Professional Documents
Culture Documents
Benoit Dostie
David E. Sahn
December 2006
Forschungsinstitut
zur Zukunft der Arbeit
Institute for the Study
of Labor
Labor Market Dynamics in Romania
During a Period of Economic Liberalization
Benoit Dostie
HEC Montral, CIRANO, CIRPE
and IZA Bonn
David E. Sahn
Cornell University
IZA
Phone: +49-228-3894-0
Fax: +49-228-3894-180
E-mail: iza@iza.org
Any opinions expressed here are those of the author(s) and not those of the institute. Research
disseminated by IZA may include views on policy, but the institute itself takes no institutional policy
positions.
The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center
and a place of communication between science, politics and business. IZA is an independent nonprofit
company supported by Deutsche Post World Net. The center is associated with the University of Bonn
and offers a stimulating research environment through its research networks, research support, and
visitors and doctoral programs. IZA engages in (i) original and internationally competitive research in
all fields of labor economics, (ii) development of policy concepts, and (iii) dissemination of research
results and concepts to the interested public.
IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion.
Citation of such a paper should account for its provisional character. A revised version may be
available directly from the author.
IZA Discussion Paper No. 2511
December 2006
ABSTRACT
Corresponding author:
Benoit Dostie
Institute of Applied Economics
HEC Montral
3000, chemin de la Cte-Sainte-Catherine
Montral, H3T 2A7
Canada
E-mail: benoit.dostie@hec.ca
*
We would like to thank participants at seminars at the World Bank, (2005) and the IZA-EBRD
Conference on: labor market dynamics, the role of institution and internal labor markets in transition
and emerging market economies, Bologna, Italy (2005) for very useful comments. We especially thank
Francesco Pastore for very detailed suggestions. For copies of the questionnaire, interviewer manual
and information on how to obtain copies of the data sets used in this paper, please see:
http://www.worldbank.org/lsms/country/romania/rm94docs.html
2
1 Introduction
Since the end of the 1980s, Eastern Europe and the former Soviet Union have
well as the allocation of labor were heavily regulated.1 It was only in 1991
that, within a broad based reform package, the government began to liberalize
the labor market by allowing wage scales, hiring and promotion criteria to be
Svejnar (1999) surveys the principal applied labor market studies in the
Central and Eastern European Countries as the countries launched the transi-
tions from central planning to a market economies (see also Boeri and Terrell
(2002)). The study of individual labor force histories can provide important in-
sights into the effect of privatization and restructuring on the labor market. By
and active labor market policies on individuals labor market history, one can
finding a job (for example, Ham, Svejnar, and Terrell (1998) on the Slovak and
Micklewright and Nagy (1999) on Hungary, and Jones and Kato (1997) on
Bulgaria).
In this paper, we take a broader view of the labor market and study transi-
and, in the case of the latter, distinguishing between those receiving and those
der, allowing us to determine whether the patterns and behaviors observed differ
for males and females. This broader perspective is important for several rea-
sons. For example, transitions in and out of self-employment are usually found
economies, and Wu (2002) also finds that rates of entry into self-employment in-
perspective allows us to identify ways in which the social safety net more
fect employment status. This issue is particularly important since Romania, like
and social assistance4 that is likely to have an impact on labor market tran-
sitions. For example, Micklewright and Nagy (1999) in Hungary find that the
most likely way to exit unemployment insurance is not by getting a job but by
Studies of labor market dynamics usually use one of two methods. Duration
models can be used if one knows how much time individuals spend in the labor
market state of interest. This is the approach taken by Micklewright and Nagy
3 Among those not receiving unemployment benefits, most have seen their benefits expire.
Official policy involved the unemployed receiving benefits for nine months after loosing a job
at a level equal to the minimum wage. After this period, supplementary benefits were provided
up to 18 months at a level of 40 percent of the minimum wage (Sahn and Younger (2000)).
4 Romania also has a rather large set of state transfers, beyond unemployment benefits, as
discussed by Sahn and Younger (2000) and Sahn and Gerstle (2001).
