returns to returning
TRANSCRIPT
J Popul Econ (2000) 13: 57±79
9992000
Returns to returning
Catherine Y. Co1, Ira N. Gang2, Myeong-Su Yun2
1 Department of Economics, University of Central Florida, Orlando, FL 32816-1400, USA2 Department of Economics, Rutgers University, 75 Hamilton Street, New Brunswick,NJ 08901-1248, USA (Fax: �1-732-932-7416; e-mail: [email protected])
Received: 4 June 1998 / Accepted: 16 April 1999
Abstract. We examine the labor market performance of return migrants usingthe Hungarian Household Panel Survey. Two distinct selection issues areconsidered in the estimation of the earnings equation; we implement a naturalmethod using MLE. The result that there is a ``premium'' to work experienceabroad for women is robust across the models we considered. For men, thereturn to working abroad is not generally signi®cant.
JEL classi®cation: J61, J24, F22
Key words: Return migration, earnings, self-selection
1. Introduction and Background
In return migration, the migrant, after spending some time in a host country,returns to his or her country-of-origin. Some migrants work until retirement inthe host country, and then retire to their home country. Some return to theirhome country and participate in the labor market. This latter group of returnmigrants is the focus of this paper.
Earlier versions of the paper were presented at the Association of Comparative Economic Studies,Chicago, January 1998, the CEPR Conference, ``European Migration: What Do We Know?,''Munich, November 1997, and at a conference on the Consequences of International Migration onDeveloping and Transition Economies, Program in Economic Policy Management, Departmentof Economics, Columbia University, February 27, 1997. The manuscript has bene®tted from dis-cussions with conference participants, and with Alan Barrett, Thomas Bauer, Roger Klein,Christoph Schmidt, Frank Vella and Yoram Weiss. We also thank the two journal referees. Re-sponsible editor: Klaus F. Zimmermann.
In this paper we focus on the relevance of the return migrants' experienceabroad for earnings generation in the home country. We study return migra-tion to Hungary, one of several Central and Eastern European countries un-dergoing transition into a market economy. A second goal of this paper is toimplement a natural method for using maximum likelihood estimation (MLE)when one needs to account for at least two sources of selection bias. Thisimplementation has been introduced earlier by Blank (1990). We presentand employ this implementation in a simple, tractable way, which is easilyreproducible in similarly structured problems.
The theory of return migration generally examines the phenomena as partof life-cycle planning. In this framework, return migration is part of optimaldecision making and is related to the savings behavior of migrants, theirinvestment in human capital acquisition in the host country, and the relativewage di¨erences between the host and home country (see, for example, Djajicand Milbourne 1988; Galor and Stark 1991). In most of the existing modelsreturn migration of individuals is achieved by assuming that the marginalutility of consumption is higher in the home country than in the host country(see Djajic and Milbourne 1988; Dustmann 1997b; Hill 1987; and Stark et al.1997). Alternative motives for return migration are developed by Dustmann(1997a) who assumes that relatively high returns to human capital investmentsmade in the host country are responsible for return migration, and by Stark(1995) who models return migration as the result of employer's learning aboutthe skills of temporary migrants.
The empirical literature has largely studied return migration from the hostcountry perspective, examining the determinants of which migrants leave andwhen, the skills of the return migrants versus those who stay, and host countrypolicies toward the migrants (Borjas and Bratsberg 1996; Dustmann 1996;Schmidt 1994). Recently, Barrett and Trace (1998) and Cohen and Haberfeld(1998) have examined, in detail, the selectivity of return migrants. Bauer andGang (1998) have analyzed the duration of migration abroad.
During the post-1989 transition period there have been a small numberof return migrations by people who left their countries for the West.1 Thesereturn migrants are motivated by either desires to retire in their home country,or to take advantage of new opportunities accorded by Hungary's transitionto a market economy. These return migrations are part of an overall trend ofincreasing international migration movements during the 1980s until about1992 from East to West (or South to North) of migrant workers, asylumseekers and political refugees (See OECD 1992). This trend leveled o¨ around1993 as host countries adopted policies to better control migrant ¯ows. Forexample, some countries tightened requirements for family reuni®cation and/or imposed quotas on immigration (See OECD 1995).
Perhaps of all the transition economies, Hungary is the most prepared totackle the challenges of a market economy. It had committed itself to the es-tablishment of a market-oriented economy (albeit a socialist one) as early as1964. The New Economic Mechanism adopted in 1968 was one of the mostradical programs of reform from Soviet style central planning. It underwentsuccessively more aggressive revisions in the 1980s. Gorbachev's glasnost andperestroika would further encourage the Hungarians to move toward a mar-ket oriented economy in the mid-1980s. However, against this background ofgradual reform, Hungary remained not fully integrated into the global econ-omy partly because of her trade links with the former Soviet block countries.
58 C.Y. Co et al.
In 1986, about 57% of Hungarian exports went to these countries and about52% of her imports were from these countries (see OECD 1991). Hungary'sclose links with the former Soviet block countries prevented her from estab-lishing unrestrained relationships with the West.
Hungary has been moving toward full integration to the global economy.By 1990, Hungarian exports to former Soviet block countries declined toabout 32% and her imports from this set of countries dropped to about 34% ofher total imports (see OECD 1991). There were some 3917 joint ventures(valued at around $900 million) with foreign partners in 1990 alone (seeOECD 1991), with as much as 14%±33% of total investment originating fromforeign sources in 1991±1994 (see Borish and Noel 1996). A movementtoward full integration into the global economy may make it possible forpeople who have been ``exposed'' to the West to receive a wage premium.In this paper we consider ``exposure'' to the West in a way that enables oneto derive a premium in the labor market, as whether the person has workedabroad in the past. Foreign work experience may (i) indicate a person'spossession of the necessary skills that can facilitate trade with the West, (ii)increase the success probability of joint ventures with Western companies, orsimply (iii) demonstrate that they have been exposed to the ``western'' way ofdoing things.2
Two selection issues arise as we investigate whether there is a wage pre-mium to foreign work experience in an economy undergoing transition andintegration to the global economy. First, those who go abroad may be a self-selected group. For example, they may have done better (or worse) during thetransformation phase regardless of whether or not they had gone abroad. Inaddition, we face the standard labor force participation (usually, working ornot working) selection issue in any wage estimation problem. We are inter-ested in ®nding the ``true'' premium to foreign work experience, i.e., the returnto having been abroad not compounded by these selection factors. We handlethese two selection issues and the earnings equation estimation jointly, byimplementing a tractable and easily reproducible maximum likelihood esti-mation.
In the next section we discuss the data that we use in detail. In Sect. 3, wereview the econometric issues and models we consider in this paper. Resultsare analyzed in Sect. 4. Section 5 concludes.
