1

2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

ECONOMIC TRANSFORMATION
AND THE REVALUATION OF HUMAN
CAPITAL HUNGARY, 19861999
Gbor Kertesi and Jnos Kll

ABSTRACT
The paper analyses skills obsolescence during transition to the market
economy, using individual wage data from Hungary, 19861999. The link
between workers age composition and firms productivity is also examined
using firm-level information. Transition started with the collapse of demand
for unskilled labor and the concomitant improvement in the relative position
of skilled workers. At later stages of the transition, when technological
change gained impetus, general appreciation of skilled labor stopped. Since
1992 the market value of skills acquired under communism has been falling.
Consistent with the wage data, the productivity estimates suggest the
devaluation of skills acquired under communism.

1. INTRODUCTION
Basic measures like enrolment in education or completed school years
suggested, some ten years ago, that Central and Eastern Europes transition to
the market economy would be promoted by a valuable and transferable stock
of human capital. Optimistic and proud references to a highly skilled labor
force were made in government manifestos and the country reports of

The Economics of Skills Obsolescence, Volume 21, pages 235273.

Copyright 2002 by Elsevier Science Ltd.
All rights of reproduction in any form reserved.
ISBN: 0-7623-0960-1

235

236
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

GBOR KERTESI AND JNOS KLL

international organizations. Less contented observers warned, however, that the

distortions of the school system and the peculiarities of work-based skills could
considerably reduce the value of the inheritance.
The socialist system placed strong emphasis on technical skills, as opposed
to business related disciplines, on both secondary and higher levels of education.
On the secondary level simple vocational training for manual occupations had
shares exceeding 50%, while upper secondary education was severely undeveloped (OECD, 1993). Work-based skills were also distorted. A considerable
part of what workers and managers learned by doing was how to deal with
input shortages, how to manage the inconsistencies of plans, how to make
transactions in a sellers market skills that loose their value when the economy
is opened and the forces of the market begin to work. New technologies were
expected to appear overnight, shortly after the liberalization of trade and
capital flows, leaving the older generation less time, compared to their Western
counterparts, for adjustment.
In 2001 the median worker of the region had about 20 years of work
experience. She or he left school at around 1980 and invested ten years in the
acquisition of work-based skills in a state enterprise. Whether these skills can
be adapted to the standards of the Western economy, which itself has led to
the depreciation of many old skills in the last two decades, is a question of
prime importance for the regions future. What is at stake is more than the
damage to growth potential. A disappointed older generation with broken
careers, pushed out of better jobs, paid lower wages would represent an
outright loss to society and put the completion of post-communist transition at
risk.
Most, albeit not all, empirical studies demonstrate that the older generation
was indeed devalued during the transition. In Poland Rutkowski (1996) and
Puhani (1997) presented evidence of falling wage returns to experience in
19871992 and 19921995, respectively. In the former Czechoslovakia Vecernik
(1995), Flanagan (1995), Chase (1997), and Sakova (1998) observed steeply
declining returns in early stages of the transition, as did Steiner and Bellmann
(1995), Krueger and Pischke (1995), and Burda and Schmidt (1997) in the
former GDR.
No decline in returns was detected by Steiner and Wagner (1997) in their
female sub-sample but, as they emphasized, their results were affected by a
growing share of public sector employees within East Germanys female labor
force. Franz and Steiner (1999) estimated falling returns for women but flat
experience-wage profiles for men both before and after the unification.1 A paper
using retrospective data by Munich et al. (1999) observed no change in the
experience-wage profile in the Czech Republic between 1989 and 1996. While

Economic Transformation and the Revaluation of Human Capital

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

237

the survey of Munich et al. is unique in providing longitudinal observations

(the respondents were asked to report their 1989 earnings in 1997) the data are
potentially subject to recall bias and, even if unbiased, are difficult to compare
with other sources for the exclusion of young workers.
Did, then, the older generation lose out during the transition? We believe
that it did, as suggested by the bulk of the literature: studies focusing on the
enterprise sector, and estimating returns to total work experience, unequivocally
suggest losses although they do show differences in magnitudes and time paths.
Some of the seemingly contradictory results can be reconciled with these
findings if specification issues are taken into account. Undoubtedly, some
disturbing details remain those discouraging general statements about the fate
of older generations and calling for further empirical research. In this paper
we seek to clarify some of the unclear details using large samples of Hungarian
workers and firms observed over a long period.2
The paper is organized as follows. Section 2 starts by presenting estimates
of returns to skills. Cross section samples of about 100,000 individuals per
annum in 19861999 are used to analyze relative wages interacting gender,
education and experience. Notes on potential biases are added in Section 3.
The alternative wage models suggest that the benchmark Mincer-type earnings functions estimated without the interaction of education and experience,
and interpreted without an eye on the levels of wages and employment, fail to
capture the complexity of change. The more detailed analysis reveals a minor
decline of returns to experience and a major general increase in the value of
school-based skills, irrespective of vintage, in the early stage of transition
(19891992). After the transformational recession the skills premium of older
workers fell but the appreciation of new skills gained impetus and continued
until recently.
Section 4 argues that the wage evolutions were initially driven by the collapse
of demand for unskilled labor rather than technological renewal. Furthermore,
rising returns to education should be interpreted in relative terms since even
skilled workers employment and real wages decreased with two-digit percentages under the transformational recession. When new technologies actually
appeared and the demand for qualified workers began to rise the appreciation
of skills was restricted to the younger generation.
Section 5 looks for a potential underlying mechanism by analyzing the link
between firms productivity and workers skills composition. Cross-section
ordinary least squares (OLS) estimates suggest a widening productivity gap
between skilled-young and skilled-old workers in large enterprises. The gap
was first observed in foreign-owned firms, but also appeared in the domestic
sector in later stages of the transition.
237

238
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

GBOR KERTESI AND JNOS KLL

Section 6 draws attention to the limits of the OLS model, presents instrumental variable estimates for panels of firms, and discusses alternative
explanations of how skilled-young labor and high productivity can be correlated.
Section 7 concludes.

2. THE EVOLUTION OF RELATIVE WAGES

Throughout the paper we apply three different specifications of earnings
regressions (see Table 1). The benchmark Mincer-type specification
comprises the key variables (schooling and experience) as educational grade
dummies and linear, quadratic, cubic and quartic terms of experience.3
(Specification 1). Specification 2 applies an augmented interactive model
with group dummies for interactions of gender, education and experience.
Finally, we estimate a simplified interactive earnings model (which combines
education and experience in a simplified manner), where the key variables
are defined in exactly the same way as in the underlying firm level productivity model (Specification 3).
We start with the benchmark regressions controlled for a large number of
wage determinants, using data from 1986, 1989 and 19921999. The regression
estimates suggest a marked increase in the wage returns to schooling, from the
onset of transition. Figure 1, depicting the time paths of returns to educational
grades, suggests that wages relative to the primary school grade grew by 25%
in the category of higher education, and about 10% in the case of secondary

Table 1.

Earnings Regressions Used in the Paper.

Specification

Key variables

Controls

1. Benchmark
Mincerian

gender dummy,
educational grade dummies,
exp, exp2, exp3, exp4
26 interactive dummies of
education, experience and
gender
gender dummy,
unskilled, skilled-youngb,
skilled-oldb

occupational grade
productivity,
capital/labor ratio,
firm size,
local unemployment
rate,
industry dummies,
region dummies,
constant

2. Augmented
interactive
3. Simplified
interactivea

a
b

Detailed information

Appendix 2

Appendix 3

Appendix 4

Occupational grade is not included as control.

Skilled: secondary school or college, young: less than the median experience.

Economic Transformation and the Revaluation of Human Capital

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

239

Fig. 1. Returns to Schooling (Base: Primary School), 19861999

(Benchmark Mincerian Model).

school background.4 The value of apprentice-based vocational training did not

change during the transition. Most of these changes took place during the
transformational recession i.e. between 1989 and 1992/1993. The rates of
return to education seem to be stabilized after 1993.
The rise in return to formal education was accompanied by the devaluation
of market experience acquired under socialism. Changes in the rates of return
are measured by the formula yt yt1, with the predictions defined as y = 1 *
exp + 2 * exp2 + 3 * exp3 + 4 * exp4. The predicted change in the returns to
experience is plotted against the years of experience in Fig. 2. Panel a suggests
that the value of labor market experience slightly increased in the last years of
state socialism. The trend reversed in 1989 when experience started to lose its
value, especially in older cohorts of the labor force. The relative premium on
20 (or more) years of labor market experience dropped by 4% in 19891992,
and 7% between 1989 and 1999.
As suggested by panel b of Fig. 2 (separate equations for each educational
group), the obsolescence of experience-based skills was stronger in the educated
part of the labor force, with workers of university or college backgrounds
suffering the largest losses. The experience-related wage premium of a person
239

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

GBOR KERTESI AND JNOS KLL

Fig. 2. Changes in the Predicted Returns to Experience, 19861999, per cent

(Benchmark Mincerian Model).