4
(1997), Earle and Pauna (1997), Ham, Svejnar and Terrell (1998), and Ham,
Svejnar and Terrell (1999), for example. Since we have no such information,
discrete choice model, a modeling option also chosen by Terrell and Sorm (2000),
Bellmann et al. (1995), Jones and Kato (1997), and Voicu (2005), among others.
the aforementioned papers by taking into account both state dependence and
ket states as explanatory variables and individual specific random effects, two
allow for correlation between different labor market states both across time and
We use three successive years of panel data from a household survey that was
conducted in Romania from 1994 to 1996. It is not typical to study labor market
Survey contains detailed data about labor market activities and various forms
compare our results to those obtained with more traditional labor force surveys.
the data and non-parametric estimation of labor market dynamics in the form
2.1 Data
For decades under totalitarian rule, the National Commission of Statistics con-
ducted a family budget survey. It was not representative of the population, both
and because there was no serious attempt to update the permanent sample of
households included from one year to the next. In the early 1990s, the Romanian
design of the Family Budget Survey. Field testing took place in early 1994, and
the survey officially went into the field in April 1994. The survey was there-
after repeated from 1995 through 1997. Each years sample is nationally and
regionally representative.6
survey ever administrated in the country, and takes place during Romanias
holds randomly selected from all districts of Romania and the city of Bucharest.
bor market activity, public transfers, and a wide range of living standard in-
panel of households that remained in the survey from one year to the next. By
matching individuals within households that were present for two consecutive
years, we were able to construct panels containing labor market information for
viduals between the ages of 15 and 65 who were in the labor force. Students
6 The survey was continued after 1997, but without the Labor Market module.
6
and housewives who report not to be searching for employment are considered
to be out of the labor force. We divide those individuals into three mutually
exclusive labor market states: employees, who are salaried and hourly workers
for both private, and more importantly state-run and operated enterprises, in-
ture; and unemployed. Note that we thus exclude pensioners. Labor market
states frequencies for men and women present in 1994-1995 and then for those
labor force for men, although only around 60 percent for women. The self-
employed represent a far greater share of the working age population among
higher for more recent years. Among men, the increase in the share of self-
employed from the 1994-95 to the 1995-96 panel is greater than for women,
although, the overall share of men who are self-employed remains far smaller
Trends in labor market status can be analyzed in the context of a simple four-
states Markov chain model linking labor market status in different years. To
get a better picture of how the unemployed are faring and how government
7
those that do, and do not receive benefits. The estimates in Tables 2 are average
stable on a year-to-year basis with about 94% of both men and women being
able to keep their job. Among the women losing their job, more entered unem-
the majority of men who lose their job become self-employed. This could be
terim stop for men who lose their job, but less so for women. Also, recall that
are employed, and that self-employment is relatively far less important for men
than women. So the large share of men entering self-employment highlights the
Over half of the men who were unemployed and receiving benefits are no
smaller share find work, although, interestingly, a greater share of them are em-
ployed than self-employed. Among those who were unemployed without ben-
efits, among men, a larger share finds jobs than those who were unemployed
with benefits. This conforms to our initial expectation that after benefits are
depleted, people would become more desperate and stop queuing for rationed
jobs and instead enter self-employment at a higher rate than those still receiving
benefits. However, the effect of not having benefits as an incentive for finding
8
work is less than expected. Interestingly, however, this incentive appears not to
apply to unemployed women without benefits, for whom 44 percent find them-
selves in the same position a year later. Overall, these results suggest that,
Focusing next on the impact of the social safety net on labor market dynam-
ics, we note that over one-third of the men who depart from employment do
not receive benefits in the year following job loss. Those people are not caught
by the social safety net. Moreover, we see that 15 percent of the unemployed
men and 18 percent of the unemployed women who were unemployed with ben-
efits exhaust their benefits without being able to find a regular job or get into
self-employment. Thus, many people are initially caught by the safety net, but
Tables 3 and 4 present summary statistics for our sample divided by labor
market status and gender. Individuals who are employed or self-employed are
older than the unemployed. The age differences are greater for women than
young. As for education, we note that the self-employed have less schooling
than those in the other categories, including the unemployed. Those employed,
have the highest education, with the mean levels being nearly the same for men
and women, 11.5 and 11.6 years, respectively. This mean is four years greater
than women who are self-employed. It is also noteworthy that while 12 percent
of the employed have higher education degrees, this is the case for less than two
percent of the persons in the three other categories, both for men and women.
