respect to the accumulation of human capital. As argued in Hanson and

Woodruﬀ (2003) the additional income from remittances may allow children to delay en-

tering the work force. Yang (2008) also ﬁnds a positive eﬀect of remittances on child

schooling and educational expenditure in Philippines using exchange rate shocks as a

source of exogenous variation for remittances. However, it has been argued as well that

migration may alter the family structure, raising child-rearing responsibilities and, there-

fore, having negative consequences on household welfare. Moreover, Acosta (2006) sus-

tains that it may well be expected that recipient families will expand their consumption

of leisure (and reduce labor supply) and increase their dependence on external transfers

accordingly.

The topic addressed here is also related to the eﬀect of credit constraints in the urban

informal sector. Woodruﬀ and Zenteno (2007) found a positive impact of remittances in

4

Mexico (they are shown to be responsible for almost 20% of the capital invested). In

the same vein, Mesnard and Ravaillon (2002) and Mesnard (2004) studied the temporary

migration decision of workers who are credit constrained in Tunisia and evaluates the

extent to which liquidity constraints aﬀect self-employment decisions of returned migrants.

There is also some evidence on this issue for the case of internal migration in India

(Banerjee and Bucci, 1994). Our paper extends these results to rural poor households.

Finally, the eﬀects of remittances on capital accumulation has also been studied at

the macroeconomic level by Glytsos (1993) and Giuliano and Ruiz-Arranz (2009) who

provide evidence that remittances tend to particularly foster growth in countries with less

developed ﬁnancial systems by helping them overcome liquidity constraints. Their results

are thus consistent with ours based on household-level data.

The rest of paper is organized as follows. Section 2 presents a model that accounts

for migration and investment decisions. Section 3 describes the unique dataset used to

construct the panel of rural households. Section 4 presents the methodology used for

constructing the asset indexes. Section 5 presents some descriptive statistics. Section 6

carries the econometric analysis showing the eﬀect of migration on asset accumulation.

Section 7 concludes.

5

2 Migration and investment decisions in a two-period

maximization problem

This section proposes a very simple model to illustrate how relatively poor families may

resort to migration as a response to credit constraints that prevent them from investing in

productive assets. In particular, the model aims at showing that poor families may, under

certain conditions, choose to send migrants so as to use their remittances to overcome

credit constraints.

We will ﬁrst start with a two-period model in which the possibility of sending migrants

is excluded. This will set a benchmark upon which we can then compare the optimal

behavior of families when they do have the opportunity to send a migrant to a richer

region or city, and receive positive remittances from the migrant.

2.1 No-migration regime

There is a continuum of rural families (or households) i∈ I who live for two periods,

t={1,2}.At the beginning of each period teach family ireceives an amount of income

equal to yt,i, where yt,i is the realization of a random variable uniformly and indepen-

dently distributed across families along the interval [1, y], where y > 1. We assume that

y1,i =y2,i =yi; that is, income realizations are persistent within families. More broadly

speaking, we could also interpret the variable yias capturing the eﬀect of family speciﬁc

productive assets (for example, diﬀerent families may own plots of land that diﬀer in

terms of their level of fertility); in the econometric terminology used below, the variable

yicaptures family-speciﬁc ﬁxed-eﬀects.

Families derive log-utility from consumption at the end of each period tand we assume

no discount factor is applied on future consumption.3All families are credit-constrained,

3No future discounting is just a simplifying assumption, useful for the algebraic derivations but without

6

and then, they cannot increase current consumption by borrowing against future income.

Families, however, have access to a storing technology (with no depreciation), hence they

may transfer present income to the future in case they wish so.

All families have access also to an indivisible investment project (an investment in

productive assets that increases productivity in the future, for example, investing in irri-

gation or buying a new tractor). In particular, in period 1 families can choose whether or

not to invest in a project that requires 1 unit of capital as investment, and yields R > 1

units of income at the end of period 2.

The families’ optimization problem may be approached by noting that it involves two

diﬀerent issues: ﬁrst, choosing whether or not to invest in the project at the beginning

of t= 1; second, choosing the optimal consumption ﬂow, conditional on the former

investment decision. We can then solve the problem for family isimply by comparing

the maximum utility achieved in each of the two possible scenarios: (a) investing in the

project; (b) not investing in it. We denote by ct,i consumption in period tand by s1,i the

amount of income stored from period 1 until period 2.

Case (a): Invest in the project. Family isolves:

max : Ui,I = ln(c1,i) + ln(c2,i ) (1)

subject to: c1,i =yi−s1,i −1,

c2,i =yi+s1,i +R,

s1,i ≥0.

It is straightforward to observe that in problem (1) the constraint s1,i ≥0 will bind

in the optimum (i.e., families would like to borrow against future income so as to smooth

any important implication. The log-utility is also assumed mainly for algebraic simplicity (in particular,

it allows us to obtain a closed-form solution for the model), and could be replaced by a general CRRA

utility function without changing the main insights of the model (as we will see below, it is important

though that utility displays decreasing absolute risk aversion).

7

consumption, but they are not able to do so). Hence, families will set optimally s∗

1,i = 0,

implying that: c∗

1,i,I =yi−1 and c∗

2,i,I =yi+R. As a result, the maximum utility achieved

by a family with income yithat invests in the project is given by:

U∗

i,I = ln (yi−1) + ln (yi+R).(2)

Case (b): No investment. Family isolves:

max : Ui,NI = ln(c1,i) + ln(c2,i) (3)

subject to: c1,i =yi−s1,i,

c2,i =yi+s1,i,

s1,i ≥0.

Since the income ﬂow is identical in both periods and future is not discounted, fam-

ilies will optimally consume the yiin each of the two periods, so as to achieve perfect

consumption smoothing. That is, c∗

1,i,NI =c∗

2,i,NI =yi, which in turn implies s∗

1,i,NI = 0.

Hence, the utility achieved by a family with income yithat decides not to invest is given

by:

U∗

i,NI = ln y2

i.(4)

Finally, families will choose to invest if and only if that allows them to obtain higher

intertemporal utility than not investing. Henceforth, we let I= 1 (I= 0) denote the

choice to invest in the productive asset (not to invest in it) in t= 1. Then, comparing

(2) and (4) implies:

I= 1 ⇔yi>R

R−1.(5)

The expression (5) stipulates that only families with (permanent) income larger than

R/(R−1) will invest in the project. The reason for this is that, in the presence of credit

8

constraints, given that utility displays decreasing absolute risk aversion, only suﬃciently

rich families are willing to give away one unit of consumption in t= 1 in order to be

able to invest and increase consumption t= 2 by Runits.4Henceforth, we assume that

y > R/(R−1), so that there exist some families who are willing to invest.