240

Economic Transformation and the Revaluation of Human Capital

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

241

with a university diploma and 1525 years of experience was lower by about
20 percentage points in 1999 than 1986.
Figure 2b draws attention to the relevance of an interactive model, which
allows for the fact that different educational and experience groups were exposed
to skills obsolescence of greatly varying degree. The interaction effects are
tested with an equation using 25 interactive dummies5 of education, experience
and gender, treating unskilled labor (08 completed classes irrespective of
experience and gender) as the reference category. In order to be able to control
the gender division of the occupational structure, and the profound gender
differentials in the extent of specialized knowledge in the post-primary
education6, Fig. 3 and its underlying equation reports the coefficients for males
and females separately. Regression parameters of the year 1986 are chosen as
uniform (zero value) starting points. Changes in wage returns relative to the
1986 values are plotted in the graphs. Panels ab repeat the results of the
benchmark Mincerian model by gender.
(i) The benchmark regression models (panels ab) hint at the stabilization of
returns to education following 1992/1993, the worst years of the transformation crisis. By contrast, the more precise interactive model (panels ch)
reveals profound differences between young and old cohorts. The experience-related gap of the pre-transitional years (19861989) was in large part
unaffected by 1992/19937, and started to narrow rapidly after 1992/1993,
and even more markedly after 1995 when the economy started to recover
from the post-1989 crisis.
(ii) Workers with college or university backgrounds, both male and female,
improved their position across all experience groups but the value
of education increased at substantially higher rates in the young
cohorts. As a result, the experience-related wage gap between the oldest
and the youngest college cohorts decreased by 2025% by 1999 (panels
gh).
(iii) The returns to secondary school increased in, and only in, the younger
cohorts of men and (particularly) women. The youngest female cohort
managed to improve its position by almost 20%, followed by workers with
610 years of experience with a 15% increase, and older females whose
market skills kept their modest value with no further appreciation. This
kind of imbalance can be a sign of change in the patterns of demand for
non-manual female employees resulting from the expansion of the tertiary
sector.
(iv) Workers who completed vocational training school did not get ahead in
general (neither the females nor the males improved their position relative
241

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

242
GBOR KERTESI AND JNOS KLL

Fig. 3. Changes in Returns Relative to the 1986 returns (Augmented Interactive Model).
i = iti86, it -s are regression parameters of the model (see Appendix 3).

Economic Transformation and the Revaluation of Human Capital

Fig. 3. Continued.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

243

243

244
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

GBOR KERTESI AND JNOS KLL

to the reference category of unqualified workers) but the wages of young

skilled workers grew by about 10% compared to their older counterparts.
Again, we observe that the age-specific changes were taking place after
the transformational recession.
Cross-Section vs. Cohort-Specific Changes
As long as we are concerned with the current market value of past human
capital investments, the cross-section earnings function estimates provide correct
orientation: the coefficients on education and experience tell, roughly, how one
type of labor is evaluated relative to another at a given point in time. The
equally interesting question of how different birth cohorts were affected by the
changes in relative wages does not belong to the main subject of the paper but
certainly deserves a brief note.
A cohorts gains or losses from transition can be approximated using the
formula g = [t +  (G, E, exp + )t (G, E, exp)][t (G, E, exp + )t (G, E,
exp)] where the s measure the regression-adjusted wage advantage of workers
with E level of education, gender G, and exp years of experience over a base
category, in years t and . The first term of g reflects actual change in returns
while the second is a measure of potential change, due to aging, under the naive
expectation that the cross-section profile of returns to human capital remains
unchanged between t and .
In order to be able to calculate g we distinguished between seven experience
cohorts and re-estimated the augmented interactive model for 1989 and 1999,
shifting both calendar time and experience forward by  = 10.8 The youngest
cohort we could follow over the life-cycle entered the labor market between
19841988, the oldest one left school in 19641968. (The calculations therefore do not capture the deteriorating position of all of these cohorts relative to
school leavers graduating after 1989).
The data presented in Appendix 5 indicate neither gains nor losses at
workers with vocational education, and males with secondary school background the deviations fall within a range of five percentage points around
the naive expectation. By contrast, male and female college graduates, and
females with secondary education, who started their career in 198488
managed to improve their position substantially, acquiring gains between 12
and 20 percentage points. Skilled womens gains from transition monotonously decreased with age, and the pattern was similar for males except for
the oldest cohort of college graduates.9 It seems that transition actually
diverted the earnings careers of high-skilled adults from paths characteristic
of the socialist system.

Economic Transformation and the Revaluation of Human Capital

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

245

3. NOTES ON POTENTIAL BIASES

In this section we discuss several sources of bias potentially affecting the wage
estimates, and touch upon factors calling for caution in their interpretation.
These include endogeneity bias inherent in models using monthly rather than
hourly earnings, bias in the pre-transition estimates due to constraints on wage
setting, selectivity bias, ability bias, and difficulties of separating demand and
supply side effects.
Endogeneity Bias.
The Wage Survey covers only full-time employees and provides information
on monthly, as opposed to hourly, earnings. The exclusion of part-time workers,
and potential endogeneity in the determination of monthly earnings, can lead
to biased estimates of returns to skills. As long as we exclude sole-proprietors
and micro-firms the risk of bias seems relatively low because part-time
employment remained infrequent by the end of the transition. In the fourth
quarter of 1998, for instance, 91.8% of the respondents in the Wage Surveys
target population reported a usual weekly worktime of 40 hours or longer;
2.7% indicated a shorter work week; a further 5.5% said worktime was
highly variable but 80% of such respondents actually worked 40 hours or
more in their main jobs on the week preceding the interview. Skill-specific
differentials in the length of the workweek were modest. A relatively high
share of university graduates (7.5%) reported highly variable working time;
highly-educated workers usual weekly worktime was also marginally (24
minutes) longer than the average. These magnitudes suggest that estimates
of returns to higher education based on monthly, rather than hourly, earnings
may be slightly upward-biased.10
Comparability of Pre- and Post-Transition Estimates.
One might argue that under central planning relative pecuniary wages could not
reflect relative scarcities since firms were constrained in wage setting by a
mandatory wage grid and employees were often remunerated in kind. The
argument holds for several socialist economies, particularly Russia, but hardly
applies to Hungary, which abandoned central planning as early as 1968. After the
reform, prohibitive taxes were levied on firms increasing their wage bill and/or
average wage by rates exceeding a centrally set limit, but managers were free to
set relative wages within the enterprise and adjust employment (apart from mass
dismissals). Lower and upper wage limits for broad occupational categories were
recommended by the authorities. The proposed ranges overlapped and job
classification practices were not effectively controlled by the planners.11 Except
245

246
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

GBOR KERTESI AND JNOS KLL

for top managers payment in kind was restricted to meal vouchers, holidays in
company resort houses, and similar benefits of marginal importance.
Selectivity Bias.
Comparisons of wages before and after the transition can be biased by change
in the strength of selectivity bias. Unfortunately we lack data for a proper test
but Appendix 7 presents an admittedly second-best method focusing on some
observable implications of selection bias. Since the start of transition the composition of wage earners has been changing in favor of workers with lower risk
of unemployment and, predictably, higher wages. The data suggest that unskilled
and elderly labor was more strongly affected by this kind of selection than other
groups. Wage models with sample selection (Heckman, 1979) would probably
detect a steeper rise in the wage advantage of skilled and young workers than
ordinary regression.
Ability Bias.
We should try to quantify, in one way or another, the potential changes in
returns to skill components which are not directly observable, that is, not
captured by simple variables like schooling or experience. A possible measurement strategy is to study wage residuals controlled for observed skill
characteristics and different sort of rents that affect the wage distribution. In a
model where these fundamental wage determinants are controlled for the
residual wage dispersion mostly reflects unobservable skill differentials across
individuals. The question is whether these residuals are stable over time. In
lack of longitudinal data there is no way to observe changes in the residuals
directly. An indirect way of measurement proposed by Juhn, Murphy and Pierce
(1993) takes the residual wage percentiles as the units of observation and
measures wages in real terms to ensure comparability over time. Mean residual
real wages are then compared over time, by percentile. This test was completed
in a former paper (Kertesi & Kll, 1997) for the periods 19861989,
19891992, and 19921995. Residual wages were remarkably stable over time,
all over the residual wage distribution12. Stability in the residual wage
distribution encourages us to rely on the measurable proxies of education and
experience in this paper.
Supply-side effects
Absolute outflows from secondary and higher education (Fig. 4) rose substantially between 1986 and 1999. Annual graduations from secondary schools grew
by more than 1/3 and outflows from colleges and universities nearly doubled.13
The concomitant increase of relative wages and the supply of skilled-young

Economic Transformation and the Revaluation of Human Capital

Fig. 4.