2. Data
We use the Hungarian Household Panel Survey (HHPS), a unique data setcollected by the Social Research Informatics Centre, Budapest University ofEconomics. The ®rst wave of the survey was drawn in 1992 (see Sik (1995) fora description). In 1993 and 1994, a question on whether an individual haslived or worked in a foreign country was included; we draw our sample fromthese two years. We restrict our sample to those individuals who are in theirworking ``life,'' that is, those between 18 and 65 years old in 1994. Out of 3145individuals, 167 were identi®ed as having worked abroad. In this paper, theterms migrant, return migrant and going abroad, are interchangeable.
Table 1 contains the means of the variables used in the analysis, for allobservations and for those who are working. The wage variable is monthlyearnings from a person's main job (natural log of monthly earnings in forints).
Returns to returning 59
Marital status, family status and Budapest at 14 are equal to one for thosewho are married, heads of household and are living in Budapest at age 14,respectively. The size of the family a person belongs to and the number ofchildren less than six in the family are also considered. Training takes a valueof one if an individual has gone through some job-related training in the pastyear. If a person receives bene®ts, such as an o½ce car, medical care, or life/
Table 1. Average values of the variables
Variables Men Women
Not beenabroad
Beenabroad
Not beenabroad
Beenabroad
Sample size, all observations 1387 111 1591 56Age 39.227 42.973* 41.013 40.839Education 10.081 11.694* 10.029 12.750*Marital status 0.645 0.775* 0.647 0.607Family status 0.683 0.856* 0.194 0.304Family size 3.609 3.523 3.417 3.161Children <6 0.259 0.198 0.245 0.179Budapest at 14 0.105 0.306* 0.121 0.286*% working 52.632 68.468* 47.706 64.286**
Sample size, those working 730 76 759 36Age 37.485 41.211* 37.613 39.444Education 10.856 11.829* 11.034 13.306*Marital status 0.722 0.829** 0.669 0.639Family status 0.796 0.882 0.216 0.361Family size 3.659 3.737 3.436 3.167Children <6 0.322 0.197 0.159 0.167Budapest at 14 0.130 0.329* 0.163 0.250Natural Log monthly earnings 9.739 9.930* 9.481 9.883*BudapestNow 0.158 0.263** 0.184 0.417*Years working 19.490 22.211** 19.171 17.472Bene®ts 0.663 0.671 0.706 0.722Training 0.112 0.145 0.141 0.306*Heavy industry 0.181 0.250 0.099 ±Construction 0.066 0.145** 0.018 ±Other industries 0.119 0.039** 0.149 0.083Utilities 0.108 0.053 0.054 0.083Health 0.040 0.026 0.129 0.222School 0.045 0.118* 0.166 0.250State 0.099 0.079 0.099 0.056Other services 0.121 0.105 0.059 0.111Trade 0.116 0.118 0.184 0.194Agriculture 0.105 0.066 0.042 ±Self-employed 0.101 0.079 0.054 ±Non-manual 0.249 0.355 0.462 0.722*FullGovOwned 0.334 0.368 0.408 0.500PartGovOwned 0.152 0.132 0.152 0.083HungarianOwned Firm 0.971 0.895* 0.980 0.917*OECD ± 0.632 ± 0.444Non-OECD countries ± 0.368 ± 0.556
Note: For both males and females, the null hypothesis tested is that the mean of those beenabroad is equal to that of those who have not been abroad. * and ** imply that the null isrejected at the 1% and 5% level of signi®cance, respectively.
60 C.Y. Co et al.
pension insurance, the variable bene®ts takes on the value one. A series ofindustry dummy variables is de®ned. An individual's employment status iscontrolled for by the introduction of dummy variables for when the individualis self-employed and when the job held is non-manual. Information on thenationality of the company a person works for is also available: the variableHungarianOwned has the value one if the individual works for a fully-Hungarian owned company. An individual can work for a company ownedexclusively (FullGovOwned) or partly (PartGovOwned) by the government.We have information on whether the individuals gained their foreign experi-ence in OECD countries (e.g., Austria, Australia, Canada Denmark, England,Germany, Italy, Japan and Sweden, and United States), or in non-OECDcountries (e.g., Africa, Czech Republic, Poland, Romania, and Slovak Re-public).
For each variable we test the null hypothesis that the mean for those whohave been abroad is equal to the mean for those who have not been abroad.Using all observations, for both men and women, those who have beenabroad have signi®cantly more years of education and more of them lived inBudapest when they were 14 years old. Men who have been abroad are sig-ni®cantly older than men who have not been abroad.
Using the sample of those currently employed, for both men and women,those who have been abroad have signi®cantly larger earnings than those whohave not been abroad; further, those who have been abroad are more edu-cated. A larger percentage of those who have been abroad are currently inBudapest (BudapestNow). Though not signi®cant, men's work experience(actual years working) is larger for those who have been abroad. A largerpercentage of men who have been abroad are employed in education-relatedoccupations. Alternatively, a smaller percentage of men who have been abroadwork in construction and other industries (food, textile and other light in-dustries); a smaller percentage of men who have been abroad work for a fully-Hungarian owned business. There are no signi®cant percentage di¨erences inwomen's choice of industry; a signi®cantly larger percentage of those whohave been abroad work in non-manual occupations and a smaller percentagework for fully Hungarian owned ®rms. Finally, relatively more women whohave been abroad have received some form of training in the past year.
Besides comparing the characteristics of those who have been and thosewho have not been abroad by gender, two observations can be made bycomparing men and women. First, there is a clear di¨erence in the type ofindustry men and women enter. Relatively more men are in the state sector,and in heavy and construction industries; while relatively more women are inhealth and school services and trade and personal services (such as ®nancialservices, tourism, etc.).3 Second, men and women di¨er in their choice ofdestination countries. Under half of the women in our sample have been toOECD countries; the rest to non-OECD countries. Men clearly preferredOECD countries ± 63% of them have been to an OECD country.
Migration by women is typically complicated by the fact they mayhave migrated unintentionally with their husbands.4 Our data set allows theidenti®cation of whether the return migrants belong to the same family uponreturning to Hungary.5 Interestingly, about a quarter of women who havebeen abroad are married to men who have been abroad. However, there isonly one working woman who belongs to the same family as a male returnmigrant.
Returns to returning 61
3. Econometric issues and models
Several issues need to be considered in estimating whether experience abroadprovides a ``premium'' for people's earnings after returning to the homecountry. Our view is that the going abroad decision (migration) is an invest-ment in ``experience gathering.'' Re-migration to Hungary provides themigrant an opportunity to reap the bene®ts (if any) of experience abroad.Moreover, the size of the earnings ``premium'' may vary by the host countrythe return migrants have visited.6
We consider a semi-log speci®cation for the earnings equation,
Y � Xb �Da� e; �1�
where Y is the natural-log of monthly earnings, b and a are coe½cientvectors and e is the stochastic term; matrix X includes variables on personalcharacteristics, and matrix D includes dummy variables accounting for for-eign experience.