In contrast, nearly one-third of women, and 23 percent of men who are the self-
employed have less than a high school education. Interestingly, a much smaller
share of men who are unemployed, both with and without benefits, as compared
9
to women, are in the category of having low levels of education. This in part
Also, not surprisingly, we find the share of urban residents among those
employed is greater than the other categories. The difference, however, is par-
ticularly dramatic for women where 76 percent of the employed are urban, in
look at the shares of unemployed with benefits and unemployed without bene-
fits by region, we find a higher share of the latter in urban areas. This would
seem to suggest that the safety net does a better job of reaching the rural unem-
ployed than those in urban areas. While the descriptive findings are of interest,
we next estimate the labor market dynamics using a discrete choice model to
3 Statistical Model
of interest, this would be of limited use if we did not control for other factors
labor market state of each individual during each time period. In this way, we
J
X
yijt = Xit j + lj dyi(t1)l + ijt , i = 1, ..., N, j = 1, ..., J, (1)
l=1
10
variables for individual i at time t, and dyi(t1)l , l = 1, ..., J, are a set of dummy
variables equal to 1 if yi(t1) = l. We assume i1t , ..., iJt are distributed type 1
extreme value so that the usual multinomial logit model results with
J
l dyi(t1)l
P
Xit j +
l=1
Pr[yit = j] = J
J
. (2)
P P y
1+ Xit m + lm di(t1)l
m=2 l=1
taken to be the base category for both past and present labor market states.
Note that in our case, we do not interpret the above probabilities as choices but
each of the labor market states conditional on observed characteristics and past
N
X
L= Li (3)
i=1
with
2 X
X
Li = dij ln Pr[yit = j] (4)
t=1 jCi
where
1
if individual i choose an alternative j
dij = (5)
0
otherwise
The inclusion of past labor market states is done in order to take into account the
individuals labor market history. It is well known that it is more likely that an
11
of order 1. Note also that the previous periods explanatory variables have an
We can also make use of the panel structure of our data set by adding a
random effect to the utility functions defined above. This allows us to take into
where uij are the individual-choice specific random effects. In order to make
the model more tractable, we use the following simplifying assumption7 for uij :
uij = j i (9)
We assume that i is normally distributed with mean zero and variance equal to
1. Note that the load factor j is also set to zero for the reference category. Thus,
the IIA hypothesis by testing the hypothesis that all parameters j are equal
to zero. Since the i are not given, the (unconditional) choice probabilities are
obtained by integrating (7) over all values of uij weighted by the density of
uij :
Z
Li () = Li (u)f (u)du (10)
mality. Since a closed form solution to the integral does not exist, we use
For the first year of our panel, we do not know the previous state. Moreover,
of endogeneity of the lagged values of the labor market status in equation (1).
Note that this is the same as equation (1) except it is obviously impossible
to include lagged values for occupations. We also decompose the error term to
include an individual specific effect in the same way as in equation (8) Note that
we obtain a different set of load factors ICj for the initial conditions, but the
individual specific effect i is the same. We maximize the full likelihood where
we assume that every labor market status and initial states are independent
4 Econometric Results
The base category in the econometric models we estimate are those employed as
wage workers (employed). Coefficient estimates are shown in Tables 6 and 7 for
men and women, respectively. Since the sign of the coefficients are not necessar-
ily the same as the sign of the marginal effects, we also present three additional
tables to better interpret the results. Tables 8 and 9 show predicted probabili-
ties of being in each labor market state computed at the average characteristics
of men and women, respectively, and Table 10 shows predicted transition rates
and simulation on those for both genders at the average characteristics of each
labor market status (simple marginal effect are also available on demand)
Our coefficient estimates in Tables 6 and 7 first and foremost underscore the im-
hypothesis that all load factors are zero is rejected, and we find statistically
significant effects for lagged labor market status. Once we control for observ-
able and unobservable characteristics, our results indicate that it is much more
likely to see both men and women with benefits remaining in that state, or
benefits indicates a slightly higher probability that they will be without benefits
in the next period, both for men and women, than with benefits. Likewise, men
and women who are unemployed without benefits are also far more likely to stay
Among those that were unemployed without benefits, as expected, the prob-
ability of finding themselves receiving benefits is much lower than in the case
14
where the initial state was being unemployed with benefits and transitioning to
than employment among men, while unemployed women are more likely to move
into employment than self-employment; and those unemployed women not re-
ceiving benefits are even less likely to take jobs as self-employed than those not
receiving benefits.