2.2 Migration allowed

Assume now that after observing the income realization yiat the beginning of t= 1, family

icould choose whether or not to send one of their members to a richer city or region in

the ﬁrst period. Sending a migrant imposes an “emotional” cost M > 0, measured in

terms of utility.5Migration is treated as a risky asset when compared with the risk-free

income in the village. The migrant may get a good job in the region he migrated to, which

yields net income υ, where 1 ≤υ < 1 + R. Instead, if migrant fails to ﬁnd a good job, he

receives net income equal to 0.6

We assume that local networks in the city where migrants move to make it easier for

them to obtain a good job.7In particular, we postulate that the migrant from family

iwill manage to ﬁnd good job with probability p(ni) = ni, where ni∈[0,1] represents

the ’network density’ that family ihas got in the recipient city. We assume that niis

uniformly distributed along the interval [0,1] in the population, and that the correlation

between niand yiin the population equals zero.

4Strictly speaking, there is no risk. Hence, the DARA property should be simply understood as an

assumption on the degree of concavity of the utility function, which in turn governs the intertemporal

elasticity of substitution, and therefore how willing agents are to transfer resources across the two periods.

5In the literature this is known as “psychological costs”, and there exists some evidence for intra-

European migration (Molle and van Mourik, 1988). We could also add to the model some pecuniary cost

attached to sending a migrant (i.e. transportation costs), although it is important for our argument that

the expected pecuniary return from sending a migrant is positive.

6The lower bound, υ≥1, essentially says that the good jobs are suﬃciently productive, making

migration (possibly) an attractive option. The upper bound, υ < 1 + R, is just posed to focus only on

those cases in which the credit constraint, si≥0, binds in the optimum (as we will see later on, υ < 1 + R

implies that total family income in t= 1 never exceeds that of t= 2).

7The role of networks on migration has been extensively studied in the literature (see for instance

Munshi, 2003, and the references therein).

9

We denote by e

U∗

ithe utility achieved by family iif they choose to send a migrant

(whereas, as before, U∗

idenotes the utility of family if they do not send a migrant).

Relatively rich families: Consider family iwith network density ni∈[0,1] and income

yi≥R/ (R−1). From the previous analysis, it follows that this family will always invest

in the project. That is, it will invest regardless of whether it chooses to send a migrant

or not, and, in the case they do send a migrant, regardless of whether the migrant ﬁnds

a good job or not. As a result, if they do not send a migrant, their utility equals that

written before in (2). On the other hand, if they do send a migrant, their utility is given

by:

e

U∗,rich

i,I =ni[ln (yi+υ−1) + ln(yi+R)] + (1 −ni) [ln (yi−1) + ln(yi+R)] −M. (6)

A family with yi≥R/ (R−1) will thus send a migrant if and only if e

U∗,rich

i,I > U∗

i,I , which

in turn leads to:

If yi≥R/ (R−1) , send migrant iﬀ: ni[ln (yi+υ−1) −ln (yi−1)] ≥M. (7)

Relatively poor families: Consider now the case of family iwith ni∈[0,1] and

yi< R/ (R−1). From the previous analysis, it follows that such a family will not invest

in the project if, after sending a migrant, this migrant fails to obtain a good job. Nor will

they invest in the project when they do not send a migrant, as this situation is isomorphic

to the no-migration regime.

The ﬁrst question to address is then the following: should a family that sent a migrant

invest in the project when the migrant obtains a good job? Consider such a family: the

two expressions below show the utility achieved by the family, ﬁrst, in the case it invests

10

in the project and, second, in the case it does not.

e

U∗,poor

i,I =ni[ln (y+υ−1) + ln(y+R)] + (1 −ni)ln y2

i−M, (8)

e

U∗,poor

i,NI =nihln yi+υ

22i+ (1 −ni)ln y2

i−M. (9)

Hence, comparing (8) and (9), it follows that families with yi< R/ (R−1) who send a

migrant will invest in the project, if and only if the migrant ﬁnds a good job and the

following condition holds:

yi>R

R−1−υR−υ

4

R−1≡by. (10)

Notice that by < R

R−1. In fact, it may well be that by < 1.8

The second question to deal with is, bearing in mind equations (8) and (9), should a

family with ni∈[0,1] and yi< R/ (R−1) send a migrant or not? Answering this question

demands comparing U∗

i,NI to e

U∗,poor

i,I for those families with yi∈by, R

R−1, whereas for those

families whose yi≤bywe must compare U∗

i,NI to e

U∗,poor

i,NI . We can thus obtain the following

two conditions:

If yi∈by, R

R−1, send migrant iﬀ: ni[ln (yi+υ−1) + ln(yi+R)−ln (y2

i)] ≥M;

(11)

If yi<by, send migrant iﬀ: nihln yi+υ

22−ln (y2

i)i≥M. (12)

Since a larger network, ni, increases the chances the migrant ﬁnds a good job (or,

in other words, the expected return from sending a migrant increases with ni), families

with a larger niwill naturally tend to be more prone to send a migrant. The following

8More precisely, by < 1 whenever R≥υ−1+υ

4.Notice, too, that both a larger Rand larger υmake

this last inequality more likely to hold. This is quite intuitive, since the (expected) return from migration

is increasing in Rand υ; in the former case indirectly through investment returns, in the latter directly

through earnings.

11

proposition states this result more formally.

Proposition 1 There exists a continuous and strictly increasing function en(y) : R++ →

R++, such that for all ni≥en(yi) :

(i) If yi∈R

R−1, y, then condition (7) holds.

(ii) If yi≥1and yi∈by, R

R−1, then condition (11) holds.

(iii) If yi≥1and yi≤by, then condition (12) holds.

Furthermore, if M≤ln(R), then for y=R

R−1, we have that 0<enR

R−1<1.

Proof. In Appendix.

Proposition 1 states that, for each family iwith income yi∈[1, y], there exists a

threshold in the network density, en(yi), such that if ni≥en(yi) this family chooses to send

a migrant. The network threshold en(y) is strictly increasing in y, implying that a larger

mass of migrants will originate from relatively poor families than from relatively rich

ones. The intuition for this is that the marginal utility of consumption is decreasing in

the level of consumption, while the disutility from migration, M, is constant for any level

of consumption. As a result, poorer families will be more eager to endure the emotional

cost M, because their marginal return of migration in terms of (expected) utility of

additional consumption is larger. Notice, ﬁnally, that Proposition 1 does not explicitly

restrict en(y)≤1. In fact, it may well be the case that en(y)>1 for some y > 1, implying

that no migrants will originate from families with incomes above that level.

From now onwards we let M≤ln(R) hold. This assumption can be read as saying

that the emotional cost of migration, M, is not too large relative to the returns from

investing in risky assets, R. Notice from the last sentence in Proposition 1 that, since

M≤ln(R) implies enR

R−1<1, then there will exist some families whose incomes are

below the threshold level R/(R−1) who will choose to send migrants.

