Earnings, Schooling, and Economic Reform: Econometric Evidence From Hungary (1986–2004)

The rise and demise of communism are two of the most important economic events of the twentieth century. One of communism's indisputable achievements was an egalitarian distribution of income, which was accomplished in large part by wage compression. Returns to skill were determined centrally, with wages set below equilibrium.

The transition from a centrally planned to a market-based economy was expected to have powerful effects. First, the liberalization of labor markets was expected to significantly raise returns to skill, as these were artificially compressed under communism. Second, at least until the collapse of communism, a large share of the labor force used vastly outdated technologies, and those specific skills would not be expected to be valued in a market economy. Third, such skill deterioration would have been accompanied by a devaluation of the labor market experience acquired during communism. Finally, the transition from a centrally planned to a market economy would involve sharp reductions in government education expenditures, which could translate into lower quality education.¹ The aim of this article is to present econometric evidence that throws light on these effects using unique data from Hungary from 1986 (before the fall of communism) to 2004 (the year Hungary became a member of the European Union).

The paucity of published studies about a country as important as Hungary is surprising. Although one of the most liberalized economies in the Soviet bloc, Hungary started the transition from communism with a gradualist approach to reform, albeit with a welcoming attitude toward foreign investors.² In addition to focusing on an economy that has not received attention commensurate with its importance,³ this article is one of the few examining labor markets in transition economies to use sufficiently large and representative samples of wage earners before, during, and after the introduction of massive economic reform.⁴ The data cover almost 4 million wage earners from 1986 to 2004 and, using a technique based on DiNardo, Fortin, and Lemieux (1996), also permit addressing the bias from selection on observables.

The results show that returns to a year of schooling increased by 75 percent, from 6.1 in 1986 to 10.7 percent in 2004. General secondary, college, and university education show the largest gains in returns from 1986 to 2004, whereas vocational and primary education show a decline in returns. Service employees experienced the largest gains in returns to skill, whereas construction and agricultural employees experienced the smallest gains.

How do these findings relate to the introduction of the massive economic reform mentioned earlier? Liberalization of the Hungarian economy was somewhat more intense than that of other transition economies (figure 1), and the results suggest that a link can be made between the progressively rising returns to schooling and the progressive liberalization of the Hungarian economy. Returns to skill increased rapidly overall, but the increase was much smaller for those employed in construction and agriculture and those with vocational (narrow) secondary education. These findings support the notion that the changes in returns to education during the transition may be taken as an indicator of the deepening of the reform process.

I. Data and The Empirical Specification

The data used in the analysis are from the Wage and Earnings Survey of the National Labor Center in Hungary and contain information on wages, education, type of employment, and other demographic details. Data for the seven years 1986, 1989, 1992, 1995, 1998, 2001, and 2004 cover the communist and transition periods.⁵ Wage earners are selected following a systematic, random-selection procedure (for details see Supplemental Appendix S.3). With assistance from the National Labor Center and the Hungarian Central Statistical Office considerable effort went into ensuring that variables were coded consistently over time, work that involved substantial recoding of the data.⁶

The monthly value of wages used in the analysis is the sum of the official base wage received and other payments to the employee (rewards given in the reference month, provisions, overtime work, shift work, and other special payments). In addition, the value of wages includes a prorated estimate of irregular payments (1/12 of irregular payments in the previous year).

Two measures of schooling are examined. The first is a vector of six dummy variables that denote the highest type of schooling completed. The school types include primary, three types of secondary (vocational, technical, and gymnasium or general), college, and university.⁷ In 2004, 17 percent of wage earners had only primary schooling or less, whereas 21 percent had a college or university education. Of the remaining 62 percent of wage earners who had completed some form of secondary schooling, slightly less than half had attended vocational school (29 percentage points of the total). The second measure of schooling is an estimate of years of school attainment, which is created by converting the data on highest school type completed into years of schooling. The average value of this variable increased from 9.7 in 1986 to 11.5 years in 2004. Potential experience is constructed as the wage earner's age minus six years and minus the number of years of schooling.

The variables designated by X include a set of eight dummy variables to control for differences across industries,⁸ a dummy variable for large firms (more than 300 employees), and a gender dummy variable to control for the large difference in wages across the sexes. The set of control variables also contains a dummy variable for each of Hungary's 19 counties and the capital, Budapest. These spatial variables control for any variation that is specific to Budapest or a specific county. In particular, these dummy variables control for spatial variation in prices, which is likely to be significant since wages and prices in Budapest are higher than in other regions. The county dummy variables will also control for region-specific differences in labor markets, which are potentially important since unemployment rates are lower in Budapest and the counties along the Budapest–Vienna highway and along the Austrian border. Similarly, the county dummy variables will control for the potential measurement problem that a year of schooling may result in different levels of human capital accumulation over different regions if there are differences in schooling quality across regions.