The estimates of equation (1) from ordinary least square (OLS) may beinconsistent. The inconsistency occurs due to two selections: a workingselection and a migration selection. The selection into working or not is awell-studied problem.7 In addition, those who have been abroad may be more(or less) productive persons regardless of the foreign experience. The errorterm in the earnings equation is related to both the working decision and themigration decision. To address these selection issues, we introduce two indexfunctions,
LFP� � Zg� u �2�
ABROAD� � Qy� z; �3�
where g and y are vectors of coe½cients and u and z are stochastic terms.Equations (2) and (3) are decision functions for working and migration,
respectively. LFP* and ABROAD* are unobservable. Nevertheless, we doobserve the dichotomous variables LFP (LFP � 1 if LFP� > 0, and LFP � 0otherwise) and ABROAD (ABROAD � 1 if ABROAD� > 0, and ABROAD� 0 otherwise).
The e¨ects of decisions of participation and migration on earnings can beestimated using either Heckman's two-step method with double selection(``Heckit'') or maximum likelihood estimation (MLE). Heckit has been widelyused when there is only one selection rule (see Heckman 1979). Though wecan extend Heckit to get 'consistent' estimators in the presence of doubleselection, Heckit becomes very cumbersome when the number of selectionrules is more than one (See Fishe et al. 1981; Ham 1982; Tunali 1986). It isbecause the formulae for the computation of self-selection correction terms(so-called l's) become complicated and the burden of computing correctedstandard error becomes enormous. The burden of computation can be re-lieved by assuming that two selections are not correlated (Fishe et al. 1981).However, this is often too strong an assumption.
MLE is an attractive method when there are double selections (e.g., seeBlank 1990). However, it has only been infrequently employed, perhaps be-cause of these two reasons: 1) researchers have gotten used to the two-step
62 C.Y. Co et al.
procedure and like to see the l's included and interpreted; and 2) since thelikelihood function typically varies from speci®cation to speci®cation, manyresearchers feel more comfortable with a standard approach and form. Herewe o¨er an MLE implementation that is tractable and easily reproduced inproblems with a similar structure. The likelihood function is relatively simplewhen the stochastic terms are assumed to follow a multivariate normal distri-bution. The earnings equation and the two decision functions are estimatedjointly using MLE. The procedure accounts for the possible correlation ofparticipation and migration decisions with earnings. The endogeneity of theparticipation and migration decisions are ignored by OLS (see Heckman(1978) and Mo½tt (1983), p. 1030). The obtained estimators are not onlyconsistent, but also have other desirable properties of MLE (they are asymp-totically e½cient and normally distributed).
The likelihood function is given as follows (The functional speci®cation isgiven in the Appendix.)
L �YP;A
Pr�u > ÿZg; z > ÿQy; e � YÿXb ÿDa�
YP;NA
Pr�u > ÿZg; zUÿQy; e � YÿXb�
YNP;A
Pr�uUÿZg; z > ÿQy�
YNP;NA
Pr�uUÿZg; zUÿQy�;
�4�
where P, NP, A, and NA are labor market participant, labor market non-participant, individual has been abroad, and individual has never beenabroad, respectively. The likelihood function shows the contribution of in-dividuals who are working and have been abroad (P, A), individuals who areworking and have never been abroad (P, NA), individuals who are not work-ing and have been abroad (NP, A), and individuals who are not working andhave never been abroad (NP, NA).
By maximizing the likelihood function, we get estimators of the indexfunctions (participation and migration decision functions, g and y), the earn-ings function (b and a), and variance and correlation coe½cients. (For identi-®cation purposes, the variance of u and z are normalized to 1.) The estimationis implemented using the SAS NonLinear Programming procedure (SAS In-stitute 1997).
Matrix X in equation (1) includes variables on personal characteristics.Whether the person in currently living in Budapest or not is included to ac-count for earnings di¨erential across locations. Typical human capital varia-bles (education, training, and experience) are expected to raise earnings. Weinclude the variable bene®ts to account for the possibility that earnings andadditional job ``quirks'' are substitutes. A series of industry speci®c dummyvariables are also included, with trade and personal services (e.g., ®nancialservices, tourism, etc.) as the reference category. Two dummy variables onemployment status are included to control for whether self-employed in-dividuals or non-manual job workers have higher earnings than the manual
Returns to returning 63
job workers. We include two dummy variables related to ®rm characteristics:whether the ®rm the individual works for is wholly Hungarian owned or not(HungarianOwned), and whether the ®rm is owned exclusively or partly bythe government.
We capture the e¨ects of foreign migration experience on earnings byintroducing dummy variables on migration in matrix D. In one speci®cation,we have only one dummy variable capturing whether an individual has for-eign experience or not. In the other speci®cation, we introduce two dummyvariables to account for the host country: OECD and non-OECD countries.8The reference category in the two speci®cations is not having been abroad.The coe½cients for experience abroad are of great interest. If positive, thenthere is a premium to experience abroad. On the other hand, a negativecoe½cient indicates that the experience gathering had negative e¨ects onperformance in the domestic labor market. If what gathered abroad are``skills,'' a negative e¨ect may imply that skills acquired abroad may be non-transferable, and that time away from the domestic labor market has hurt theworker.
The matrix Z in equation (2) includes, age, educational attainment, maritaland family status, the size of the family and the number of children below 16years old. Age and its square term are included to test the notion that proba-bility of work rises with age only up to a point, and then declines. Investmentsin formal education are made with expectations of higher future earnings, sohigher levels of education should increase the probability of working. Maritaland family status are also controlled for in the work status equation. The co-e½cients for the size of the family and the number of children below 16 yearsold are expected to be positive.
The matrix Q in the abroad equation above includes age, educationalattainment and the locality in which an individual was raised.9 Intuitively,younger individuals are more likely to go abroad and return migrate becausethey potentially have a longer life span during which to reap any returns fromwork experience abroad. However, since we do not know when the decision tomigrate was made, the data only allows us to assess who might have migratedat some point in time. In this context, the coe½cient for age is expected to bepositive since older individuals have had more opportunities to decide tomigrate.10 Budapest at 14 is included to capture di¨erences in the propensityto migrate and return migrate between those living in the capital and thoseliving elsewhere. More educated individuals may have lower moving costssince they can gather information on job availability, etc., much more e½-ciently (see Schwartz 1973). Moreover, education may bring an opportunity togo abroad, and, in some cases, allow for additional years of education forthose who went abroad. However, the additional years of education an indi-vidual may undertake while abroad is dependent on the number of years ofeducation he or she has when the migration decision was made.
4. Analysis of results
OLS estimates for men and women and the results are presented in the ®rstcolumns of Tables 2 and 3, respectively. Here we consider the speci®cation inwhich a single dummy variable captures whether an individual has foreignexperience or not. The results indicate that the di¨erence in earnings between
64 C.Y. Co et al.
Ta
ble
2.