Flows of the Educational System, 19702000 (Number of Students).

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

247

247

248
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

GBOR KERTESI AND JNOS KLL

labor leads us to believe that the revaluation of human capital was predominantly
demand-driven.

4. THE BROADER CONTEXT: TWO STAGES

OF THE TRANSITION
The wage equations demonstrated that the value of skills rose between 1986
and 1992/1993 in all experience groups at almost the same rate. In interpreting
the general rise in returns to education during the transformational recession,
one should consider the fact that nearly all groups of the labor force lost in
terms of job opportunities and real wages the appreciation of skills should
be interpreted in relative terms.
Real wages decreased, albeit unevenly, by two digit percentages all along
the wage distribution, except for the 100th percentile, in early years of the
transition. Low wage workers (10th percentile) suffered a 30% loss in
19891995 but the decline at the other end of the wage distribution was large
too (20% in the 90th percentile). Figure 5 gives the percentage changes in net
real wages by percentile over the 19891992, 19891995 and 19951999
periods.

Fig. 5.

Percentage Changes in Net Real Earnings by Percentiles in 19821992,

19891995, 19951999.

Economic Transformation and the Revaluation of Human Capital

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

249

As Table 2 shows the early years were also characterized with large scale job
destruction. Workers with different skills were unevenly affected but the
available figures suggest net job destruction even in the skilled part of the market.
As much as 48% of unskilled jobs disappeared between 19901995 but the
market for skilled LABOR also had to face a 11% contraction in that period.14
The years of general decline were followed by an era of divergent evolutions in 19951999. Skilled jobs were created at a similar rate to those that had
been lost during the transition15, and real wages in the upper range of the wage
distribution started to rise. This was not the case with unskilled jobs. The number
of unskilled jobs stagnated after 1995 and real wages decreased further in the
lower ranges of the wage distribution.
With the passing of the transformational recession, which brought about the
collapse of demand for unskilled labor, substantial changes took place in the
evaluation of human capital. As shown in the previous section, the skills
premium of older workers failed to increase further while the appreciation of
new skills gained an impetus that continued until recently. In Sections 5 and 6
we present evidence suggesting that the widening gap between the value of old
and new skills was consistent with their relative productivity. These differences
gathered importance when the market for skilled labor started to grow and new
technologies appeared in the economy.

5. ESTIMATES OF RELATIVE PRODUCTIVITY

We present a simple model with the aim to test the hypothesis that changes in
technology and markets raised the productivity of younger educated workers
relative to older skilled employees. We estimate productivity equations (derived
from Cobb-Douglas production functions with heterogeneous labor input) of
the form:
Table 2.

Employment by Gender and Education 1990, 1992, 1995, 1999

(Thousands).
Male

Female

Year

unskilled

skilled

1990
1992
1995
1999

1,803
1,358
1,225
1,228

845
860
824
875

unskilled
1,380
929
759
702

skilled

unskilled

skilled

1,055
936
869
1,006

3,190
2,287
1,984
1,930

1,900
1,864
1,693
1,881

Source: Central Statistical Office in: Fazekas (2000), pp. 247249.

249

Total

250
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

GBOR KERTESI AND JNOS KLL

3

log y =  +  i = 2 i log li +  log k + .

(1)

y being firm level productivity (value added per worker), li the share of the
different types of skilled labor within the firm (skilled-young (l2) and skilledold (l3), the base category being the share of unskilled labor: l1)16, k stands for
the capitallabor ratio approximated with the net value of fixed assets per
worker. Parameters of particular interest are the productivity elasticities with
respect to l2 and l3, that is, the differences between skill groups defined on the
basis of educational background and experience. The estimation results and
descriptive statistics are presented in Appendix 6.
In order to see how the parameters change over time we start with crosssection ordinary least squares (OLS) estimates of equation [1]. OLS can produce
biased and/or inconsistent estimates if firm-specific residuals are correlated with
the explanatory variables, or, if the models variables are endogenously
determined. These problems will be addressed in Section 6 where the OLS
results will be compared with panel estimates using instrumental variables (IV)
and alternative explanations will be briefly explored.
Errors in measuring capital may also bias the coefficients. There is little doubt
that the value of assets was imperfectly measured under socialism when no
capital market existed. The coefficients of the capital/labor ratio are most
probably downwards-biased in 1986 and 1989 due to pure measurement error.
(Other coefficients would be affected only if measurement errors were systematically related to productivity or skills composition). By the end of the transition
period the capital market was in place. More than 80% of the large firms in
the sample were private in 1999, more than 50% incorporated. Many of them
went public and those which did not are also interested in presenting meaningful
data on their assets for creditors and investors. Despite problems with the capital
term its inclusion in all equations can be justified in several ways. First,
including ki to some equations and dropping it from others would make our
estimates non-comparable over time. Second, the bias from dropping the capital
term should be predictably larger than the one from measurement error. Third,
a comparison of OLS and IV estimates (Section 6) can signal whether severe
measurement problems exist.
The equations are estimated for a restricted sample of medium-sized and
large firms. Restriction is required because we shift from individual to firmlevel observations. Information on the internal composition of a firms labor
force is available on the basis of a ten% random sample of its employees, so
we have to restrict our attention to enterprises where the number of workers in
the Wage Survey is sufficiently large.17 The sample thus excludes small and
medium size firms but covers the vast majority of firms employing 300 or more

Economic Transformation and the Revaluation of Human Capital

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

251

workers. The response rate of the Wage Survey has been deteriorating since
1992 but this size category was largely unaffected. A comparison with firm
counts published in the statistical yearbooks suggested a response rate of 75.8%
in 1992 and 86.3% in 1998.
Figure 6 shows the time paths of the coefficients estimated for young and
old skilled labor. The productivity yield attributed by the model to skilled-young
labor input was rapidly growing in 19861999 while the productivity of skilledold labor input was declining in 19921999 to a point that in the latter year it
did not differ significantly from the productivity yield of unskilled labor (this
was chosen as the base category).
Before taking these first results as proof, let us examine the productivity
yields in more detail. If the appreciation of new skills is explained by the efficient matching of new technologies and young workers, one would expect
younger employees to be more productive and better paid in firms applying
advanced equipment and new work standards. Ideally, one would like to study
the experience-related differences in productivity and wages by comparing a
modern and an obsolete sector distinguished using firm-level information.
As a second best solution, since no enterprise-level information is available on

Fig. 6. Productivity Elasticities (Estimated Coefficients from Equation 1),

19861999.
Base Category: Unskilled Labor.

251

252
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

GBOR KERTESI AND JNOS KLL

technology, and industrial branches are too heterogeneous for a meaningful

classification, we use foreign ownership to proxy the modern sector.
Majority foreign-owned enterprises constitute a sizeable part of the Hungarian
economy: in 1999 they had a 40.5% share in employment, 51.2% in fixed assets,
61.6% in depreciation, 63.0% in value added and 78.6% in exports, taking all
firms in the Wage Survey one hundred18. Foreign-firm workers operate relatively
new equipment of unusually high value (Table 3). In addition they are required
to adapt to the as-yet unfamiliar Western corporate culture, many of them are
expected to speak foreign languages, and to undertake training courses regularly.19 Exceptions do exist (several foreign enterprises are known to have high
demand for extended worktime and maximum physical effort rather than novel
skills) but there is hardly a better choice than this if one looks for a representative of modernity in the contemporary Hungarian economy.
Re-estimating the productivity equation for domestic and foreign firms
separately yields the results presented in Fig. 7. The estimates cover the period
19921999 during which the percentage of workers employed by foreign firms
rose from 10% to 40%.
In domestic firms we observe that a higher share of skilled workers is
conducive to higher productivity but the productivity yields attributed to the
skilled-young vs. the skilled-old were similar and stable over time, at least until
1996. By contrast foreign-owned firms apparently acquire high productivity
Table 3.