12

The next step is to study how migration decisions interact with investment decisions.

In particular, we are interested in studying whether families send migrants with the aim

to increase their capacity to invest in the projects. By merging the migration results in

Proposition 1 with the preceding discussion in this section, we can summarize households’

optimal decisions concerning migration and investment in the following corollary.

Corollary 1

(i) If R≥υ−1+υ

4. Then by≤1, and:

a) For any y∈R

R−1, y: If ni≥en(y)and yi=y, family isends a migrant. If ni<en(y)

and yi=y, family idoes not send a migrant. Family ialways invests in the project.

b) For any y∈1,R

R−1: If ni≥en(y)and yi=y, family isends a migrant and invests in

the project if and only if the migrant ﬁnds a good job. If ni<en(y)and yi=y, family i

does not send a migrant and does not invest in the project.

(ii) If R < υ−1+υ

4. Then by > 1, and:

a) For any y∈R

R−1, y: If ni≥en(y)and yi=y, family isends a migrant. If ni<en(y)

and yi=y, family idoes not send a migrant. Family ialways invests in the project.

b) For any y∈by, R

R−1: If ni≥en(y)and yi=y, family isends a migrant and invests in

the project if and only if the migrant ﬁnds a good job. If ni<en(y)and yi=y, family i

does not send a migrant and does not invest in the project.

c) For any y∈[1,by]: If ni≥en(y)and yi=y, family isends a migrant. If ni<en(y)and

yi=y, family idoes not send a migrant. Family inever invests in the project.

The results from Corollary 1 can be visually summarized in Figure 1. The key insight

of the corollary can be gleaned from point b), both for cases (i) and (ii) therein. The

result in b) says there exist some families who use migration as a mechanism to mitigate

credit constraints that prevent them from investing in projects that would raise their

intertemporal income. Essentially, those families send a migrant, betting on the chance

13

that this migrant ﬁnds a good job, which would increase their total income in t= 1 and,

thus, place them in better position to undertake the unit investment that yields R > 1

units of income in t= 2.

2.3 Eﬀect of migration on investment decisions

We now study the eﬀect of migration on families’ investment decisions. The migration

eﬀect results from calculating the diﬀerence in investment decisions between migrant and

non-migrant families. First consider E[I|m= 1, y]−E[I|m= 0, y], where Iand m

are indicator functions regarding investment and migration decisions, respectively. In

relation to the empirical results in this paper, we refer to this model as ﬁxed-eﬀects

(FE) model, because by conditioning on ywe are controlling for the family-speciﬁc FE.

Note from Corollary 1 that, for any y≥R

R−1, families choose I= 1 irrespective of their

migration choice; while (in case (ii) of the corollary), for y < by, families always set I= 0,

regardless of their migration choices. It follows then that migration has only an eﬀect on

the investment behavior of families with by≤y < R

R−1; in particular:

EI|m= 1,by≤y < R

R−1

| {z }

>0

−EI|m= 0,by≤y < R

R−1

| {z }

= 0

>0 (13)

Equation (13) makes it explicit that migration exerts a positive eﬀect on investment

decisions.9However, notice that a key feature of the problem is the fact that intrinsic

family characteristics need to be taken into account when evaluating the eﬀect of migration

on investment. In fact, if those characteristics are not controlled for, the measured eﬀect

9The analytical expression for EhI|m= 1,by≤y < R

R−1iis given by:

R/(R−1)

Zby

[1 −en(yi)] dyi

−1R/(R−1)

Zby

1 + en(yi)

2[1 −en(yi)] dyi.

14

of migration on investment may turn out to be incorrect, because by simply comparing

the average behavior of families with and without migrants, we may also be capturing the

inﬂuence of other variables that somehow correlate with migration decisions.

To make this last argument more precise, consider now the overall association between

migration and investment in the population; this results from calculating the diﬀerence,

E[I|m= 1] −E[I|m= 0]. In parallel with the empirical results, we refer to this model

as ordinary least-squares (OLS) eﬀect. After some algebra we obtain

E[I|m= 1] −E[I|m= 0] = Pr hby < y < R

R−1m= 1i·EhIm= 1,by < y < R

R−1i

| {z }

positive

+

nPr hy≥R

R−1m= 1i−Pr hy≥R

R−1m= 0io

| {z }

,

negative

(14)

where Pr y≥R

R−1m= 1<Pr y≥R

R−1m= 0follows from the fact that the threshold-

function en(y) is monotonically increasing in y.

The ﬁrst thing that can be observed from (14) is that it is no longer true that families

with migrants tend to invest more than families without migrants; that is, E[I|m= 1] −

E[I|m= 0] ≶0. Furthermore, we can also show that OLS eﬀect is always smaller that

the FE eﬀect. We refer to this diﬀerence as the OLS bias.

Proposition 2 The OLS bias is negative, that is:

[E(I|m= 1) −E(I|m= 0)]−EI|m= 1,by≤y < R

R−1−EI|m= 0,by≤y < R

R−1<0

(15)

Proof. Note: The following proof is conducted for the case in which by≤1. The proof

for the case in which by > 1 is almost identical to this one, and it is available from the

authors upon request.

15

The expression (15) can be re-ordered as follows:

OLS bias = EI|m= 0,by≤y < R

R−1−E(I|m= 0)

| {z }

A

−EI|m= 1,by≤y < R

R−1−E(I|m= 1)

| {z }

B

.

(16)

Recalling (13), we can observe that the ﬁrst member of (16) simpliﬁes to:

A= 0 −Pr y≥R

R−1m= 0=−Pr y≥R

R−1m= 0.

In the case of the second member of (16), we have:

B=EI|m= 1,by≤y < R

R−1−Pr by≤y < R

R−1m= 1EIm= 1,by≤y < R

R−1

−Pr y≥R

R−1m= 1

=EI|m= 1,by≤y < R

R−11−Pr by≤y < R

R−1m= 1

| {z }

Pry≥R

R−1m=1

−Pr y≥R

R−1m= 1

=−Pr y≥R

R−1m= 11−EI|m= 1,by≤y < R

R−1

Therefore, we can in the end obtain:

A−B=−Pr y≥R

R−1m= 0+ Pr y≥R

R−1m= 11−EI|m= 1,by≤y < R

R−1

which is always strictly negative for the combined eﬀect of the following two properties:

1) The monotonicity of en(y) implies that: Pr y≥R

R−1m= 0>Pr y≥R

R−1m= 1.

2) The fact that EI|m= 1,by≤y < R

R−1<1. This is because, among the families with

by≤y < R/(R−1) and send migrants, only in those cases in which the migrant manages

to ﬁnd a good job (which occurs with probability ni) do families invest in the project.