E¨
ects
of
mig
rati
on
on
earn
ings
,m
en
OL
SM
LE
Lo
gea
rnin
gsL
og
earn
ings
Ch
oic
eo
fla
bo
rm
ark
etp
art
icip
ati
on
Ch
oic
eo
fm
igra
tio
n
Co
nst
an
t8
.92
4*
(0.1
35)
9.1
90*
(0.1
55)
ÿ3.6
00*
(0.3
89)
ÿ3.0
26*
(0.2
42)
Ag
e0.1
34*
(0.0
19)
0.0
09*
(0.0
03)
Ag
e2=1
00
ÿ0.2
05*
(0.0
23)
Ma
rita
lst
atu
s0.2
03*
**
(0.1
06)
Fa
mil
yst
atu
s0.8
03*
(0.1
18)
Fa
mil
ysi
zeÿ0
.008
(0.0
32)
Ch
ild
ren<
60.0
10
(0.0
79)
Bu
da
pes
ta
t1
40.5
61*
(0.1
27)
Bu
da
pes
tNo
w0
.15
0*
(0.0
42)
0.1
49*
(0.0
43)
Ed
uca
tio
n0
.06
4*
(0.0
09)
0.0
54*
(0.0
10)
0.1
31*
(0.0
16)
0.1
03*
(0.0
19)
Yea
rsw
ork
ing
0.0
21
*(0
.005)
0.0
11**
(0.0
05)
Yrs
wrk
2=1
00
ÿ0.0
33*
(0.0
11)
ÿ0.0
09
(0.0
12)
Ben
e®ts
0.0
15
(0.0
38)
0.0
12
(0.0
40)
Tra
inin
g0.0
41
(0.0
50)
0.0
32
(0.0
37)
Hea
vy
ind
ust
ry0
.19
3*
(0.0
60)
0.1
88*
(0.0
59)
Co
nst
ruct
ion
0.3
33
*(0
.073)
0.3
31*
(0.0
84)
Oth
erin
du
stri
es0
.14
5*
*(0
.066)
0.1
38**
(0.0
65)
Uti
liti
es0.1
96*
(0.0
68)
0.1
89**
(0.0
75)
Hea
lth
ÿ0.0
79
(0.0
92)
ÿ0.0
73
(0.0
77)
Sch
oo
lÿ0
.02
0(0
.086)
ÿ0.0
19
(0.0
90)
Sta
te0
.07
3(0
.068)
0.0
67
(0.0
66)
Oth
erse
rvic
es0
.16
9*
(0.0
62)
0.1
66**
(0.0
77)
Ag
ricu
ltu
reÿ0
.02
8(0
.066)
ÿ0.0
31
(0.0
66)
Sel
f-em
plo
yed
0.1
60**
(0.0
62)
0.1
50***
(0.0
87)
No
n-m
an
ual
0.1
30*
(0.0
49)
0.1
34*
(0.0
48)
Hu
nga
rian
Ow
ned
ÿ0.3
27*
(0.0
84)
ÿ0.3
27*
(0.0
90)
Fu
llG
ovO
wn
edÿ0
.02
1(0
.038)
ÿ0.0
18
(0.0
37)
Pa
rtG
ovO
wn
ed0
.06
2(0
.047)
0.0
59
(0.0
40)
Returns to returning 65
Ta
ble
2.
(co
nti
nu
ed)
OL
SM
LE
Lo
gea
rnin
gsL
og
earn
ings
Ch
oic
eo
fla
bo
rm
ark
etp
art
icip
ati
on
Ch
oic
eo
fm
igra
tio
n
Ab
roa
d0
.03
8(0
.052)
0.0
11
(0.0
74)
se
0.4
28*
(0.0
20)
rue
ÿ0.3
52*
(0.0
72)
rze
0.0
30
(0.0
47)
ruz
0.0
39
(0.0
75)
Ad
j.R
20
.26
9F
-sta
t1
5.1
34
N8
06
1498
No
te:
Th
en
um
ber
sin
pare
nth
eses
are
the
stan
dard
erro
rs.
*,
**
an
d***
mea
nst
ati
stic
all
ysi
gni®
can
ta
t1%
,5%
an
d10%
resp
ecti
vel
y.
66 C.Y. Co et al.
Ta
ble
3.
E¨
ects
of
mig
rati
on
on
earn
ings
,w
om
en
OL
SM
LE
Lo
gea
rnin
gs
Lo
gea
rnin
gsC
ho
ice
of
lab
or
mark
etp
art
icip
ati
on
Ch
oic
eo
fm
igra
tio
n
Co
nst
an
t8
.74
8*(0
.126)
8.6
63*
(0.1
62)
ÿ4.6
36*
(0.3
67)
ÿ3.8
21*
(0.3
67)
Ag
e0.2
40*
(0.0
21)
0.0
06
(0.0
05)
Ag
e2=1
00
ÿ0.3
38*
(0.0
26)
Ma
rita
lst
atu
s0.2
80*
*(0
.109)
Fa
mil
yst
atu
s0.4
24*
(0.1
29)
Fa
mil
ysi
zeÿ0
.057*
**
(0.0
34)
Ch
ild
ren<
6ÿ0
.712*
(0.0
85)
Bu
da
pes
ta
t1
40.2
62*
**
(0.1
54)
Bu
da
pes
tNo
w0
.12
6*(0
.036)
0.1
22*
(0.0
34)
Ed
uca
tio
n0
.06
2*(0
.008)
0.0
64*
(0.0
09)
0.1
11*
(0.0
15)
0.1
53*
(0.0
27)
Yea
rsw
ork
ing
0.0
21*
(0.0
04)
0.0
25*
(0.0
06)
Yrs
wrk
2=1
00
ÿ0.0
37*
(0.0
11)
ÿ0.0
47*
(0.0
14)
Ben
e®ts
0.0
83*
*(0
.033)
0.0
88*
(0.0
34)
Tra
inin
g0.0
49
(0.0
40)
0.0
50
(0.0
41)
Hea
vy
ind
ust
ryÿ0
.02
5(0
.054)
ÿ0.0
26
(0.0
56)
Co
nst
ruct
ion
ÿ0.0
05
(0.1
06)
ÿ0.0
05
(0.0
89)
Oth
erin
du
stri
esÿ0
.08
6**
*(0
.047)
ÿ0.0
89**
(0.0
45)
Uti
liti
esÿ0
.06
0(0
.066)
ÿ0.0
62
(0.0
70)
Hea
lth
ÿ0.0
97*
**
(0.0
52)
ÿ0.1
02**
(0.0
47)
Sch
oo
lÿ0
.09
7**
*(0
.051)
ÿ0.0
99**
(0.0
48)
Sta
teÿ0
.05
6(0
.055)
ÿ0.0
60
(0.0
57)
Oth
erse
rvic
esÿ0
.07
4(0
.062)
ÿ0.0
73
(0.0
80)
Ag
ricu
ltu
re0
.08
0(0
.074)
0.0
78
(0.0
91)
Sel
f-em
plo
yed
0.0
22*
(0.0
69)
0.0
22**
(0.0
94)
No
n-m
an
ual
0.1
24*
(0.0
39)
0.1
25*
(0.0
40)
Hu
nga
rian
Ow
ned
ÿ0.2
91*
(0.0
91)
ÿ0.2
88*
(0.0
81)
Fu
llG
ovO
wn
edÿ0
.04
2(0
.035)
ÿ0.0
44
(0.0
35)
Pa
rtG
ovO
wn
ed0
.12
5*(0
.042)
0.1
22*
(0.0
43)
Returns to returning 67
Ta
ble
3.