Foreign and Domestic Firms Selected Variables.

Value added/workera
Year

domestic

1992
1993
1994
1995
1996
1997
1998
1999

528
769
1,085
1,555
1,855
2,804
2,602
3,368

foreign F-ratio
1,008
1,540
2,065
3,118
4,942
6,022
6,445
7,529

33.1
63.5
26.8
27.2
12.1
25.2
27.3
28.6

Fixed assets (net)/workera

domestic

foreign

F-ratio

1,202
1,889
2,627
3,150
3,250
4,815
4,033
6,308c

1,151
2,524
2,769
3,955
5,820
7,804
7,766
7,253

0.1n
2.3n
0.1n
0.9n
6.9
7.4
7.9
0.3n

Depreciation rateb
domestic foreign F-ratio
0.113
0.093
0.085
0.096
0.109
0.115
0.134
0.151

0.132
0.109
0.119
0.131
0.141
0.149
0.153
0.181

1.2n
2.3n
20.4
16.7
11.0
19.5
3.8
6.4

Million forints.
Firms reporting a rate higher than one excluded. The number of excluded firms: 1 in 1994, 2 in
1992, 3 in 1996 and 1999.
c
Three domestic firms reported extremely high capital/labor ratios (28 times the average) in this
year. If these firms are omitted the domestic mean becomes 4,410 and F = 10.8.
n
Insignificant at the 0.05 level. The F-ratios test the equality of the means by one-way analysis
of variance.
b

Fig. 7. Productivity Elasticities (Esrtimated Coefficients from Equation 1) by Ownership, 19861999.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

253

253

Base Category: Unskilled Labor.

Economic Transformation and the Revaluation of Human Capital

254
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

GBOR KERTESI AND JNOS KLL

gains by employing young and educated workers. In foreign firms the

contribution of skilled-young workers to the firms productivity was markedly
higher than that of older cohorts throughout 19921999. Higher shares of
skilled-old labor increased the productivity of a foreign firm in 1992 but the
estimated yields of this group were continuously decreasing and even fell below
the level estimated for the base category after 1996.
Foreign firms seem to have played a decisive role in the economy-wide rise
in skilled-young workers productivity. Taking into account the widening
productivity gap between young and old skilled labor in the economy as a
whole, on the one hand, and the aforementioned ownership-specific differentials on the other, one can conclude that the economy-wide increase was a result
of pure compositional change until 1996. Foreign direct investment, and the
more efficient matching of new technologies and new skills in foreign, rather
than domestic, enterprises, was the driving force behind the appreciation of
younger generations for half a decade.
The returns to skills in domestic firms started to follow the foreign-firm
pattern only at the end of the nineties as shown by panel (b) of Fig. 7: during
the three years between 1996 and 1999 the base differential in productivity
yields of the two types of skilled labor almost tripled (growing from 0.14 to
0.40 by 1999).

Fig. 8. Wages at Foreign and Domestic Firms.

Coefficients from the simplified interactive model (Appendix 4).

Economic Transformation and the Revaluation of Human Capital

Fig. 9. The Distribtuion of Skilled Labor by Experience in Foreign and Domestc Firms (1992, 1999).

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

255

255

256
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

GBOR KERTESI AND JNOS KLL

The ownership-specific differentials in productivity are clearly reflected in

relative wages. Consistent with the predictions of the productivity model,
skilled-young workers employed in foreign enterprises have higher wages,
relative to their unskilled and skilled-old colleagues, than their counterparts in
domestic firms.20
High demand for young, educated workers is also reflected by the age composition of the skilled labor force compared across sectors. Foreign firms demand
has been shifted towards the young cohorts since the times of the first observations as shown by the panels of Fig. 9. The size distribution of the labor
force was only slightly different in 1992, skilled workers with 310 years of
experience were modestly over-represented in foreign firms, but the bias for
young skilled labor is apparent from the density functions of 1999. Note that
the cohorts heavily over-represented in foreign enterprises are not school-leavers
but workers with some work experience.

6. ISSUES OF SPECIFICATION
AND INTERPRETATION
The results presented earlier were based on cross-section OLS models which
rely on restrictive assumptions, do not fully utilize the information available in
the database, and relate to an ever-changing population of firms. In this section
we estimate IV panel models, which allow for correlation between residual
productivity and the firms skill composition on the one hand, and endogeneity
on the other. To start with, we write equation (1) in a more general form:
y =  + Xit  +
i + it

(2)

where y is (log) productivity, X the matrix of independent variables (capital

labor ratio, skill shares), and t denotes time. The residuals are composed of a
firm-specific component
i and the standard residual it. When
i and Xit are
correlated the model cannot determine how much of the change in productivity
to assign to  (the true effect of skill) vs. how much to assign to the correlation.
This, and possible endogeneity of the age composition21, calls for the use of
IV methods.
The time-series cross-section sample to be used for estimation ought to be
long enough to provide information on the within effects. At the same time it
has to be short, otherwise the model would hurt pooling restrictions which
became evident from the cross-section results. As a compromise we estimate
equation (2) for the period 19961999 using a balanced panel of 275 enterprises, and an unbalanced panel of 433 firms observed at least twice without a

Economic Transformation and the Revaluation of Human Capital

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

257

gap. The choice of years was motivated by the cross-section estimates

suggesting that the productivity differential between skilled-young workers and
their elderly counterparts became substantial only after 1995.
The time dimension of the panel has implications for the choice of instruments. The time period is too short for instrumenting Xit with lagged dependent
variables. Instead we shall use industry means of the capitallabor ratio and
skill shares (Zit) as instruments for the firm-level variables.22 If both
i and Xit
were strongly industry-specific Zit would clearly be a wrong choice. Therefore
before estimating the IV model we seek to clarify how the correlation between

i and Xit is related to firms industrial affiliation.

The upper block of Table 4 presents the cross-section results for the balanced
panel, and compares the fixed and random effects estimates of (2). The crosssection coefficients follow a similar time path as those presented in Section 5.
The Hausman specification test reveals significant differences between the fixed
and random effects models a typical symptom of correlation between
Table 4.

Estimates for a Balanced Panel of 275 Firms 19961999

(Exogeneous Regressors).

Dependent:
log(value added/worker)
(A) Firm-level variablesa
logk (capital/labor)
logl2 (skilled-young)
logl3 (skilled-old)
constant
aR2

Cross-sections
1996

1997

Panel 19961999
1998

1999

fixed

randomb

0.2219*
0.1893*
0.0296
0.1600
0.3741d

0.2657*
0.3088*
0.0651
0.6230
0.3261

0.3205*
0.3521*
0.0590
0.7890
0.3902

0.2623*
0.4674*
0.0736
1.100
0.4020

0.2664*
0.4714*
0.0283
0.9610
0.4031

0.1342*
0.0477
0.0477
0.5181
0.0296c

(B) Deviation from industry meansa

logk (capital/labor)
0.3420*
logl2 (skilled-young)
0.2632*
logl3 (skilled-old)
0.0422
constant
0.6739
aR2
0.1851

0.3836*
0.3724*
0.1181
0.0390
0.3460

0.3506*
0.3695*
0.0371
0.0444
0.3560

0.3189*
0.3835*
0.0529
0.0249
0.3431

0.3509* 0.3618*
0.0752 0.2168*
0.0851
0.0092
0.1759 0.1631
0.0847c 0.2656d

Hausman specification tests for the equality of the fixed and random effects estimates: (chi-squared):
(A) 119.7 (0.0000), (B) 59.9 (0.0000)
a

Equations (A) use the firm-level variables, (B) use the log deviation of firm level variables from
variable means in the respective two-digit industries.
b
Generalized least squares.
c
Within R2.
d
Overall R2.
*
Significant at 0.01 level.