16

The OLS bias arises because the OLS regression underestimate the eﬀect of migration

on investment. This occurs because the family-speciﬁc level of income (yi) and the mi-

gration decision cannot be separated. In consequence, it is important to control for the

level of income yior other family-speciﬁc characteristics to get an unambiguous eﬀect.

17

3 Data

We make use of a unique new dataset available for poor rural households in Mexico.

The data was collected for administrative purposes by the Oportunidades (ex Progresa)

program.10 Thanks to retrospective information, we managed to construct a panel of

households based on three surveys. In December 2006, the Instituto Nacional de Salud

P´ublica conducted a survey11 of recipient households in the rural localities where the Opor-

tunidades program started in 1997 with a 10% random sample, stratiﬁed by state. This

database is then matched to another survey, the ENCASEH (Encuesta de Caracteristicas

Socioeconomicas de los Hogares), carried out in 1997 and 1998, and to the ENCRECEH

(Encuesta de Recertiﬁcaci´on de los Hogares) carried out in 2001. This allows us to build

a balanced panel database composed of three time observations (1997, 2001 and 2006) for

4,365 households from 130 rural localities.

This constructed database includes detailed information on each beneﬁciary household,

including household demographics, income level and sources, education and several types

of assets. It also includes locality-level data, mainly regarding infrastructure. Although

it was not designed to evaluate migration patterns the database contains a few questions

about household members that migrated. Moreover, from the income data we obtain

information about remittances. Given the risk of attrition bias in our estimation, we

compared the distributions between the balanced panel of 4,365 and the unbalanced panel.

The distributions of the kernel density estimates appear to be very close to each other

and this is conﬁrmed by the results of Kolmogorov-Smirnov tests that we run on the

hypothesis that the distributions of the balanced and unbalanced panels are the same for

10Launched in Mexico in 1997, it is a program whose main aim is to improve the process of human

capital accumulation in the poorest communities by providing conditional cash transfers on speciﬁc types

of behavior in three key areas: nutrition, health and education. Nevertheless, these households are also

targeted by other social programs.

11Encuesta de “Re-evaluaci´on de localidades incorporadas en las primeras fases del Programa (1997-

1998).” INSP, 2006.

18

some key variables. The null hypothesis cannot be rejected across all tests12.

In sum the dataset seems well suited for the purposes of the paper, because it allows

us to capture diverse information on households along with the time dimension that is

useful to control for the household ﬁxed-eﬀects. However, it should be noted that this

database may not be representative of rural Mexico because it was designed to cover a

particular subset of the population (i.e. those receiving Oportunidades). Therefore the

conclusions from the empirical results may only apply to this group.

12Not shown but available from the authors upon request.

19

4 The construction of an asset index

The ﬁrst step in the empirical analysis is to reduce the household assets into unidimen-

sional measures. This requires either complete knowledge of the market value of each

asset owned or the construction of an asset index. Given that the prices of many assets

owned by households are often unknown or diﬃcult to determine, we construct the as-

set index using the methodology used by Adato et al. (2006): the household income13

is regressed on the household’s stock of assets. The household asset index is then the

household income predicted from the estimated coeﬃcients in the ﬁrst year (1997), which

are used to extrapolate to every year. The equation we estimate is of the form:

yi,1997 =β0+β1x1i,1997 +β2x2i,1997 +ST AT Ei+ei,1997 ,(17)

where yi,t is the per-capita income by household, x1i,t is a vector of household assets we are

interested in, x2i,t is a vector of other household characteristics and STATE correspond

to state dummy variables. The asset index is then constructed as

Ai,t =ˆ

β1x1i,t.(18)

The asset index is standardized by its standard deviation. This simpliﬁes the interpre-

tation of the regression analysis results (i.e. a regression coeﬃcient of one means one

standard deviation of the index).

We consider three asset indexes and four categories of assets:

-AP: Productive assets: owner of a truck, agricultural land, irrigated land, working

animals;

13Income aggregates were created and broken down into ﬁve categories: agricultural wage employ-

ment, non-farm wage employment, self employment, transfers and other (including income from rent and

interests).

20

-ANP : Non-Productive (leisure) assets: ownership of radios, TV, refrigerator, gas

stove, washing machine and vehicles;

-AT: Total assets: APand ANP ;

- Other dwelling and household characteristics such as: electricity, earth ﬂoor, roof

weak, domestic animals, own house, years of education of the household head.

We compute the asset indexes for the diﬀerent periods in the panel in Table 1. The

table shows that there is a marked increase in asset accumulation for all households (HH)

during the ten-year period. In Figure 2 we present density plots for migrant and non-

migrant households for each type of asset. Overall, the ﬁgures show that there are no

considerable diﬀerences across migrant and non-migrant HHs.

21

5 Descriptive statistics

According to the Bank of Mexico, Mexican migrants have remitted in 1998 an amount

of income that equals approximately 1.5% of Mexican GDP. Household level surveys also

show that remittances tend to play a key role on the survival and livelihood strategies for

many (typically rural) poor households (Rapoport and Docquier, 2005). We take advan-

tage of our detailed panel database to describe the economic role played by remittances

in the rural poor households. Tables 1, 2 and 3 present summary statistics of the vari-

ables of interest for the balanced panel of Mexican rural households. This information is

presented for the pooled database and disaggregated for the three diﬀerent periods of the

panel: 1997, 2001 and 2006.

We construct a dummy variable at the household level that indicates whether the

household has at least one member who is a migrant (i.e., working in another locality,

state or abroad). As can be observed in Table 2, in 1997, 5% of the households had a

migrant member, while 3% had a member in the US. These percentages are somewhat

reduced in 2001 (3% and 2%, respectively), but increase considerably in 2006 (10% and

7%, respectively). These results show that even when we follow the same households over

a long period of time (10 years), there is considerable variation in migration statistics at

the household level.

Other summary statistics appear in Table 3. The table shows that remittances rep-

resent less than 10% of the total income in the household (0.6/7.7). Surprisingly, this

ratio is very similar for households with current member/migrants and for those without

(the reason for this is that remittances may come from past migrants). The (pooled)

average household has a household head with 3.3 years of schooling and has 1.4 male

adults in the labor force. Both schooling and labor participation increase in 2006. The

table also reports community level variables that will be used as an instrumental variable

in the next section. HH w/mig / #HH (at com.) is the proportion of households at the

22

community level with at least one household member being a migrant. HH w/USmig /

#HH (at com.) represents a similar ratio but for the case when the migrant lives in the

US. As explained in the next section, the instrumental variable will work well if there is

enough variation both across levels and across type of households. A visual inspection of

the table reveals that this is indeed the case.