(co
nti
nu
ed)
OL
SM
LE
Lo
gea
rnin
gs
Lo
gea
rnin
gsC
ho
ice
of
lab
or
mark
etp
art
icip
ati
on
Ch
oic
eo
fm
igra
tio
n
Ab
roa
d0
.22
7*(0
.065)
0.3
44*
(0.1
06)
se
0.3
66*
(0.0
16)
rue
0.1
55
(0.1
28)
rze
ÿ0.1
53***
(0.0
92)
ruz
0.0
08
(0.1
03)
Ad
j.R
20
.33
0F
-sta
t1
9.6
46N
79
51647
No
te:
Th
en
um
ber
sin
pare
nth
eses
are
the
stan
dard
erro
rs.
*,
**
an
d***
mea
nst
ati
stic
all
ysi
gni®
can
ta
t1%
,5%
an
d10%
resp
ecti
vel
y.
68 C.Y. Co et al.
men who have been abroad and men who have not been abroad is not statis-tically signi®cant. Women who have been abroad earn about 25% more thanthose who have not been abroad.11 However, the OLS estimates do not mea-sure the true e¨ect of migration on earnings because of the two self-selection issues we discussed in the previous section. We use maximum likeli-hood estimation to ®nd the parameter estimates for the earnings equation.The MLE procedure also provides estimates for the coe½cients of the twoselection functions. These MLE are presented in columns 2 to 4 of Tables 2and 3. Column 2 shows parameters of the earnings function, and columns 3and 4 show coe½cients for the labor market participation and migrationchoice equations.12
For both men and women, increased years of schooling and being inBudapest at age 14 signi®cantly increases the probability of going abroad.The e¨ect of education is larger for women than for men, while being inBudapest at age 14 is larger for men than for women. Age is signi®cant formen but not for women.
The decision to work for men is positively related to age, being married,being the head of the household, and increased schooling. Neither family sizenor number of children is signi®cant. For women, larger family size and morechildren under 6 years of age is negatively related to working. The other re-sults on women are similar to those for men.
Our estimates of the earnings equation using maximum likelihood are di¨-erent from those using OLS. For men, the MLE coe½cient on foreign expe-rience is smaller than the OLS coe½cient. MLE shows that some of thepositive e¨ects of going abroad on earnings in the OLS re¯ect the e¨ect ofself-selection into going abroad. This may be due to unmeasurable personalfactors that make the return migrant's productivity higher (consequentlyearnings are larger). The ®nding implies that those who have been abroadwould earn higher earnings even if they have not been abroad. The coe½cientfor the abroad variable for men is statistically insigni®cant. It is possible thatskills acquired abroad have some wage premium, however, the lost ``contact''of having gone abroad may have caused lower wages because of less chancefor building networks, etc. These two opposing e¨ects may have canceled eachother out.
For women, the MLE coe½cient on having been abroad is larger than theOLS coe½cient. One explanation of the larger MLE coe½cient estimate canbe found from the negative correlation between unobserved characteristics inthe earnings and migration equations in Table 3. Women who have beenabroad may lack some desirable unobserved earnings capabilities, and this isnot captured in the OLS estimates. However, by going abroad they acquireother characteristics (e.g., foreign experience) which the labor market rewardsin the form of higher earnings. The earnings premium for women who haveforeign experience is not only statistically signi®cant at the 1% level but thecoe½cient is economically large. Female return migrants earn a premium of40%.
Though we do not have a de®nitive answer why women who have foreignexperience receive such a large premium, we do not ®nd it surprising either.The period covered by the analysis is during the early stage of the transfor-mation process. While this period was characterized by falling output, incomeand employment (see Kattuman and Redmond 1997), knowledge about how``Western'' economies operate is valuable. There is some anecdotal evidence
Returns to returning 69
that as Hungary moved toward a more open economy, women are, relative tomen, negatively a¨ected. This is because they have more tenuous connectionto their jobs (see Weil 1993), hence are underpaid (relative to men with thesame quali®cations). Women who have not been abroad may have receivedwage cuts; those who have been abroad may have ``skills'' that are moresought after in an economy undergoing transition to a market economy, and,hence, do not experience wage cuts (or experience smaller cuts). The ``pre-mium'' that we observe may have two components: 1) a premium from ``skill''acquired abroad and, 2) women with foreign experience have not su¨-ered from wage cuts during the transition phase.
Regarding the other coe½cient estimates from maximum likelihood esti-mation, both men and women currently living in Budapest earn at least 13%more than those in other locations; for each additional year of education,earnings are at least 5% higher and each additional year of experience trans-lates to at least a 1% premium, ceteris paribus. Not surprisingly, men inconstruction earn 39% more than those in the base industry trade and per-sonal services (e.g., ®nancial services and tourism, etc.); those in utilities andheavy industries earn 20% more. Those in other industries (e.g., food, textile,and other light industries) earn 15% more. On the other hand, women inhealth, school and other industries earn signi®cantly less than those in tradeand personal services. The e¨ects range from 9% to 10%. Everything else thesame, those working for wholly Hungarian owned ®rms earn signi®cantly less± men by 28% and women by 25%. As pointed out in section one, our data aretaken from a period when Hungary embarked on full integration to the globaleconomy and the period is characterized by signi®cant joint ventures withforeign ®rms. Workers in foreign-owned ®rms might have higher skills thatwe, as researchers, cannot observe and/or ®rms with some foreign ownershipmay be willing to pay higher wages to attract the ``best'' in the workforce.
Surprisingly, individuals working for partly government owned ®rms earnabout 12% more than those in fully privately owned ®rms. The ability of thesetypes of enterprises to pay higher wages perhaps indicates some advantage tonot fully privatizing formerly state owned enterprises. Some governmentownership allows a ®rm to take advantage of established partnerships withsimilar ®rms while some private ownership ensures the development of moree½cient operations. It should be noted that Hungary did not embark onmassive privatization of government owned enterprises during the early 1990s.Instead, it followed a gradual approach. It created the ``. . . State PropertyAgency (SPA) to sell ®rms at full value to cash paying buyers, many of themforeign ®rms, either on the stock market, in open auctions, or after negotia-tions with the SPA.'' (Rosser and Rosser 1996, p. 322). This approach, de-manding full value, suggests that those privatized during the period are thosewith potential. An alternative explanation is that partly government owned®rms are paying their employees a premium to discourage the ``best'' onesfrom leaving and joining either the newly set up enterprises or those fully pri-vatized.