257

258
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

GBOR KERTESI AND JNOS KLL

firm-specific residuals and explanatory variables.23 To test if the correlation is

explained by industry effects we re-estimate the model using the log deviation
of firm-level variables from industry means on both sides of the equation. As
shown in the lower panel of Table 4 the Hausman test results in a highly
significant chi-squared value in this model too, suggesting that a considerable
part of the correlation between and Xit should be explained by factors other
than industrial affiliation.
Using industry means as instruments, the random effects IV model estimates
higher productivity of young skilled workers in both the balanced and the
unbalanced panels. (Table 5). The elasticities estimated for skilled-young
workers fall close to 0.5 while the coefficients are not significantly different
from zero with their elderly counterparts.24
The cross-section OLS equations undoubtedly simplify reality but their central
qualitative conclusion is supported by the more carefully specified panel model.
The productivity gap hypothesis nevertheless requires further empirical
justification. There are many ways in which a positive correlation can develop
between productivity and share of skilled-young workers. Some of the
alternative scenarios, such as compensation for an increased risk of jobloss25,
can be rejected on the basis of the available evidence but many others remain.
Higher wages paid in efficient firms may actually reward employees for shift
work, loyalty, outstanding ability, or higher work intensity factors we could
not address in the analysis. Errors in measuring capital may also have biased
the results to an unknown extent.
Attempts to prove that a causal link exists between age composition and
productivity can be criticized from another, non-technical point of view:
arguably, the debate over causality is over-particular from a long run perspective. The correlation between high productivity and low share of elderly workers
Table 5.

Instrumental Variable Panel Estimates, 19961999a.

b
logk
(capitallabor
ratio)b

logl2
(skilled-young)

logl3
(skilled-old)

Balanced panel
275 firms, 1,100 obs.

0.2746
(5.3)

0.4759
(2.8)

0.1341

Unbalanced panel
433 firms, 1,495 obs.

0.2344
(5.3)

0.4796
(5.1)

0.1694
(1.1)

Dependent:
log (value added
per worker)

a
b

Swamy-Arora method derived by Baltagi and Chang (1994).

Real value at 1996 prices.

constant

aR2

0.5477

0.3672

1.409

0.3623

(0.5)

Economic Transformation and the Revaluation of Human Capital

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

259

is a robust fact which, irrespective of how it came into being, implies that older
workers are less likely to learn how to use modern technologies. This will
inevitably erode their skills (further), and establish a causal link between
workers age composition and firms efficiency. We believe that the correlation
observed today is more than a statistical artifact resulting from compositional
effects and unobserved selection processes skills obsolescence already had a
major contribution. Future research will be able to explore the process easily
and more convincingly.

6. CONCLUSION
Like most studies on Central and Eastern Europe we found a general rise in
returns to education between 1989 and 1992, when technological change was
minimal, and the forces of the market had just started to work. This, we believe,
was simply a mirror image of the collapse of demand for unskilled labor. When
market institutions were already at work, and modern technologies started to
flow in, the general appreciation of education stopped but the returns to
experience continued to decline. Technological renewal apparently contributed
to the appreciation of young and educated labor in this period. We found that
these workers are paid increasing wages and presented evidence suggesting that
their skills are actually worth more in a modern environment. By contrast,
neither productivity nor wages grew for the older cohorts of educated workers
after 1992. The obsolescence of skills acquired under state socialism had, and
will continue to have, far-reaching implications for transition countries ones
ranging from reduced growth potential to disappointment with market-based
liberal democracy.

NOTES
1. Their choice of controlling the wage equation for tenure may have affected their
pre-unification estimates because labor turnover was particularly low in the GDR, calling
into question the distinction between general and firm-specific experience.
2. An introduction of data sources and descriptive statistics are presented in
Appendix 1.
3. We use higher than second order experience terms in order to be able to follow
cohort specific changes in the earnings profiles over time.
4. For the sake of consistency the charts and the text always refer to the estimated
coefficients () as presented in Appendices 24. The precise measure of the effect of
education (and other dummy variables) on earnings should be expressed as e which
differs from the coefficient at high values of .
5. The 26 groups are defined by combining gender, three levels of education
(vocational, secondary, higher) and four experience groups (05, 610, 1120, 2130
259

260
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

GBOR KERTESI AND JNOS KLL

years). Two additional categories comprise old workers (more than 30 years of experience, without distinguishing by gender and schooling) and unskilled labor (without
distinguishing by gender and experience). The latter category is the group of reference
in the regressions. In the augmented interactive model the same controls are used as in
the benchmark specification.
6. Female workers typically graduate from general secondary schools (37% in 1999)
as opposed to vocational training schools (19%). The opposite holds for male workers
(24 vs. 42%).
7. Exceptions are female college graduates and workers with completed secondary
school. Even in these groups about half of the narrowing of the base years differential
took place after 1992/93.
8. The underlying assumption that the experience-earnings profiles of 1989 properly
approximated the expected life-cycle paths seems justified in view of the stability of
relative wages under socialism.
9. The relatively high gains of this group may be explained by early retirement of
many high-educated males, aged already 4550 at the start of transition. A precise
clarification of the reasons would require a selectivity bias correction model.
10. Authors calculations based on the Labor Force Survey, available on request.
11. Cukor (1989) provides ample evidence of how firms reclassified their workers in
response to labor shortages between 1973 and 1976.
12. The only exception was the 100th percentile where there was an upward jump
in the mean residual real wage between 1989 and 1992. We suspect that the spectacular wage rise of the best-paid employees in this period represented a sort of
privatization gain, which is an exceptional event from the point of view of the present
paper.
13. Outflows from vocational and secondary schools were also affected by demography. The share of secondary schools in total outflows increased sharply but the cohorts
reaching graduation age (1718 years) were smaller and smaller in subsequent years of
the transition period. In evaluating the magnitudes of possible supply side effects the
absolute numbers in graduations are relevant.
14. Consistent time series on employment are available only from 1990 therefore we
cannot have 1989 as a starting date.
15. The net skilled job creation rate was 11% in 19951999.
16. Skilled means completed secondary school or incomplete or completed college
or university, unskilled means incomplete or completed primary or incomplete
secondary school. Young means experience less than the median experience, old
means median experience or more.
17. The critical sample size on the level of a particular firm was set at 30 workers.
By this restriction firms with less than 300 workers are excluded. Those employing
slightly more than 300 are more likely to drop out than larger firms. To reduce the
resulting bias the computations in the restricted sample were weighted. Weights were
defined as the ratio of firms in the original and the restricted samples within groups
formed by firm size and ownership.
18. A firm was defined majority foreign in case foreigners share in equity exceeded
50%.
19. According to a recent survey comprising 264 domestic and 78 foreign-owned
firms the former spent 2.4% of their total investment on training while the latter spent
14.2% in 2000. A difference of similar magnitude was observed in the Czech Republic

Economic Transformation and the Revaluation of Human Capital

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

261

by Filer et al. (1995) For an introduction of the Hungarian survey see EBRD (2000).
The quoted figures have been calculated from the original data file.
20. These results stem from the simplified interactive regression models presented in
Appendix 4.
21. Capital-abundant modern firms are likely to start businesses in highly profitable
sectors, can afford to pay high wages, and attract job seekers who tend to be younger
and better educated than the average incumbent employee. Firm creation can lead to
spurious correlation between age and productivity in this way.
22. The annual industry means were calculated for 42 industries using data of all
firms observed in the Wage Survey excluding those in the panels. Monetary aggregates
were discounted using producer price indices for three, two and one digit industries.
23. The estimates also suggest that the link between share of skilled-young labor and
productivity is dominated by the between effect, that is, the linkage between intertemporal firm-specific means of the variables.
24. Note that the IV coefficients are different from those of the non-instrumented
panel model. (Compare the estimates for the balanced panel in Tables 4 and 5). The
equality of the key parameters of k, l1, and l2 is rejected at conventional levels of significance.
25. The aforementioned survey of the EBRD suggested lower risk of dismissal in
foreign firms (average annual rate of 3.4% in 19971999 as opposed to 6.9% in domestic
enterprises).

ACKNOWLEDGMENTS
The authors are grateful for comments by Gbor Krsi (Institute of Economics
and CEU, Budapest); an anonymous referee; and participants of the conference
Understanding Skills Obsolescence: Theoretical Innovations and Empirical
Applications organized by the Research Center on Skills, Knowledge and
Organizational Performance (SKOPE), University of Oxford, Department of
Economics and the Research Center for Education and the Labor Market (ROA),
Maastricht University, May 1112, 2001.

REFERENCES
Baltagi, B. H., & Chang, Y. (1994). Incomplete panels: A comparative study of alternative estimators for the unbalanced one-way error component regression model. Journal of
Econometrics, 62, 6789.
Burda, M., & Schmidt, C. M. (1997). Getting behind the East-West German wage differential.
Discussion Paper No. 250. University Heidelberg.
Chase, S. R. (1997). Markets for communist human capital: Returns to education and experience
in the Czech Republic and Slovakia. The Davidson Institute Working Paper Series, No. 109,
Ann Arbor.
Cukor, E. (1989). Wage differentials among blue collar workers, (Kereseti klnbsgek a fizikai
dolgozk krben). Kzgazdasgi Szemle, 36(3), 346362.
EBRD (2000). Transition report 2000 (Chapter 5). EBRD, London.