The controls for firm size and industry, as well as the county fixed effects, reduce the potential for omitted-variable bias in the estimation of equation (1). Having data that were collected using the same survey instrument also significantly improves the credibility of measured changes. That avoids the question, common when data come from different sources, of whether changes over time reflect actual changes in the population or simply the use of different survey instruments. These are important advantages to using the Wage and Earnings Survey data.

The primary disadvantage of using the survey data is that the choice of variables is small and thus the ability to empirically address violations of the ordinary least squares (OLS) assumptions is limited.⁹ In particular, if education is correlated with the residual, which can occur for several reasons, including if people select into (or out of) the sample based on some characteristic correlated with education, then the OLS estimator is biased. For example, if people with high returns are more likely to be wage earners and those with expected low returns opt out of the sample, this would induce correlation and result in sample-selection bias.¹⁰ This source of bias is typically corrected by modeling the selection decision, which requires data on the individuals who have opted out of the wage market. Because the survey provides information only on wage earners, this approach is unavailable. However, unique features of the labor market in 1986 provide important information that can be exploited to reduce sample-selection bias.

In centrally planned economies workers had limited ability to select in or out of the wage market. In principle, everyone of working age was required to work, official unemployment was close to zero, and the opportunity to choose to work in nonwage employment was highly limited. This lack of freedom to select out of the wage market implies that the pre-transition, 1986 estimates of wage equation (1) will be less susceptible to sample-selection bias.¹¹

Access to the 1986 pre-transition data is a unique feature of this analysis and helps to mitigate selection bias in the later, post-1986 years. The principal assumptions are that sample-selection bias was minimal in 1986 and that the decision to participate in the wage market after 1986 is correlated with age, gender, and schooling demographics.¹² Changes in these demographic variables in the post-1986 data are assumed to come from people selecting in and out of the wage market. The data are then reweighted to have the same demographic composition as in 1986, thereby purging labor supply changes from the data. This reweighting results in a counterfactual conditional wage distribution that corrects for (reduces) overt sample-selection bias.

Consider for example that low-educated young males constitute a larger proportion of the sample in 1986 than in 1995. This method adjusts their 1995 weights upward (in this example) to ensure that they represent the same proportions across both years. One difficulty with this approach is that it does not allow for any true population changes in the sex, age, and education composition of the sample.¹⁴ Although this affects the interpretation of the results, it is in some ways a desirable characteristic. Changes in returns to education can result from changes in the composition of the labor force and from the way the labor market rewards education, conditional on labor market characteristics. Reweighting the data to the 1986 demographic composition purges changes in the labor supply from the analysis so that the focus is on market changes in the demand for wage labor and on whether firms are responding to reforms by providing greater returns to investment in human capital.

II. Results

To examine how returns to schooling changed from 1986 to 2004, county fixed-effects estimates of equation (1) are provided in tables 1–4. Panel B of table 1 provides fixed-effects estimates for all firms with more than 50 employees. Panel A and all the other tables report fixed-effects estimates for the same sample weighted according to the formula above and therefore corrected for selection on observables. The standard errors listed in all tables are corrected for heteroscedasticity of unknown form through the use of the sandwich variance estimator.¹⁵ The disadvantage, as noted by Kauermann and Carroll (2001), is that the sandwich variance estimator is inefficient and often results in estimated standard errors that are too conservative (large). Given the large sample sizes, the cost of the sandwich or robust variance estimates are not qualitatively important, and the benefit of consistency is desirable.

The results from Panel B of table 1 show that the uncorrected returns to a year of schooling increased 123 percent, from a return of 6.1 in 1986 to 13.6 percent in 2004, and the increase is statistically significant. Once corrected for sample selection on observables, the increase in returns is smaller, suggesting the existence of the positive correlation between education and the decision to participate in the wage sector that was discussed above. Panel A shows that the selection-corrected return to schooling for wage earners from firms with 50 or more employees increased by only 75 percent, from 6.1 in 1986 to 10.7 percent in 2004. This result supports the hypothesis that central planners undervalued education and that the market has quickly corrected this undervaluation.¹⁶

Comparing the panels shows that sample-selection bias is positive and quite large throughout the period of analysis. The direction of the bias is consistent with the notion that people who expect to receive higher returns in the wage market choose to enter, whereas those who expect lower returns opt out. The decision of workers to select in and out of the sample appears to happen quickly. By 1992, the magnitude of the bias is more than 10 percent and remains above this level throughout the 1990s.¹⁷

Table 1 also provides evidence of decreasing returns to experience throughout the transition from a centrally planned to a market economy. This result appears to support the notion that labor market experience acquired under communism loses value after the introduction of economic reform.¹⁸ There is still little research focusing on the returns to experience in the labor market before, during, and after reform (Fleisher, Sabirianova, and Wang 2005), an area in need of in-depth research.

Under communism, a substantial share of the labor force was employed in large state-owned industrial enterprises. Government subsidies enabled these firms to survive with minimal technological modernization. Consequently, at the outset of transition a considerable share of the workforce had skills that were relevant for production technologies that were obsolete in the rest of the world (certainly in the Organization for Economic Co-operation and Development countries). Returns to the skills of manufacturing workers would thus be expected to be lower than returns to the skills of workers in other sectors (especially services). Table 2 shows estimates of the returns to skill from 1986 to 2004 in eight sectors.