To better understand the source of the wage premium for women havinggone abroad, instead of using a binary variable for experience abroad, wedi¨erentiate the host countries by introducing two dummy variables for host``regions.'' Our maxtix D now includes two dummy variables to account fordi¨erences in how alternative host countries a¨ect earnings in Hungary. Thetwo dummies account for when an individual has been to an OECD country
70 C.Y. Co et al.
and for when an individual has been to a non-OECD country. The reference``region'' is not having gone abroad. The likelihood function, equation (4),still has four components; however, the error term in the ®rst component ofthe likelihood function should be adjusted accordingly. Maximizing this ``re-vised'' likelihood function gives us the estimates of all coe½cients in the threeequations simultaneously.13 The results of this estimation are in Tables 4and 5.
From the MLE estimates, we ®nd that the coe½cient estimates are similarto those when host countries are not di¨erentiated. Focusing on the foreignexperience variable, we again ®nd that men earn no wage premium over thosewho have not been abroad. Women who have been to OECD countries earn a67% premium over those who have not been abroad. It is not surprising to®nd that the non-OECD coe½cient is insigni®cant. As we argued in sectionone, in the context of an economy undergoing transformation and full inte-gration to the global economy, foreign work experience in OECD countriesmight have a higher wage premium. This could be a premium for actual skillslearned, knowledge of foreign language or exposure to Western work ethic.Returns to foreign experience are driven by the rewards to women who wentto OECD countries.
5. Conclusion
We address the issue of the returns to foreign experience, asking what di¨er-ence foreign experience makes to the earnings of those return migrants whoenter the labor force once back in Hungary. Using the Hungarian HouseholdPanel Survey, we ®nd that there are large di¨erences in the returns to foreignexperience across gender and among host countries in which the experienceoccurred. Two distinct selection issues are considered in the estimation of theearnings equation. Rather than using Heckman's two-step technique, we im-plement a maximum likelihood estimation. MLE is easy to implement owingto the progress of inexpensive and high speed computing, and allows us jointlyto account for our two selection issues in our earnings estimation. Indeed it issurprising to us that this is not the preferred methodology.
Two factors may be behind the result why foreign experience matters forwomen and not for men. First, there is a clear dichotomy in the types of in-dustry men and women enter. The results suggest that the types of industriesmen enter (e.g., heavy industries and construction) do not o¨er any wagepremium for foreign experience; while the industries women enter are exactlythose industries where foreign experience matters (e.g., ®nancial services).Second, say there is wage premium to having gone abroad, the insigni®cantabroad coe½cient for men suggests that ``lost'' contact through having goneabroad may have resulted in lower wages. These opposing e¨ects lead to aninsigni®cant abroad coe½cient.
Returns to returning 71
Ta
ble
4.
E¨
ects
of
mig
rati
on
on
earn
ings
,m
enw
ith
ho
stco
un
trie
s
OL
SM
LE
Lo
gea
rnin
gsL
og
earn
ings
Ch
oic
eo
fla
bo
rm
ark
etp
art
icip
ati
on
Ch
oic
eo
fm
igra
tio
n
Co
nst
an
t8
.91
9*
(0.1
35)
9.1
88*
(0.1
55)
ÿ3.5
99*
(0.3
89)
ÿ3.0
25*
(0.2
42)
Ag
e0.1
33*
(0.0
19)
0.0
09*
(0.0
03)
Ag
e2=1
00
ÿ0.2
05*
(0.0
23)
Ma
rita
lst
atu
s0.2
03*
**
(0.1
06)
Fa
mil
yst
atu
s0.8
04*
(0.1
18)
Fa
mil
ysi
zeÿ0
.008
(0.0
32)
Ch
ild
ren<
60.0
11
(0.0
79)
Bu
da
pes
ta
t1
40.5
61*
(0.1
27)
Bu
da
pes
tNo
w0
.14
8*
(0.0
42)
0.1
47*
(0.0
44)
Ed
uca
tio
n0
.06
4*
(0.0
09)
0.0
54*
(0.0
10)
0.1
31*
(0.0
16)
0.1
03*
(0.0
19)
Yea
rsw
ork
ing
0.0
21
*(0
.005)
0.0
11**
(0.0
05)
Yrs
wrk
2=1
00
ÿ0.0
33*
(0.0
11)
ÿ0.0
09
(0.0
12)
Ben
e®ts
0.0
15
(0.0
38)
0.0
12
(0.0
40)
Tra
inin
g0.0
42
(0.0
50)
0.0
34
(0.0
37)
Hea
vy
ind
ust
ry0
.19
1*
(0.0
60)
0.1
86*
(0.0
60)
Co
nst
ruct
ion
0.3
30
*(0
.073)
0.3
27*
(0.0
84)
Oth
erin
du
stri
es0
.14
5*
*(0
.066)
0.1
38**
(0.0
65)
Uti
liti
es0.1
95*
(0.0
68)
0.1
88**
(0.0
75)
Hea
lth
ÿ0.0
79
(0.0
92)
ÿ0.0
73
(0.0
77)
Sch
oo
lÿ0
.02
2(0
.086)
ÿ0.0
21
(0.0
90)
Sta
te0
.07
6(0
.068)
0.0
71
(0.0
66)
Oth
erse
rvic
es0
.16
9*
(0.0
62)
0.1
66**
(0.0
77)
Ag
ricu
ltu
reÿ0
.02
6(0
.066)
ÿ0.0
29
(0.0
65)
Sel
f-em
plo
yed
0.1
59**
(0.0
62)
0.1
49***
(0.0
87)
No
n-m
an
ual
0.1
29*
(0.0
49)
0.1
33*
(0.0
48)
Hu
nga
rian
Ow
ned
ÿ0.3
25*
(0.0
84)
ÿ0.3
25*
(0.0
91)
Fu
llG
ovO
wn
edÿ0
.01
9(0
.039)
ÿ0.0
16
(0.0
37)
Pa
rtG
ovO
wn
ed0
.06
4(0
.047)
0.0
61
(0.0
40)
72 C.Y. Co et al.
OE
CD
0.0
74
(0.0
65)
0.0
49
(0.0
96)
No
n-O
EC
Dco
un
trie
sÿ0
.02
2(0
.082)
ÿ0.0
60
(0.1
01)
se
0.4
29*
(0.0
20)
rue
ÿ0.3
56*
(0.0
72)
rze
0.0
33
(0.0
48)
ruz
0.0
38
(0.0
75)
Ad
j.R
20
.26
9F
-sta
t1
4.4
87
N8
06
1498
No
te:
Th
en
um
ber
sin
pare
nth
eses
are
the
stan
dard
erro
rs.
*,
**
an
d***
mea
nst
ati
stic
all
ysi
gni®
can
ta
t1%
,5%
an
d10%
resp
ecti
vel
y.
Returns to returning 73
Ta
ble
5.