261

262
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

GBOR KERTESI AND JNOS KLL

Fazekas, K. (Ed.) (2000). LABOR market report, 2000 (Munkaerpiaci Tkr, 2000), Institute of
Economics, National Employment Fund, Budapest.
Filer, R., Schneider, O., & Svejnar, J. (1995). Wage and non-wage LABOR cost in the Czech
Republic: The impact of fringe benefits. CERGE-EI, Working Paper Series 77, Prague.
Flanagan, R. J. (1995). Wage structures in the transition of the Czech economy. IMF Working
Paper WP/95/36, March.
Franz, W., & Steiner, V. (1999). Wages in the East German transition process Facts and explanations. ZEW Discussion Paper No. 9940, Mannheim.
Juhn, C., Murphy, K. M., & Pierce, B. (1993). Wage inequality and the rise in return to skill.
Journal of Political Economy, 101(3), 410442.
Kertesi, G., & Kll, J. (1997). Real wages and earning inequalities: 19861996 (Relbrek s
kereseti egyenltlensgek). Kzgazdasgi Szemle, 44(78), 612634.
Kertesi, G., & Kll, J. (1999). Economic transformation and the return to human capital. Budapest
Working Papers on the Labor Market, 1999/1996. Institute of Economics, Budapest
University of Economics, Budapest.
Krueger, A. B., & Pischke, J. S. (1995). A comparative analysis of East and West German labor
markets before and after unification. In: R. Freeman & L. Katz (Eds.), Differences and
Changes in Wage Structures (pp. 405445). Chicago: University of Chicago Press.
Munich, D., Svejnar, J., & Terrell, K. (1999). Returns to human capital under the communist wage
grid and during the transition to a market economy. CERGE-IE Discussion Paper No.
199929 (November), CERGE-IE, Prague.
OECD (1993). Education at glance. Paris: OECD.
Puhani, P. (1997). All quiet on the wage front in Poland? ZEW Mannheim, mimeo.
Rutkowski, J. (1996). High skills pay off: The changing wage structure during economic transition in Poland. Economics of Transition, 4(1), 89112.
Sakova, S. (1998). Changes and differences in earnings structures, unpublished thesis, Central
European University, Economics Department, Budapest.
Steiner, V., & Bellmann, L. (1995). The East-German wage structure in the transition to a market
economy. Labor, 9(3), 539560.
Steiner, V., & Wagner, K. (1997). East-West German wage convergence How far have we got?
ZEW Discussion Paper No. 9725, Mannheim.
Vecernik, J. (1995). Changing earnings distribution in the Czech Republic: Survey evidence from
19881994. Economics of Transition, 3(3), 355371.

APPENDIX 1: THE WAGE SURVEY

The National Labor Centers Wage Survey (1986, 1989, 19921999) covers
representative samples of firms employing 20 or more workers (10 or more
workers in 19951999) and 10% random samples of their full-time employees.
Descriptive statistics of the estimation sample are shown in Table A1.
The cases are weighted to ensure representativity. An individual weight (w1)
stands for the number of workers represented by a respondent given the sampling
quota within his/her firm. An enterprise-level weight, added by the authors of
this paper, is intended to correct the bias from firm-level non-response.
Comparing the composition of the target population and the sample by firm

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
Variables

1986

263

log(gross earnings/m):
mean
8,74
st. dev.
0,37
Gender:
male
60,07
female
39,93
Education:
primary
48,65
vocational
24,41
secondary
21,31
college
5,63
Experience:
mean
22,22
st. dev.
11,63
log(value added/w):
mean
1,65
st. dev.
0,61
log(net fixed assets/w):
mean
1,36
st. dev.
0,89
Firm size:
1120

2150
0,48
51300
11,23
3011,000
29,35
1,0013,000
27,17
3,001 +
31,77
3,244
116,205

1989

1992

1993

1994

1995

1996

1997

1998

1999

9,18
0,45

9,84
0,51

10,02
0,52

10,27
0,56

10,40
0,55

10,55
0,57

10,75
0,61

10,93
0,63

11,07
0,62

61,60
38,40

60,11
39,89

58,96
41,04

58,48
41,52

59,99
40,01

61,99
38,01

62,92
37,08

60,72
39,28

61,30
38,70

44,45
27,94
22,17
5,43

32,03
33,45
27,07
7,45

31,85
34,70
27,06
6,39

26,08
32,74
31,33
9,85

25,03
33,82
31,33
9,82

26,18
38,91
27,17
7,74

24,96
37,07
29,14
8,83

23,65
37,19
29,48
9,68

22,75
37,99
29,94
9,32

22,68
11,42

22,02
10,58

21,90
10,63

21,84
10,42

21,73
10,60

21,52
10,65

21,28
10,69

21,49
10,69

21,88
10,95

1,13
0,64

0,75
0,82

0,48
0,76

0,18
0,81

0,06
0,80

0,32
0,79

0,48
0,89

0,66
0,92

0,78
0,92

0,99
0,93

0,50
1,27

0,33
1,29

0,39
2,40

0,00
1,40

0,11
1,39

0,32
1,49

0,38
1,63

0,54
1,46


1,85
14,14
28,72
28,04
27,25


6,20
30,00
26,36
16,86
20,58


8,39
36,32
24,90
14,97
15,41


9,93
36,80
25,72
12,38
15,17

7,34
11,25
35,46
19,85
11,12
14,98

8,13
10,99
38,48
17,77
12,13
12,50

9,59
13,51
25,55
19,57
15,35
16,43

7,62
11,90
30,77
21,05
14,50
14,16

9,31
8,93
29,90
22,30
13,40
16,17

4,082
111,293

6,498
86,935

7,243
85,833

7,776
94,639

7,456
90,717

7,796
97,918

7,922
88,208

7,909
102,102

8,541
102,547

263

No. of firms
No. of workers

The Estimation Sample.

Economic Transformation and the Revaluation of Human Capital

Table A1.

264
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

GBOR KERTESI AND JNOS KLL

size and two-digit industry weights were attached (w2) to each size-industry
cell. Calculations were made using the compound weights (w1 w2).
Highest educational attainment can be primary school (8 years in school),
vocational (11 years, provides no certificate required at entry to higher
education), secondary (12 years), college (16 or 17 years depending on the type
of institution). Experience was approximated on the basis of age and the level
of education (age years in school 6). Unemployment was measured at the
level of 170 labor office districts. The earnings figures comprise all work-related
payments made by the enterprise in May and 1/12 of the premia, bonuses and
rewards paid in the preceding year. All statistics refer to pre-tax earnings.

Economic Transformation and the Revaluation of Human Capital

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

265

APPENDIX 2: BENCHMARK MINCERIAN EARNINGS

MODEL (19861999)
Table A2.

Estimation results.

Dependent: log of gross monthly earnings

Independent variables

1986

Constant
8,0504
Gender:
Male
0,2838
Schooling:
Vocational training school
0,1203
Secondary school
0,1359
College
0,3592
Experience:
Linear
0,0522
Quadratic/100
0,2187
Cubic/10000
0,4654
Quartic/1000000
0,4149
Occupation:
Non-manual
0,0957
Managerial
0,5436
Productivity:
log(value added/worker)
0,0565
Negative value added
0,0601
Capital/labor ratio:
log(net fixed assets/worker) 0,0350
Firm size: (No. of workers)
1020

2150
0,0005n
3011,000
0,0312
1,0013,000
0,0502
3,001 +
0,0772
Unemployment:
log(unemployment rate)

Industry:
49 dummies, F-test
108,25
Region:
15 dummies, F-test
131,47
No. of observations
Adjusted R2
F- test
Heteroscedasticity
Omitted variables
Normality of residuals

116,205
0,4588
1,033,11
710,191
148,08
1,579,83

1989

1992

1993

1994

8,4343

8,9185

9,0968

9,1676

0,2995

0,2234

0,2317

0,2378

0,1157
0,1460
0,4410

0,1339
0,2197
0,5597

0,1323
0,2308
0,5971

0,1288
0,2194
0,5981

0,0514

0,0538

0,0482

0,0481

0,2034

0,2554

0,2226

0,2285

0,4130

0,6259

0,5416

0,5760

0,3579

0,6000

0,5182

0,5569

0,1757
0,8700

0,2218
0,7505

0,2480
0,7039

0,2451
0,8334

0,0808

0,1177

0,1615

0,1270

0,0681

0,1099

0,0903

0,1068

0,0225

0,0200

0,0168

0,0131

0,0142n
0,0478
0,0764
0,0900

0,0546

0,0669

0,0382

0,0404
0,0716
0,1294

0,0582
0,1059
0,1558

0,1294
0,1822
0,2318

0,0553

0,0714

0,0811

105,97

93,46

56,03

80,64

152,39

57,39

35,78

30,55

86,935
0,5275
940,88
574,001
269,20
5,397,33

85,833
0,5058
825,08
858,773
234,17
7,523,75

94,639
0,5324
898,97
702,04
122,85
3,833,55

111,293
0,4639
881,07
848,551
251,54
3,259,07

265

266
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

GBOR KERTESI AND JNOS KLL

Table A2.
Independent variables

Continued.