Model results also indicate that those in self-employment are far more likely
to remain in that state than become unemployed with benefits. In the case
of women, they are also considerably less likely to transition into unemploy-
ment without benefits than remain self-employed. Both men and women are
more likely to transition into employment than the unemployed who are more
as seen by the fact that all the dummy variables have negative and significant
coefficients, relative to the left out category of having completed less than mid-
that among women, but not men, more educated individuals are much more
This is shown by the much larger negative coefficients on higher education for
pronounced for men. But the most important finding overall is the importance
the predicted probabilities of being in the four states. In the case of education,
we find that the predicted probability of being in a state other than employed
is nearly zero for those with higher education. Similarly, men with high school
unemployed, although, this is not the case for women, where the predicted
probability of not being employed is around 25 percent. Among those who have
only completed primary school or less the predicted probability of being self-
employed is also much higher for women than men, 0.52 and 0.36, respectively.
Perhaps more troubling is the high predicted probabilities of men and women
being unemployed among those who have completed less than secondary school.
the gender difference between unemployment with and without benefits. Specif-
ically, we observe that women with primary or less education are approximately
one-third more likely to not be receiving benefits than receiving benefits, while
for middle and high school degree holders, their predicted probabilities of re-
ceiving benefits are slightly higher than not receiving benefits. This pattern
Our model results indicate an asymmetrical impact of marriage for men and
employed, but we find the opposite result for women. Not surprisingly, we
also find that it is much less likely that an individual will be engaged in self-
also the case that urban dwellers are less likely to be unemployed. Overall, the
The negative coefficient on age suggests that older individuals are also more
a declining impact of age. Simulating the effects of age on the predicted proba-
bility of working, we find an interesting pattern that for both men and women,
there is a U-shaped relationship between age and being unemployed and self
employed, with the exception of women who are unemployed with no benefits.
Our models also predict that the vast majority of the urban labor force,
the rural labor force. What is quite striking is the low predicted probability of
women in rural areas being employed, 36 percent, versus 78 percent for men.
Finally, we also find that those in the labor force from larger households are
more likely to be unemployed, and even more so, self-employed, relative to being
employed, but the effect is statistically significant for women only. Perhaps this
captures the fact that larger households also tend to have more than one person
in the labor market, who both is more selective in the choice of a job, or, more
employed.
Table 10 reports predicted transition rates from our econometric model. Those
numbers, found in bold text, are computed at the average for each labor market
that are taken from Tables 6 (for men) and 7 (for women). The table also indi-
cates how the transitions rates vary with the different observed characteristics
of the individuals.
17
the number of stayers, especially for women, where the model indicates that
83.9 percent of the women remain as employed, but in fact, this is the case for
94.1 percent of them. Our prediction overshoots the actual number of individu-
als transitioning from employment to all other states, both for men and women.
women, where we predict that 11.2 percent of the women who were employed in
the previous state will be self-employed in the current state, while in fact, this
The model fares much worse in predicting transitions out of the other em-
illustrate, our model predicts that an unemployed male worker with benefits
has a 74.3% chance of being employed in the next period, while the actual tran-
sition rate is just 25.9%. The prediction error is slightly smaller in the case of
women (58.9% (predicted) versus 26.0% (actual)). Put another way, our model
predicts that the difference in the probability of men being employed in the next
17% (91.3% - 74.3%) while the observed difference is 68.2% (94.1% 25.9%).
rently being employed is in the case of men who were previously self-employed,
predicted to be 63.1 percent. In fact, this is the case for only 10.2 percent of
the men. Thus, the difference in the probability of men being employed in the
next period between those that are currently employed and self employed is 28
mated in all cases, both for men and women. Transitions to self-employment
from unemployment, both with and without benefits, are also greatly underes-
timated for men, although, this is not the case for women where the difference
between the actual and predicted transitions to self-employment are quite small.
overestimated, but again, this type of movement is relatively rare for both men
vidual effects. This underlines the importance of the role of unobserved hetero-
geneity and the importance of taking this into account in this type of model.
4.3 Simulations
effects and instead focus on those results. However, there is also an important
limitation of focusing on marginal effects: since they are averaged over all in-
dividuals in the sample, they hide important differences due to the fact that
are markedly different, and often dramatically so, as shown in the descriptive
lowing hypothetical question: how would transition probabilities change for the
If an unemployed male with benefits was given the average education level of
the three more years of education that are found among the employed, their
self-employed women had the additional four years of education that is found
among employed women, they would be 35 percent more likely to transition into
employment.
years younger than employed individuals. Increasing the age of men by five years
raises their probabilities of moving into a job as an employee by 10.4 and 8.9
probability points for the unemployed with and without benefits, respectively.