E¨
ects
of
mig
rati
on
on
earn
ings
,w
om
enw
ith
ho
stco
un
trie
s
OL
SM
LE
Lo
gea
rnin
gs
Lo
gea
rnin
gsC
ho
ice
of
lab
or
mark
etp
art
icip
ati
on
Ch
oic
eo
fm
igra
tio
n
Co
nst
an
t8
.75
7*(0
.126)
8.6
63*
(0.1
60)
ÿ4.6
34*
(0.3
68)
ÿ3.8
17*
(0.3
65)
Ag
e0.2
39*
(0.0
21)
0.0
06
(0.0
05)
Ag
e2=1
00
ÿ0.3
38*
(0.0
26)
Ma
rita
lst
atu
s0.2
81*
(0.1
09)
Fa
mil
yst
atu
s0.4
25*
(0.1
29)
Fa
mil
ysi
zeÿ0
.057*
**
(0.0
37)
Ch
ild
ren<
6ÿ0
.713*
(0.0
85)
Bu
da
pes
ta
t1
40.2
57*
**
(0.1
53)
Bu
da
pes
tNo
w0
.12
5*(0
.036)
0.1
21*
(0.0
34)
Ed
uca
tio
n0
.06
1*(0
.008)
0.0
63*
(0.0
09)
0.1
11*
(0.0
15)
0.1
53*
(0.0
27)
Yea
rsw
ork
ing
0.0
22*
(0.0
04)
0.0
25*
(0.0
06)
Yrs
wrk
2=1
00
ÿ0.0
38*
(0.0
11)
ÿ0.0
49*
(0.0
14)
Ben
e®ts
0.0
87*
(0.0
33)
0.0
91*
(0.0
33)
Tra
inin
g0.0
48
(0.0
40)
0.0
49
(0.0
42)
Hea
vy
ind
ust
ryÿ0
.02
3(0
.054)
ÿ0.0
24
(0.0
56)
Co
nst
ruct
ion
ÿ0.0
01
(0.1
05)
ÿ0.0
01
(0.0
89)
Oth
erin
du
stri
esÿ0
.08
4**
*(0
.047)
ÿ0.0
87***
(0.0
46)
Uti
liti
esÿ0
.05
6(0
.066)
ÿ0.0
59
(0.0
71)
Hea
lth
ÿ0.0
95*
**
(0.0
52)
ÿ0.1
00**
(0.0
47)
Sch
oo
lÿ0
.09
2**
*(0
.051)
ÿ0.0
94***
(0.0
48)
Sta
teÿ0
.05
3(0
.055)
ÿ0.0
58
(0.0
58)
Oth
erse
rvic
esÿ0
.08
0(0
.062)
ÿ0.0
79
(0.0
77)
Ag
ricu
ltu
re0
.08
2(0
.073)
0.0
80
(0.0
91)
Sel
f-em
plo
yed
0.2
22*
(0.0
69)
0.2
25**
(0.0
94)
No
n-m
an
ual
0.1
23*
(0.0
39)
0.1
24*
(0.0
40)
Hu
nga
rian
Ow
ned
ÿ0.2
98*
(0.0
90)
ÿ0.2
95*
(0.0
77)
Fu
llG
ovO
wn
edÿ0
.03
9(0
.035)
ÿ0.0
41
(0.0
35)
Pa
rtG
ovO
wn
ed0
.12
5*(0
.042)
0.1
22*
(0.0
43)
74 C.Y. Co et al.
OE
CD
0.4
17*
(0.0
95)
0.5
23*
(0.1
50)
No
n-O
EC
Dco
un
trie
s0.0
78
(0.0
85)
0.1
88
(0.1
15)
se
0.3
64*
(0.0
16)
rue
0.1
67
(0.1
31)
rze
ÿ0.1
41
(0.1
00)
ruz
0.0
09
(0.1
03)
Ad
j.R
20
.33
6F
-sta
t1
9.2
51N
79
51647
No
te:
Th
en
um
ber
sin
pare
nth
eses
are
the
stan
dard
erro
rs.
*,
**
an
d***
mea
nst
ati
stic
all
ysi
gni®
can
ta
t1%
,5%
an
d10%
resp
ecti
vel
y.
Returns to returning 75
Appendix
The stochastic terms �u; z; e� follow a joint normal distribution with mean zeroand the following variance-covariance matrix:
S �s2
u suz sue
s2z sze
s2e
26643775;
where s2u and s2
z are normalized to 1.The likelihood function for non-participants is simply the bivariate distri-
bution of u and z with correlation coe½cient ruz. For participants, thelikelihood function can be factored into the conditional distribution of u and zgiven e, and marginal density of e. Conditional means of u and z given e aredenoted as muje and mzje respectively, and computed as follows:
muje � rue
su
see; and
mzje � rze
sz
see;
where rue and rze are correlation coe½cients between u and e, and between zand e.
The conditional variance and covariance of u and z given e are denotedas s2
uje; s2zje and suzje respectively. The correlation coe½cient between u and z
conditional on e is denoted as ruzje. They are computed as follows:
s2uje � s2
u �1ÿ r2ue�;
s2zje � s2
z �1ÿ r2ze�;
suzje � �ruz ÿ ruerze�susz; and
ruzje �suzje
sujeszje:
The likelihood function is given as follows:
L �YP;A
CZg� muje
suje;Qy� mzje
szje; ruzje
� �� 1
se� f YÿXb ÿDa
se
� �Y
P;NA
CZg� muje
suje;ÿQy� mzje
szje;ÿruzje
� �� 1
se� f YÿXb
se
� �Y
NP;A
C ÿZg
su;Qy
sz;ÿruz
� �Y
NP;NA
C ÿZg
su;ÿQy
sz; ruz
� ��4 0�
76 C.Y. Co et al.
where f and C are the standard univariate normal density and the standardbivariate normal distribution function.
Endnotes
1 According to o½cial Hungarian statistics, about 121,000 Hungarians emigrated between 1963to 1988 (see OECD (1992)) and about 19,000 returned to Hungary between 1973±1991 (about34% returned in the last three years alone). Similar re-migrations are observed in Poland and inthe Czech and Slovak Republics.
2 This argument is related to the literature on immigration and trade. For example, Gould(1996) tests and ®nds evidence that immigration increases foreign market information (that is,it decreases the transaction costs of trade), thereby increasing U.S. trade with immigrant sourcecountries. Using a similar argument, it is not unlikely that return migrants can also facilitateHungarian trade with their former host countries.
3 This distribution is not an artifact of the data that we use. It is consistent with data from theInternational Labour Organization that identi®ed women to hold a signi®cant percentage oftrade and personal services jobs (see Hubner et al. 1993).
4 We thank one of the referees for pointing this out.5 Although the return migrants may belong to the same family upon returning, we do not know
if they migrated as a family. In other words, we can identify family membership during thesurvey year, not the formation of families.