1995

1996

1997

1998

1999

Constant
9,4365
Gender:
Male
0,2196
Schooling:
Vocational training school 0,1108
Secondary school
0,1882
College
0,5461
Experience:
Linear
0,0454
Quadratic/100
0,2187
Cubic/10000
0,5394
0,4985
Quartic/1000000
Occupation:
Non-manual
0,2142
Managerial
0,7453
Productivity:
log(value worker/worker)
0,1698
Negative value added

Capital/labor ratio:
log(net fixed assets/worker) 0,0174
Firm size: (No. of workers)
1020
0,2114
2150
0,1142
3011,000
0,0795
1,0013,000
0,1176
3,001 +
0,1673
Unemployment:
log(unemployment rate)
0,0839
Industry:
49 dummies, F-test
68,61
Region:
15 dummies, F-test
15,13

9,5614

9,7322

9,7852

9,6720

0,2025

0,1929

0,1921

0,2088

0,1303
0,1999
0,6004

0,1297
0,2101
0,6263

0,1220
0,2154
0,6342

0,1183
0,2069
0,6272

No. of observations
Adjusted R2
F test
Heteroscedasticity
Omitted variables
Normality of residuals

90,717
0,5201
835,44
3,551,24
235,60
3,547,12

0,0549

0,0593

0,0646

0,0636

0,3004

0,3262

0,3883

0,3834

0,8134

0,8705

1,0912

1,0734

0,8038

0,8466

1,1013

1,0806

0,2389
0,8461

0,2281
0,7979

0,2334
0,8291

0,2491
0,8021

0,2170

0,1962
0,0319**

0,2256
0,0543

0,2301
0,1064

0,0243

0,0072

0,0068

0,2506

0,2868

0,2926

0,2749

0,1408

0,1692

0,1808

0,1901

0,0784
0,1085
0,1397

0,0796
0,1405
0,1202

0,0981
0,1710
0,1545

0,1242
0,1884
0,2179

0,0843

0,0725

0,0831

0,0970

68,61

69,81

69,59

71,34

15,13

16,27

18,90

14,90

97,918
0,5472
917,92
5,688,65
178,08
3,494,94

88,208
0,5591
906,14
3,435,75
174,96
3,515,77

102,102
0,5710
1,192,02
4,923,05
258,54
3,607,09

102,547
0,5771
1,243,58
4,894,68
277,51
4,534,56

0,0995

0,0058*

Unmarked coefficients are significant at 0,0001. Significant at * 0,001 ** 0,05 n not significant.
Notes: OLS regressions with heteroscedasticity corrected standard errors.
Base categories: women; completed or incomplete primary school; manual workers; firms with
51300 workers.

Economic Transformation and the Revaluation of Human Capital

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

267

APPENDIX 3: AUGMENTED INTERACTIVE EARNINGS

MODEL (19861999)
Table A3.

Estimations Results.

Dependent: log of gross monthly earnings

Interactive variables

1986

1989

1992

1993

1994

education experience (years)

vocational

secondary

college

vocational

secondary

college

05
610
1120
2130
05
610
1120
2130
05
610
1120
2130

0,1004

05
610
1120
2130
05
610
1120
2130
05
610
1120
2130

0,2351

male and female,

experience > 3 0
No. of observations
Adjusted R2
F test
Heteroscedasticity
Omitted variables

0,1031
0,2088
0,2581
0,1384
0,0997
0,2256
0,3300
0,0112n
0,2553
0,4315
0,5305

MALES
0,1313
0,1009
0,1984
0,2555
0,1456
0,0801
0,2285
0,3448
0,0454***
0,2676
0,4766
0,6014

FEMALES
0,2615
0,1433
0,1682
0,0756
0,1049
0,0078n
0,0198n
0,3709
0,3900
0,2080
0,2571
0,0685
0,0983
0,0805
0,0598
0,0698
0,0927
0,1155
0,1855
0,2808
0,3570
0,4246
0,4782

0,1136

0,0747

0,0665

0,0909
0,1902
0,2498
0,0681**
0,1218
0,2655
0,3607
0,1572
0,3834
0,5579
0,6696

0,0878
0,1951
0,2516
0,0469
0,1280
0,2846
0,3699
0,2406
0,4528
0,5519
0,6886

0,1169
0,1972
0,2492
0,0422**
0,1270
0,2635
0,3600
0,2295
0,4390
0,5619
0,6894

0,2085

0,1969

0,2339

0,1112

0,1244

0,1719

0,0583

0,0677

0,0900

0,0249**
0,2374
0,1052
0,0174**
0,1224
0,1834
0,2896
0,4373
0,5436

0,0263**
0,2243
0,0997
0,0168***
0,1290
0,2101
0,3424
0,4980
0,6070

0,0015n
0,2432
0,1215
0,0344

0,0899
0,1894
0,3419
0,4539
0,6159

0,2892

0,2879

0,3135

0,3200

0,3403

116,205
0,3647
649,39
187,121
52,101

111,293
0,3998
576,21
234,891
29,371

86,935
0,5023
617,41
448,511
59,931

85,833
0,4802
616,97
628,163
85,28

94,639
0,5106
696,18
481,45
80,83

267

268
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

GBOR KERTESI AND JNOS KLL

Table A3.
Interactive variables

1995

Continued.
1996

1997

1998

1999

education experience (years)

05
610
1120
2130
05
610
1120
2130
05
610
1120
2130

0,0745

05
610
1120
2130
secondary
05
610
1120
2130
college
05
610
1120
2130
male and female,
experience > 30

0,1985

0,2997

No. of observations
Adjusted R2
F- test
Heteroscedasticity
Omitted variables

90,717
0,4999
651,44
3,474,02
182,43

vocational

secondary

college

vocational

0,0771
0,1597
0,2199
0,0788
0,1060
0,2233
0,3217
0,1785
0,4097
0,5130
0,6238

0,1142
0,0777
0,0114n
0,2363
0,1231
0,0454

0,0592
0,2001
0,3774
0,4149
0,4988

MALES
0,0605
0,0713
0,1663
0,2309
0,0893
0,1159
0,2291
0,3205
0,1956
0,4736
0,5564
0,6317

0,0577

0,0559

0,0432

0,0564
0,1674
0,2353
0,0816
0,1087
0,2361
0,3301
0,2748
0,5312
0,5750
0,6691

0,0816
0,1668
0,2292
0,0622***
0,1159
0,2507
0,3275
0,2775
0,5273
0,6220
0,6511

0,0808
0,1716
0,2251
0,0306
0,1287
0,2600
0,3309
0,3072
0,5777
0,6214
0,6842

0,1607

0,1513

0,1351

0,1014

0,1065

0,0716

0,0660

0,0655

0,0629

0,0052n
0,2282
0,0693
0,0190+
0,0934
0,2182
0,4161
0,4513
0,5942

0,0076n
0,1879
0,0540
0,0268*
0,0977
0,2345
0,4646
0,5025
0,5546

0,3061

0,3057

0,2907

0,2817

97,918
0,5303
716,36
5,645,25
138,21

88,208
0,5445
737,18
3,357,66
156,04

102,102
0,5570
948,85
4,934,77
237,19

102,547
0,5622
1,006,96
4,909,06
236,07

FEMALES
0,1803
0,1184
0,0718
0,0148n
0,2397
0,0982
0,0556
0,0783
0,1725
0,3679
0,4348
0,5634

0,0077n
0,1857
0,0587
0,0345

0,0815
0,2299
0,4584
0,4735
0,5268

Unmarked coefficients are significant at 0,0001.