23

6 Econometric analysis

Let Ait be an asset index for family iand year t. We are mostly interested in household-

speciﬁc asset dynamics, that is in Gi,t ≡Ai,t −Ai,t−1. Let Mi,t be a variable that cap-

tures the migration-related nature of the household; Xit be household characteristics; and

(µi+it) be an error component with household ﬁxed-eﬀects and idiosyncratic temporary

shocks. We consider the following asset dynamics equation:

Gi,t =αAi,t−1+βMi,t +δXi,t +µi+i,t (19)

We are mostly concerned with β≡∂E[Gi,t |Ai,t−1,Mi,t,Xi,t−1,µi,η t]

∂M , which denotes the con-

ditional eﬀect of migration on asset accumulation. We extend this analysis to a multi-

dimensional measure of assets A={AP, ANP }, where APdenotes productive assets and

ANP non-productive assets. As argued above, the question we want to address here is the

eﬀect of migration on the type of assets that families accumulate.

We study the eﬀect of migration on asset accumulation using three diﬀerent measures

of migration. First, we consider a dummy variable for households that declare having

at least one migrant member, Migrant HH (see Table 4). Second, we use the number

of migrants in the household, Number of Migrants by HH (see Table 5). Third, we use

remittances per capita (see Table 6). In each case, we separately study the eﬀect migration

on: (i) total assets, (ii) productive assets, and (iii) non-productive assets.

6.1 Endogeneity

Several endogeneity issues need to be addressed in order to avoid potential biases in

this estimator. First, households may respond to adverse or positive shocks () changing

the number of migrants or the nature of migration (temporal vs. permanent). Second,

selection bias may occur if migrant households are intrinsically diﬀerent from non-migrant

24

ones.14 Acosta (2006) uses migration networks and history (at the village or household

level) as instruments for migration (or remittances) postulating that these variables have

a positive impact on the opportunity to migrate but no additional impact on income,

schooling, or nutrition at home. McKenzie and Sasin (2007) argue that these instruments

are suitable to study the migration impact at the originary location as in our case.

Following previous work on this subject, the IV strategy we follow uses the percentage

of migrants (to all destinations and to the US, separately) at the community level as

an instrument for the household level decision. Because we have a panel data, we can

include this variable together with the ﬁxed eﬀects at the household level. Therefore, our

identiﬁcation strategy relies on variation in aggregate migration across time and space.

The Sargan test for overidentiﬁcation in the following tables has an average p-value of 0.1

for total and productive assets, and 0.4 for non-productive assets. As a result they do not

reject the null hypothesis of exogeneity of the instrumental variables.15 Moreover, both

instruments are signiﬁcant on the ﬁrst stage of the regression with high F-values.

6.2 Second stage

In all cases the OLS eﬀect of migration on assets accumulation is negative and statistically

signiﬁcant. However, when we include the household-level FE this eﬀect becomes non-

signiﬁcant, except for non-productive assets where it continues to display a negative sign

and signiﬁcant. The FE results also show that total and productive assets may have a

positive correlation with migration. The diﬀerences between OLS and FE are in line with

those outlined before in Section 2. That is, when comparing (13) with ( 14), the eﬀect of

14Regarding the relationship between migration and self-selection, Borjas (1987, 1991) has formalized

the endogeneity of the migration decision, showing that the welfare impact of immigrants is crucially

dependent on the degree of transferability of their unobservable and observable variables, and that aﬀects

the labour market.

15However, when we use migration to the US as our endogenous variable, Sargan tests reject the null

hypothesis of exogeneity of the IV. Therefore, we only evaluate the eﬀect of total migration.

25

FE should be bigger than that of OLS.

Next, we follow the IV strategy described above. Both total assets and productive

assets become positive and statistically signiﬁcant while non-productive assets is, in gen-

eral, negative and statistically signiﬁcant. A striking feature is actually the magnitude

of the eﬀect. The coeﬃcient of the migrant dummy variable can be interpreted as the

change in standard deviation units of the corresponding asset. Therefore this shows that

having a migrant household increases total asset accumulation by 0.8 standard deviation

units. Moreover, one additional household migrant contributes to 0.2 total assets stan-

dard deviation units. Finally, doubling the amount of remittances per capita increases

assets by 1.2/10 of a standard deviation.

The magnitude and sign of the eﬀect on productive assets follow closely that of total

assets. Having a migrant household increases productive asset accumulation by 0.8 stan-

dard deviation units. Moreover, one additional household migrant contributes to 0.2 total

assets standard deviation units. Finally, doubling the amount of remittances per capita

increases assets by 1/10 of a standard deviation. However, there is a negative and sta-

tistically signiﬁcant eﬀect on non-productive asset accumulation of a similar magnitude.

We consider that the negative coeﬃcient in non-productive assets is also an interesting

result in itself. It suggests that some families with migrants reduce their spending in

non-productive asset so as to leave additional funds available for the accumulation of pro-

ductive assets.16 This result can in fact be related to our model in Section 2. There, we

have shown the existence of a minimum initial level of wealth that is necessary to hold in

order to invest. Families at the margin of y=R/(R−1), who now choose to invest as a

consequence of migration, may reinforce the magnitude of their project by concomitantly

reducing consumption.17

16Unfortunately, the surveys do not have current consumption.

17Strictly speaking, this does not occur in our (highly) stylized model because we assume that υ≥1 (see

equation (8)) together with a ﬁxed level of investment. However, letting υ > 0 would straightforwardly

lead to the result that households at the margin of y=R/(R−1) will reduce consumption to help raising

26

Overall the results show that migration can be seen as a long-term investment for the

household. Therefore, the income sent back home by the migrant is used to accumulate

productive assets, rather than non-productive assets. These ﬁndings appear across all the

diﬀerent speciﬁcations presented in the tables.

funds for investment, when the migrant ﬁnds a good job and 0 < υ < 1.

27

7 Conclusion

This paper aims at explaining the link between migration and asset dynamics for a panel

of poor rural households in Mexico over the period 1997-2006. Our results suggest that

migration may be used by households as a mechanism to accelerate asset accumulation

in productive assets. The general idea is that remittances may help alleviate credit con-

straints for poor households, thus allowing them to invest in productive assets that would

be optimal under complete markets. Furthermore, our estimations also suggest that fam-

ilies who send migrants with the intention to channel remittances towards investment

in productive assets, concomitantly reduce their accumulation of non-productive assets,

possibly to further contribute to raising funds for physical investment.