6 We would also like to consider how long the return migrants were working abroad and theyear they returned. These would lead to varying periods of re-assimilation and in¯uencethe observed e¨ect of experience abroad on earnings. Unfortunately, the data set does notcontain this information. We also do not have pre-migration information on individuals. Nordo the data inform us about what skills are learned abroad or the particular work experienceabroad.
7 Correction for the bias that arises due to workers' self-selecting themselves into work is stan-dard for women, but not for men. However, the men in our sample have a working rate of54%, while the women have a working rate of 48%. This is very low for prime-aged males, soconsidering the work decision for males here is appropriate.
8 In an earlier version of the paper, we also distinguished between European and non-EuropeanOECD destination countries in addition to non-OECD countries. However, as pointed out bya referee, given the relatively small number of individuals who went to non-European OECDcountries, the data are not rich enough to warrant such distinction. We have therefore madeour categorization into OECD versus non-OECD countries. The results that we present in thenext section are qualitatively similar to our original speci®cation.
9 We excluded marital and family status, family size and the number of children from the mi-gration equation (they are included in the labor market participation equation) because thesepertain to changeable characteristics. For example, a person's marital status during the deci-sion to go abroad may not be the same as we observe from the data in 1994. We do not havepre-migration information on those variables.
10 We thank an anonymous referee for pointing this out.11 These percentages are calculated following Kennedy's (1981) suggestion that the percentage
change in semilog models when the independent variable is a dummy is exp�b ÿ 0:5V�b�� ÿ 1,where b is the estimated coe½cient and V�b� is the variance of b. All dummy variable co-e½cients are converted using this formula and these values are used in the discussion.
12 The MLE results are very robust. We estimate the models with (presented in this paper) andwithout industry dummy variables for the full sample, and with/without industry dummy var-iables for the subsample excluding self-employed. We also estimate these models for those aged25 or more to avoid the e¨ects of education decision. The results of these tests can be obtainedfrom the authors upon request.
13 Ideally, we would like to have index functions accounting for which regions migrants go to.Data limitations prevent this. However, we can di¨erentiate in the earnings part of the likeli-hood function between host regions, while accounting for selection using a generalized ``goneabroad'' speci®cation.
Returns to returning 77
References
Barrett A, Trace F (1998) They're Coming Back: Return Migration into Ireland in the 1990's.Manuscript, Economic and Social Research Institute, Dublin, Ireland
Bauer T, Gang I (1998) Temporary Migrants from Egypt: How Long Do They Stay Abroad?Manuscript, Department of Economics, Rutgers University, New Brunswick NJ. http://www-snde.rutgers.edu/scripts/Rutgers/wp/rutgers-listwp.exe?9811
Blank RM (1990) Are Part-Time Jobs Bad Jobs? In: Burtless G (ed) A Future of Lousy Jobs?Brookings Institution, Washington, DC 123±155
Borjas GJ, Bratsberg B (1996) Who Leaves? The Outmigration of the Foreign-Born. Review ofEconomics and Statistics 78:165±176
Borish M, Noel M (1996) Private Sector Development During Transition: the Visegrad Countries.World Bank Discussion Papers, Washington, DC
Cohen Y, Haberfeld Y (1998) Self-Selection and Return Migration: Returning Israelis from theU.S. during the 1980's. Manuscript, Department of Labor Studies, Tel Aviv University, Israel
Djajic S, Milbourne R (1988) A General Equilibrium Model of Guest-Worker Migration: ASource-Country Perspective. Journal of International Economics 25:335±351
Dustmann C (1996) Return Migration: The European Experience. Economic Policy 22:215±249Dustmann C (1997a) Di¨erences in the Labor Market Behavior Between Temporary and
Permanent Migrant Women. Labour Economics 4:29±46Dustmann C (1997b) Return Migration and the Optimal Migration Duration. Manuscript,
University College, London, EnglandFishe RPH, Trost RP, Lurie PM (1981) Labor Force Earnings and College Choice of Young
Women: An Examination of Selectivity Bias and Comparative Advantage. Economics ofEducation Review 1:169±191
Galor O, Stark O (1991) The Probability of Return Migration, Migrants' Work E¨ort, andMigrants' Performance. Journal of Development Economics 35:339±405
Gould D (1996) Immigrant Links to the Home Country: Implications for Trade, Welfare and FactorReturns. Garland, New York
Ham JC (1982) Estimation of a Labour Supply Model with Censoring Due to Unemployment andUnderemployment. Review of Economic Studies 49:335±354
Heckman J (1978) Dummy Endogenous Variables in a Simultaneous Equation System. Econo-metrica 46:931±959
Heckman J (1979) Sample Selection Bias as a Speci®cation Error. Econometrica 47:153±161Hill JK (1987) Immigrant Decisions Concerning Duration of Stay and Migratory Frequency.
Journal of Development Economics 25:221±234Hubner S, Maier F, Rudolf H (1993) Women's employment in central and eastern Europe: status
and prospects. In: Fisher G, Standing G (eds) Structural Change in Central and EasternEurope: Labour Market and Social Policy Implications. OECD, Paris, 213±240
Kattuman P, Redmond G (1997) Income Inequality in Hungary 1987±1993. Manuscript,University of Durham, England
Kennedy PE (1981) Estimation with Correctly Interpreted Dummy Variables in SemilogarithmicEquations. American Economic Review 71:802
Mo½tt R (1983) An Economic Model of Welfare Stigma. American Economic Review 73:1023±1035
OECD (1991) OECD Economic Surveys: Hungary. OECD, ParisOECD (1992, 1993, 1995) Trends in International Migration. Annual Report. OECD, ParisRosser J, Rosser M (1996) Comparative Economics in a Transforming World Economy. Irwin,
ChicagoSAS Institute (1997) SAS/OR Technical Report: The NLP Procedure. SAS Institute, Cary, NCSchmidt CM (1994) The Country of Origin, Family Structure and Return Migration of
Germany's Guest-Workers. Vierteljahreshefte zur Wirtschaftsforschung 1/2:119±125Schwartz A (1973) Interpreting the e¨ect of distance on migration. Journal of Political Economy
81:1153±1169Sik E (1995) Measuring the Unregistered Economy in Post Communist Transition. Eurosocial
Report, 52, Vienna, AustriaStark O (1995) Return and Dynamics: The Path of Labor Migration when Workers Di¨er in their
Skills and Information is Asymmetric. Scandinavian Journal of Economics 97(1):55±71
78 C.Y. Co et al.
Stark O, Helmenstein C, Yegorov Y (1997) Migrants' Savings, Purchasing Power Parity, and theOptimal Duration of Migration. International Tax and Public Finance 4(3):307±324
Tunali I (1986) A General Structure for Models of Double-Selection and an Application to a JointMigration/Earnings Process with Remigration. Research in Labor Economics 8(B):235±283
Weil G (1993) Economic Reform and Women: A General Framework with Speci®c Referenceto Hungary. In: Moghadam V (ed) Democratic Reform and the Position of Women in Transi-tional Economies. Claredon Press, Oxford, 280±301
Returns to returning 79