Significant at * 0,001 ** 0,01 *** 0,05 + 0,1 n not significant.
Notes: OLS regressions with heteroscedasticity corrected standard errors.
Base category: Education = completed or incomplete primary school, irrespective to gender and
experience.
Controls: same as in the benchmark Mincerian model (see Appendix 2)

Economic Transformation and the Revaluation of Human Capital

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

269

APPENDIX 4: SIMPLIFIED INTERACTIVE EARNINGS

MODEL (19861999)
Table A4.

Estimation Results.

Dependent: log of gross monthly earnings

DOMESTIC FIRMS
Interactive variables

1992

1993

1994

1995

skilled-young
skilled-old

0,2675
0,5053

0,2707
0,5036

0,3170
0,5586

0,2641
0,4919

81,301
0,3927
9,8293

75,791
0,3872
9,9952

82,768
0,3811
10,2428

65,754
0,3573
10,3331

Interactive variables

1996

1997

1998

1999

skilled-young
skilled-old

0,2517
0,4888

0,2307
0,4714

0,2613
0,4959

0,2579
0,4796

77,733
0,4031
10,5108

68,013
0,4401
10,6838

75,415
0,4424
10,8141

75,321
0,4277
10,9606

No. of observations
Adjusted R2
Mean ln (monthly wage)

No. of observations
Adjusted R2
Mean ln (monthly wage)

FOREIGN FIRMS
Interactive variables

1992

1993

skilled-young
skilled-old

0,3753
0,5089

0,3543
0,4919

0,3745
0,5422

0,3382
0,4849

5,493
0,3992
10,0673

9,815
0,4004
10,2099

11,490
0,3919
10,4685

24,650
0,3952
10,6292

Interactive variables

1996

1997

1998

1999

skilled-young
skilled-old

0,3385
0,5032

0,3363
0,4968

0,3496
0,4723

0,3361
0,4307

19,814
0,4196
10,8283

20,195
0,4092
11,0455

26,687
0,4435
11,2282

27,226
0,4627
11,3634

No. of observations
Adjusted R2
Mean ln (monthly wage)

No. of observations
Adjusted R2
Mean ln (monthly wage)

1994

1995

All coefficients are significant at 0,0001.

Notes: OLS regressions with heteroscedasticity corrected standard errors.
Skilled young: secondary or college, less than 22 years of experience.
Skilled old: secondary or college, 22 years of experience or more.
Base category: Completed or incomplete primary school, irrespective of experience.
Controls: as in the benchmark Mincerian model (see Appendix 2) except occupational groups.

269

270
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

GBOR KERTESI AND JNOS KLL

APPENDIX 5: COHORT-SPECIFIC CHANGES IN

RETURNS TO EXPERIENCE OVER THE LIFE-CYCLE
The gains and losses displayed in the table have been calculated as:
g = 100{[99 (G, E, exp + 10)  89 (G, E, exp)]
 [89

(G, E, exp + 10)  89 (G, E, exp)]}

where the -s measure the regression-adjusted wage advantage of workers with
E level of education, gender G, and exp years of experience over the base
category, in 1989 and 1999. The coefficients were estimated with the augmented
interactive model using 42 dummies for the interactions of two genders, three
levels of education, and seven experience groups. Workers with completed or
incomplete primary school background were treated as the reference.
Table A5. Life-Cycle Gains and Losses from Change in
Experience-Earnings Profiles between 1989 and 1999.
Labor market
entry

19841988
19791983
19741978
19691973
19641968

Experience
in 1989

15
610
1115
1620
2125

Vocational training
school
males females
2,5
2,8
4,0
3,1
2,1

6,0
3,7
1,6
3,7
4,4

Secondary school
males

females

3,1
4,3
0,6
2,2
3,4

12,0
4,5
3,8
0,3
0,4

College
males females
20,1
11,7
12,2
5,3
11,7

15,6
11,4
4,2
8,1
4,0

Economic Transformation and the Revaluation of Human Capital

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

271

APPENDIX 6: PRODUCTIVITY MODEL

Table A6.

Log Value Added per Worker.

ALL FIRMS

Independent Variables
Constant
log share of skilled-young
log share of skilled-old
log capital/labor ratio
No. of firms
Adjusted R2
Independent Variables
Constant
log share of skilled-young
log share of skilled-old
log capital/labor ratio
No. of firms
Adjusted R2

1986

1989

1992

1993

1,2928

1,2463

1994

2,2069

1,5558

0,2418
0,1175
0,1822

0,1591
0,1273
0,2428

0,1894
0,0723n
0,2145

0,2180
0,1069**
0,1833

0,4025
0,0170n
0,1354

971
0,2111

748
0,2170

600
0,1241

567
0,1736

506
0,2341

1995

1996

1997
*

0,8274

1998
**

0,4725

1,3857

0,4975

1999
*

0,5156

0,4461**

0,2510
0,0623n
0,2545

0,3227

0,3441

0,4385

0,0180n

0,0111n

0,0291n

0,2834

0,3417

0,2897

0,5355
0,1047**
0,2628

470
0,2666

477
0,3419

400
0,3665

476
0,4045

441
0,4677

DOMESTIC FIRMS
Independent Variables
Constant
log share of skilled-young
log share of skilled-old
log capital/labor ratio
No. of firms
Adjusted R2
Independent Variables
Constant
log share of skilled-young
log share of skilled-old
log capital/labor ratio
No. of firms
Adjusted R2

1992

1993

1994

1,2976

1,2451

1,4589

1995
0,9548

0,1658*
0,0788n
0,1982

0,1405*
0,1575*
0,1502

0,3346
0,0903n
0,1040

0,1157***
0,1868*
0,1887

543
0,1104

478
0,1324

420
0,2071

312
0,1777

1996

1997

1998

1999

0,6556

0,6073

0,6085

0,6032**

0,2431
0,0978n
0,2257

0,2492
0,0905n
0,2756

0,3103
0,1057***
0,2121

0,4225
0,0418n
0,1818

345
0,2923

276
0,3160

310
0,3518

269
0,3995

Unmarked coefficients are significant at 0,0001.

Significant at *0,01 **0,05 n not significant.

271

272
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

GBOR KERTESI AND JNOS KLL

Table A6.

Continued.

FOREIGN FIRMS
Independent Variables
Constant
log share of skilled-young
log share of skilled-old
log capital/labor ratio
No. of firms
Adjusted R2
Independent Variables
Constant
log share of skilled-young
log share of skilled-old
log capital/labor ratio
No. of firms
Adjusted R2

1992

1993

1994

1995

1,5904*

0,8008*

0,4548*
0,1748n
0,1800n

0,5979*
0,1018n
0,1874

0,6069
0,0461n
0,3077

0,3721
0,0514n
0,2397*

57
0,2601

89
0,4876

86
0,3612

158
0,3607

1996

1997

1998

1999

0,3343n

0,0804n

0,3491n

0,2204n

0,3946
0,0642n
0,3424

0,3015*
0,0308n
0,3961

0,4657
0,0824n
0,3517

0,1265n

132
0,3881

124
0,3245

166
0,3390

172
0,4322

1,7159

Unmarked coefficients are significant at 0,0001.

Significant at *0,01 **0,05 n not significant.

1,8830

0,4980
0,3613

Economic Transformation and the Revaluation of Human Capital

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

273

APPENDIX 7: ASSESSING SELECTIVITY BIAS

For lack of suitable data the standard Heckman technique of selectivity-bias
corrected wage regressions was not available for us. (Wages are not observed
in the Hungarian Labor Force Survey). We used an alternative, and admittedly
second-best, solution to assess how the wage distribution is affected by the
selection of wage earners. Using two consecutive waves of the Labor Force
Survey from 1993, when unemployment was at its peak, we estimated the
determinants of individual job loss. A dummy reporting whether the person
who was employed in the first quarter of 1993 lost his/her job by the next
quarter (1 = yes, 0 = no) was used as the dependent variable. The independent
variables included gender, age, schooling, occupation and residence. The
parameters of the probit were used to predict individual risks for employees
observed in the Wage Surveys. The distribution of workers by the predicted
individual risks were compared for subsequent points in time and for different
groups of workers. In cases where we observed a decrease in the average
predicted risk of a group and a simultaneous increase in the groups relative
wage, we considered selectivity bias as a potential underlying reason. (A
decrease of the average risk indicates that workers with a high risk of job loss,
or with high reservation wages, tend to leave the group and this may lead to
an increase of observed wages). This simple test could draw the attention to
groups where self-selection was strengthening over time.

273

274
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

GBOR KERTESI AND JNOS KLL