An important caveat concerning our analysis is that it has abstracted from general

equilibrium interactions, so as to focus exclusively on the direct eﬀect of migration on

capital accumulation via remittances. One speciﬁc general equilibrium eﬀect that may

be particularly relevant in our context is the fact that migration decisions will necessarily

aﬀect the aggregate labor supply at the home village. On the one hand, migration lowers

aggregate labor supply at the village level, which in turn would raise equilibrium wages

and household incomes (see Jaimovich, 2010, for a growth model where this mechanism

is at play). However, looking at the household level, sending out a migrant also means

losing one of their workers (and, possibly, the most productive worker). Furthermore, it

may well be the case that the wealth eﬀect brought about by the migrant leads household

members who remain at the village to increase their leisure consumption. In that regard,

two remarks apply here. First, although we acknowledge that these eﬀects imply that

migration may inﬂuence accumulation also by other channels other than remittances, we

are agnostic concerning the overall sign of these additional eﬀects. Second, the above

general equilibrium eﬀect on the wage, which could be expected to induce an upwards

bias on the eﬀect of remittances, will be of signiﬁcant magnitude only if the total number

28

of migrants from the rural village varies substantially across our years of observations. In

that respect, the results in Table 2 show that the percentage of families with at least one

migrant ranges within 3% to 10% of the sampled households.

In a similar vein the eﬀect of migration and remittances are both confounded. We

should expect that remittances increase the probability of capital accumulation as it

relaxes credit constraints. On the other hand, migration would decrease that probability

because of the loss of household members and/or less incentives to work. Both eﬀects

could be further exploited, as for example studying whether results change or not when

we analyze the impact of remittances for the sub-sample of migrant households vis-a-vis

non-migrant households.

29

Appendix

Proof of Proposition 1.

Step 1: Let yi∈R

R−1, yand deﬁne:

en1(yi)≡M

ln (yi+υ−1) −ln (yi−1).(20)

Notice ﬁrst that en1(yi)>0 and ﬁnite, since both the numerator and denominator in (20)

are strictly positive and ﬁnite. Secondly, diﬀerentiating (20) with respect to yiyields:

den1

dyi

=M

[ln (yi+υ−1) −ln (yi−1)]21

yi−1−1

yi+υ−1>0,

where the result en0

1(yi)>0 follows from the fact that yi−1< yi+υ−1. Finally, since

the left-hand side in (7) is strictly increasing in ni, it immediately follows that for any

ni>en1(yi) condition (7) holds.

Step 2: Let yi≥1 and yi∈by, R

R−1and deﬁne:

en2(yi)≡M

ln (yi+υ−1) + ln(yi+R)−ln (y2

i).(21)

Firstly, en2(yi)>0 and ﬁnite, because both the numerator and denominator in (21) are

strictly positive and ﬁnite. Secondly, diﬀerentiating (21) with respect to yiyields:

den2

dyi

=M

[ln (yi+υ−1) + ln(yi+R)−ln (y2

i)]22

yi

−2yi+R+ 2(υ−1)

y2

i+R(yi−1) + yi(υ−1) + υR >0,

(22)

where en0

2(yi)>0 obtains after some algebra on the second term in right-hand side of (22),

which leads to the condition that en0

2(yi)>0 iﬀ yi(R−1) + yiυ+ 2R(υ−1) >0.Lastly,

since the left-hand side in (11) is strictly increasing in ni, it immediately follows that for

30

any ni>en2(yi) condition (11) prevails.

Step 3: Let yi≥1 and yi≤byand deﬁne:

en3(yi)≡M

ln yi+υ

22−ln (y2

i).(23)

As in the previous two cases, en3(yi)>0 and ﬁnite, as both the numerator and denominator

in (23) are strictly positive and ﬁnite. Next, diﬀerentiating (23) with respect to yiyields:

den3

dyi

=M

hln yi+υ

22−ln (y2

i)i2 2

yi

−2yi+υ

y2

i+υ2

4+yiυ!>0,(24)

where en0

3(yi)>0 obtains after some algebra on the second term in right-hand side of (24),

which leads to the condition that en0

3(yi)>0 iﬀ υ2

2+yiυ > 0.Finally, since the left-hand

side in (12) is strictly increasing in ni, it trivially follows that for any ni>en3(yi) condition

(12) holds.

Step 4: Let now,

en(yi) =

en1(yi) if R

R−1≤yi≤y,

en2(yi) if yi≥1 and by < yi<R

R−1,

en3(yi) if yi≥1 and yi≤by.

Replacing yi=R

R−1into (20) and (21), we can observe after some simple algebra that

en1R

R−1=en2R

R−1. Similarly, from the deﬁnition of byin (10), replacing yi=byinto (21)

and (23), it follows that en2(by) = en3(by). As a consequence, it follows that en(yi) portrays

a continuous and strictly increasing function and en(yi) : R++ →R++.

Step 5: Finally, to prove that enR

R−1<1, notice that plugging yi=R

R−1into (20) leads

31

to:

en1R

R−1=M

ln 1

R−1+υ−ln 1

R−1=M

ln 1+υ(R−1)

R−1

1

R−1=M

ln [1 + υ(R−1)].

Therefore, en1R

R−1<1 iﬀ M < ln [1 + υ(R−1)], which is guaranteed by M≤ln(R)

together with υ≥1 and R > 1.

32

References

Acosta P. (2006) “Labor Supply, School Attendance, and Remittances from International Mi-

gration: The Case of El Salvador,” World Bank Policy Research Working Paper N.3903.

Adams R. (1998) “Remittances, Investment, and Rural Asset Accumulation in Pakistan,”

Economic Development and Cultural Change 47:1, 155-173.

Adato, M., Carter, M.R. and May, J. (2006) “Exploring poverty traps and social exclusion

in South Africa using qualitative and quantitative data,” Journal of Development Studies 42:2,

226 - 247.

Banerjee, B. and Bucci, G. A. (1994) “On-The-Job Search After Entering Urban Employ-

ment: An Analysis Based On Indian Migrants,” Oxford Bulletin Of Economics And Statistics

56, 1, 33-47.

Banerjee, A. and Newman, A., (1998) ”Information, the Dual Economy, and Development,”

Review of Economic Studies, 65, 631-653.

Borjas, G. J. (1987) “Self-Selection And The Earnings Of Immigrants,” American Economic

Review 77, 531-53.

Borjas, G. J. (1991) “Immigration And Self-Selection,” in Abowd and Freeman 27-96.

Carter, M. R. and Barrett C. (2006) “The Economics of Poverty Traps and Persistent

Poverty: An Asset-Based Approach,” Journal of Development Studies 42, 178-199.

Das Gupta, M. (1987) “Informal Security Mechanisms and Population Retention in Rural

India,” Economic Development and Cultural Change 36, 101-120.

Durand, J. and Massey D. S. (2001) “Mexican Migration to the United States: a critical

review,” Latin American Research Review 27, 3-42.

Ghatak, S., Levine, P. and Wheatly Price, S. (1996) “Migration Theories and Evidence: An

Assesment,” Journal of Economic Surveys 10, 159-198.