urban setting is much lower than for employed individuals. If the share of cur-
rently unemployed males was the same as the share of males currently employed
living in urban areas the probability of the currently unemployed moving into an
employed state would increase by 10.9 and 6.3 probability points, respectively,
for men, it should be noted that for women, age and living in an urban set-
employment than education. For example, for women unemployed with bene-
fits, the impact of reaching the education level of employed women raises their
ing into an urban area raises it by 22.0 probability points. We also note the
resides in rural or urban areas for the probability of transitioning into employ-
the present level of 8% to the 76%, which is the probability observed among the
The model predicts that unemployed men with benefits have an 8.9% chance
while losing their benefits. Both those probabilities are higher for women at
the probability of staying unemployed (with benefits) for men (-4.7 relative to
the base probability of 8.9), for women, state dependence and age play the
most important role (- 5.7 and -5.3, respectively, relative to the base probability
are on average four years younger than employed individuals, their predicted
probability of staying in this state is double what it would be if they were older
(6.4 instead of 3.4 for men and 8.2 instead of 4.3 for women). Finally, looking at
not surprisingly, that state dependence plays an important role. But among
the probability of keeping benefits for men (picking up the fact that employed
The simulations also show the particular role state dependence plays for un-
diminish the probability of staying unemployed with no benefits by 4.0 and 5.7
probability points for men and women, respectively. Compared to the base
predicted transition rates of 7.2% and 8.9%, these are economically significant
numbers.
The most important finding is that the model predicts that women are gen-
employment as are men. We also find that individuals living in an urban setting
are much less likely to make this transition, and the negative geographical ef-
fect is particularly high for women. Age and education also play an important
role for both men and women in terms of the probability of transitioning into
self-employment. If, for example, unemployed women without benefits had the
same age and education characteristics of the older and more educated women
who are employed, it would roughly reduce by three quarters the probability of
5 Conclusion
In this paper we evaluate how employment transitions interact with the social
We first compute transition matrices that give a complete picture of the mobil-
ity process between different labor market states, distinguishing the experience
the employment transitions with a dynamic mixed multinomial logit with en-
We find that both unobserved heterogeneity and state dependence are im-
geneity seems to play a much bigger role as observed characteristics, and past
labor market states explain in many cases only one-third or sometimes even
unobserved heterogeneity.
are more likely to live in urban areas. Education is also paramount in terms
of being employed. Among those who lose their job, most transition into being
receiving benefits are more likely to be younger, male, and living in urban
areas, suggesting that the social safety net is functioning better in rural areas
benefits.
Among those that find themselves unemployed, age, gender and education
educated women, for example, are much more likely to exit unemployed for
also note that women are much more likely overall to transition out of un-
note that a large share of those unemployed who do receive benefits, exhaust
Like being unemployed, we find individuals with less education are more
likely to be self-employed than employed. Among women, but not men, the
self-employed also have less education than not just the employed, but the un-
that only the past labor market status has an impact on the current labor
market status. To test this assumption would require a much longer panel or
detailed information about the length of time spend in each labor market status.
Another useful distinction our data does not allow for is the one between formal
and informal work. It is expected transition rates would differ between these
References
[2] Bellmann, L., S. Estrin, and H. Lehmann (1995). The Eastern German la-
24
bor market in transition: Gross flow estimates from panel data. Journal
[4] Earle, J.S. and C. Pauna (1998). Long-term unemployment social assistance
235.
[6] Ham, J. C., J. Svejnar, and K. Terrell (1998). Unemployment and the social
1142.
[7] Ham, J. C., J. Svejnar, and K. Terrell (1999). Womens unemployment dur-
[8] Heckman, James and James Walker. 1990. The relationship between wages
and income and the timing and spacing of births: evidence from Swedish
[9] Jones, D. C. and T. Kato (1997). The nature and determinants of labor
[10] Lubyova, M. and J. van Ours (1997). Unemployment dynamics and the re-
Paper #93 ,Cornell Food and Nutrition Policy Program , Ithaca, NY.
[14] Sahn, D. E. and A. Gerstle (2001). Child allowances and allocative deci-
[16] Svejnar, J. (1999). Labor markets in the transitional Central and East Eu-
ence.
[17] Terrell, K. and V. Sorm (2000). Sectoral restructuring and labor mobility:
[18] Terrell, K. and V. Sorm (1999). Labor market policies and unemployment
60.
[20] Wu, X. (2002). Embracing the market: Entry into self-employment in tran-