Giuliano, P. and Ruiz-Arranz, M. (2009) “Remittances, ﬁnancial development, and growth,”

Journal of Development Economics 90, 144–152.

33

Glytsos, N. P. (1993) “Measuring The Income Eﬀects of Migrant remittances: A Method-

ological Approach Applied to Greece,” Economic Development and Cultural Change 42, 131-168.

Hanson, G.H. and Woodruﬀ, C. (2003) “Emigration and Educational Attainment in Mexico,”

mimeo.

Harris, J.R. and Todaro, M.P. (1970) “Migration, Unemployment And Development: A

Two-Sector Analysis,” American Economic Review 60, 126-142.

Jaimovich, E. (2010) “Adverse Selection and Entrepreneurship in a Model of Development,”

Scandinavian Journal of Economics 112, March 2010.

Lewis, W.A. (1954) “Economics Development With Unlmimited Supplies Of Labor,” Manch-

ester School of Economic and Social Studies 22, 139-191.

Lloyd-Ellis, H. and Bernhardt, D. (2000) ”Enterprise, Inequality and Economic Develop-

ment,” Review of Economic Studies 67, 147-168.

Lucas, R.E.B. and Stark, O. (1985) “Motivations to Remit: Evidence from Botswana,’

Journal of Political Economy 93, 901-918.

McKenzie, D. and Sasin, M.J. (2007) “Migration, Remittances, Poverty, and Human Capital:

Conceptual and empirical challenges,” World Bank Policy Research Working Paper N.4272.

McKenzie, D. and Woodruﬀ, C. (2006) “Do entry costs provide an empirical basis for poverty

traps? Evidence from Mexican microenterprises,” Economic Development and Cultural Change

55, 3-42.

Mesnard, A. (2004) ”Temporary migration and capital market imperfections,” Oxford Eco-

nomic Papers 56, 242–262.

Mesnard, A. and Ravallion, M. (2006) “The wealth eﬀect on new business startups in a

developing economy,” Economica 73, 367–392.

Molle, W. and Van Mourik, A. (1988) “International movements of labour under conditions

of economic integration: the case of Western Europe,” Journal of Common Market Studies 26,

317-42.

34

Munshi K. (2003) “Networks in the modern economy: Mexican Migrants in the U.S. Labor

Market,” Quarterly Journal of Economics 118, 549-599.

Montgomery, J. (1991) “Social Networks and Labor-Market Outcomes: Toward an Economic

Analysis,” American Economic Review 81, 1408–1418.

Rapoport, H. and Docquier, F. (2005) “The Economics of Migrants’ Remittances”, IZA

Discussion Paper No. 1531, chapter prepared for the Handbook on the Economics of Reciprocity,

Giving and Altruism, Elsevier- North Holland, forthcoming (Editors: L.-A. Gerard-Varet, S.-C.

Kolm and J. Mercier-Ythier).

Rapoport, H. (2002) “Migration, credit constraints and self-employment: A simple model of

occupational choice, inequality and growth.” Economics Bulletin 15, 1-5

Rosenzweig, M.R. and Wolpin, K.I. (1993) “Credit market constraints, consumption smooth-

ing, and the accumulation of durable production assets in low-income countries: invesment in

bullocks in India,” Journal of Political Economy 101, 223–244.

Rozelle, S., Taylor, J.E. and DeBrauw, A. (1999) “Migration, Remittances, and Agricultural

Productivity in China,” American Economic Review 89, 287-291.

Rozelle, S., Taylor, J.E. and DeBrauw, A. (2003) ”Migration and Incomes in Source Com-

munities: A New Economics of Migration Perspective from China,” Economic Development and

Cultural Change 52, 75-101.

Sen, A.K. (1966) ”Peasants and Dualism with or without Surplus Labour,” Journal of Po-

litical Economy 74, 425-450.

Stark, O. (1991) “The Migration Of Labor,” Cambridge, Mass.: Basil Blackwell.

Stark, O. and Levhari, D. (1982) “On Migration and Risk in LDCs,” Economic Development

and Cultural Change 31, 191-196.

Woodruﬀ, C. and Zenteno, R. (2007) “Migration networks and microenterprises in Mexico,”

Journal of Development Economics 82, 509-528.

Yang, D.. (2008) “International Migration, Remittances, and Household Investment: Evi-

35

dence from Philippine Migrants’ Exchange Rate Shocks,” Economic Journal 118, 591-630.

36

Table 1: Asset Indexes, by HH migrant status

All HH HH with HH without

migrants migrants

All years

Asset Index 0.5 0.503 0.499

[0.45] [0.471] [ 0.447 ]

Non-productive Asset Index 0.418 0.422 0.417

[0.357] [0.363] [ 0.356]

Productive Asset Index 0.1 0.085 0.102

[0.198] [0.214] [0.196]

N 13,095 1,443 11,652

1997

Asset Index 0.388 0.387 0.388

[0.44] [ 0.452] [0.438]

Non-productive Asset Index 0.344 0.342 0.345

[ 0.336] [0.345] [0.335 ]

Productive Asset Index 0.037 0.027 0.038

[0.203] [0.2] [0.203]

2001

Asset Index 0.478 0.474 0.478

[0.445] [0.466] [0.442]

Non-productive Asset Index 0.391 0.387 0.392

[0.363] [0.359] [0.363]

Productive Asset Index 0.123 0.103 0.126

[0.189] [0.214] [0.185]

2006

Asset Index 0.634 0.649 0.632

[0.43] [0.457] [0.427]

Non-productive Asset Index 0.517 0.536 0.515

[0.348] [0.357] [0.346]

Productive Asset Index 0.142 0.126 0.143

[0.186] [0.214] [0.182]

37

Table 2: Summary statistics: migration

1997 2001 2006

Migration HH 0.05 0.03 0.10

Migration HH to the US 0.03 0.02 0.07

Number of HH 4365 4365 4365

38

Table 3: Summary Statistics

HH All HH HH w/mig HH wo/mig

Variable Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.

All years

Per capita inc 7.7 1.9 7.847 1.70 7.667 1.95

Remittances 0.6 3.3 1.694 8.17 0.462 1.96

Yrs educ (head) 3.362 2.079 3.977 2.23 3.271 2.45

HH male adults 1.436 1.18 1.784 1.31 1.393 1.15

#HH w/mig / #HH (at com.) 0.012 0.027 0.021 0.05 0.011 0.02

#HH w/USmig / #HH (at com.) 0.01 0.025 0.018 0.05 0.009 0.01

1997

Per Capita inc 7.289 2.536 7.424 2.275 7.272 2.566

Remittances 0.4 2.262 0.4 2.235 0.4 2.265

Yrs educ (head) 3.273 2.296 3.662 2.089 3.216 2.32

HH male adults 1.256 1.03 1.426 1.099 1.235 1.02

#HH w/mig / #HH (at com.) 0.011 0.031 0.018 0.05 0.011 0.028

#HH w/USmig / #HH (at com.) 0.008 0.026 0.015 0.05 0.008 0.021

2001

Per capita inc 7.776 1.502 7.925 1.198 7.757 1.535

Remittances 0.503 1.836 0.305 1.398 0.528 1.882

Yrs educ (head) 3.245 2.376 3.85 2.155 3.156 2.394

HH male adults 1.29 1.046 1.674 1.204 1.243 1.015

#HH w/mig / #HH (at com.) 0.008 0.021 0.015 0.049 0.007 0.013

#HH w/USmig / #HH (at com.) 0.006 0.019 0.012 0.048 0.005 0.011

2006

Per capita inc 7.997 1.503 8.193 1.334 7.972 1.521

Remittances 0.883 4.914 4.315 13.436 0.454 1.725

Yrs educ (head) 3.567 2.609 4.42 2.381 3.44 2.617

HH male adults 1.763 1.364 2.254 1.477 1.702 1.337

#HH w/mig / #HH (at com.) 0.017 0.029 0.029 0.068 0.015 0.019

#HH w/USmig / #HH (at com.) 0.014 0.027 0.026 0.067 0.013 0.016

39

Table 4: Growth of the Asset Index - Migrant Household

(1) (2) (3)

OLS FE IV-FE

ALL ASSETS

Asset Indext−1-0.569*** -1.334*** -1.357***

(0.00986) (0.0126) (0.0154)

Migrant HH -0.140*** 0.0601 0.827***

(0.0414) (0.0455) (0.279)

HH male adults -0.0715*** 0.0836*** 0.0607***

(0.00783) (0.0128) (0.0156)

R20.282 0.729 0.712

Sargan Test 0.0942

First Stage F-Test 63.81

PRODUCTIVE ASSETS

Asset Indext−1-0.553*** -1.349*** -1.367***

(0.00981) (0.0125) (0.0148)

Migrant HH -0.136*** 0.0596 0.689***

(0.0413) (0.0444) (0.267)

HH male adults -0.0799*** 0.0538*** 0.0347**

(0.00781) (0.0125) (0.0151)

R20.274 0.738 0.726

Sargan Test 0.0965

First Stage F-Test 64.87

NON PRODUCTIVE ASSETS

Asset Indext−1-0.657*** -1.491*** -1.485***

(0.0120) (0.0161) (0.0165)

Migrant HH -0.135*** -0.162*** -0.678**

(0.0467) (0.0543) (0.300)

HH male adults -0.0516*** -0.0304** -0.0113

(0.00885) (0.0150) (0.0187)

R20.258 0.664 0.657

Sargan Test 0.434

First Stage F-Test 75.36

Observations 8.730 8.730 8.730

Households 4.365 4.365

Notes: Robust standard errors in parentheses. * signiﬁcant at 10%; ** signiﬁcant at 5%; *** signiﬁcant at 1%. See text for

variable deﬁnitions.

40

Table 5: Growth of the Asset Index - Number of Migrants by Household

(1) (2) (3)

OLS FE IV-FE

ALL ASSETS

Asset Indext−1-0.569*** -1.333*** -1.357***

(0.00986) (0.0125) (0.0158)

Number of Migrants by HH -0.0290*** 0.00847 0.220***

(0.00911) (0.0101) (0.0768)

HH male adults -0.0703*** 0.0841*** 0.0510***

(0.00790) (0.0129) (0.0180)

R20.282 0.729 0.702

Sargan Test 0.0812

First Stage F-Test 41.93

PRODUCTIVE ASSETS

Asset Indext−1-0.554*** -1.348*** -1.367***

(0.00981) (0.0124) (0.0150)

Number of Migrants by HH -0.0299*** 0.00826 0.182**

(0.00909) (0.00982) (0.0732)

HH male adults -0.0784*** 0.0543*** 0.0266

(0.00788) (0.0126) (0.0174)

R20.274 0.738 0.719

Sargan Test 0.0844

First Stage F-Test 42.80

NON PRODUCTIVE ASSETS

Asset Indext−1-0.657*** -1.491*** -1.483***

(0.0120) (0.0161) (0.0169)

Number of Migrants by HH -0.0211** -0.0281** -0.186**

(0.0103) (0.0120) (0.0820)

HH male adults -0.0516*** -0.0312** -0.00241

(0.00893) (0.0151) (0.0213)

R20.258 0.663 0.650

Sargan Test 0.480

First Stage F-Test 49.78

Observations 8.730 8.730 8.730

Households 4.365 4.365

Notes: Robust standard errors in parentheses. * signiﬁcant at 10%; ** signiﬁcant at 5%; *** signiﬁcant at 1%. See text for

variable deﬁnitions.

41

Table 6: Growth of the Asset Index - Remittances per capita

(1) (2) (3)

OLS FE IV

ALL ASSETS

Asset Indext−1-0.570*** -1.334*** -1.352***

(0.00986) (0.0125) (0.0155)

Remittances per capita 0.000863 0.00849** 0.124***

(0.00341) (0.00378) (0.0447)

HH male adults -0.0750*** 0.0841*** 0.0660***

(0.00776) (0.0128) (0.0157)

R20.281 0.729 0.671

Sargan Test 0.117

First Stage F-Test 19.05

PRODUCTIVE ASSETS

Asset Indext−1-0.554*** -1.348*** -1.362***

(0.00982) (0.0124) (0.0147)

Remittances per capita 0.000579 0.00608* 0.104**

(0.00340) (0.00369) (0.0423)

HH male adults -0.0833*** 0.0546*** 0.0390***

(0.00774) (0.0125) (0.0150)

R20.274 0.738 0.696

Sargan Test 0.115

First Stage F-Test 19.42

NON-PRODUCTIVE ASSETS

Asset Indext−1-0.658*** -1.492*** -1.481***

(0.0120) (0.0161) (0.0177)

Remittances per capita -0.0122*** -0.00499 -0.104**

(0.00384) (0.00453) (0.0480)

HH male adults -0.0553*** -0.0354** -0.0163

(0.00877) (0.0149) (0.0182)

R20.258 0.663 0.626

Sargan Test 0.454

First Stage F-Test 21.75

Observations 8.730 8.730 8.730

Households 4.365 4.365

Notes: Robust standard errors in parentheses. * signiﬁcant at 10%; ** signiﬁcant at 5%; *** signiﬁcant at 1%. See text for

variable deﬁnitions.

42

Figure 1: Migration and investment decisions

43

Figure 2: Kernel density estimates for asset indexes

44