The largest increases in returns to schooling are found in the service industries (excluding those in public services), with returns increasing 91 percent from 1986 to 2004. Correspondingly, the share of college-educated workers in the service industry doubled from 6 to 12 percent, suggesting an increase in demand for greater skills. Returns declined in construction (45 percent) and increased slightly in agriculture (27 percent). Similarly, the share of college-educated workers in these two industries declined from 42 to 6 percent. These results are consistent with the notion that the planned economy undervalued skilled labor used in the production of nonphysical goods and services.

There are several possible explanations for the varying returns to types of schooling during transition. One is that different types of schooling produce different skill sets, and these skills may be more or less well suited to the needs of the new market economy. A related explanation is that the government traditionally steered students into certain types of schools, and this planned aspect of the economy no longer provided the correct mix of skills. Both explanations are based on the idea that the changing market environment produced changes in the market value of certain skills. These hypotheses ignore the fact that under the planned economy returns to skills were set by planners and were not determined by the market. Prior to the transition wage setting was used to favor certain industries and certain types of labor. Labor who had been trained in technical and vocational schools and was involved in the production of certain goods tended to be more highly valued, whereas labor who had been more academically trained and less likely to be working in the physical production of goods was less highly valued. Presumably, the market economy rewards the value added by labor and is indifferent as to whether the added value is in a physical commodity or a service.

Although there was a large increase in the return to a year of schooling overall during this period, the wage premium to primary and vocational schooling declined between 1986 and 2004 (table 3). Wage earners who completed university, college, or secondary general education experienced the largest percentage changes in the wage premium. These results are consistent with the view that the planned economy undervalued labor used in the production of nonphysical goods and services relative to the market economy.

One indication that students are responding to the changing structure of returns by school type is that the share of students in general education increased from 24 in 1990 to 28 percent in 1997 (Fretwell and Wheeler 2001). This finding provides empirical evidence supporting the theoretical argument of Nelson and Phelps (1966) and Schultz (1975) that general education may enhance an individual's ability to adapt to a changing market environment. In contrast, the value of training in specific technical skills is more dependent on market fluctuations. When skills training is well targeted to the specific demands of the market, returns are high; when market and technology conditions change, there will be a time lag before curricula can adjust to provide the newly demanded mix of skills.

The final issue explored is whether there is evidence of qualitative changes in schooling after 1989. One hypothesis is that after 1989 schools responded to changing market needs and provided more marketable skills. A competing theory is that the quality of education has deteriorated because of the large declines in education budgets for many countries, including Hungary, during the transition period. Total public expenditures on education as a percent of gross domestic product increased in Hungary from 5.7 in 1989 to 6.6 percent in 1992, but then fell by 35 percent during the next 5 years, reaching a low of 4.4 percent in 1997 (Berryman 2000). As a result of the declining expenditures on education after 1992, studies note that many teachers have had to take on second jobs (Fretwell and Wheeler 2001) and that academic performance has been declining (World Bank 1997).

Although the Wage and Earnings Survey data include no direct measures of school quality, it is possible to provide limited supporting evidence.¹⁹ Until 1992, all wage earners acquired their schooling before the transition or while education expenditures were increasing. By 2004, the youngest wage earners in the sample had acquired most of their schooling during the post-1989 years, and much of it during the post-1992 period of declining expenditures. Contrasting the returns for the youngest wage earners in the sample with those for all others permits observation of whether the market considers schooling attained in the transition years more or less highly (table 4).

In 1986, the returns to schooling for wage earners, 20 years old and younger, were about 61 percent less than the returns for wage earners more than 20 years old. In the early years of transition, this gap narrowed, and by 1992 the difference stood at 17 percent. The difference in returns increased over the next 6 years, and by 1998 wage earners who were schooled in the post-1989 years had returns that were 163 percent less than the returns for those who received most of their schooling before 1989. During the first few years of transition, the youngest wage earners experienced the largest increases in the returns to schooling; after 1992, the returns to education stagnated for the youngest workers whereas it continued to increase for older workers.²⁰ This is consistent with the hypothesis of declining school quality and, if correct, could have negative repercussions for future economic growth and the earnings potential of the generation that received its education during the post-transition years.²¹

There are, however, additional plausible explanations for this finding. It might be that over time young people are increasingly likely to pursue post-secondary schooling. If so and if these are the most able, then it would likely be that the more able youth were sorted out of the under-20 cohort earnings equation. Another possibility is that the higher estimated returns for the older group may be due to the increasing share of more educated (college and university) workers in this group. A third possible explanation is that the results reflect changes in returns to experience that are difficult to distinguish from changes in the quality of education. Finally, young workers may be the most adversely affected by market regulations that constrain job creation, and the slowdown and slight reversal of liberalization in the late 1990s could be taken as support for this hypothesis. Although the data do not permit discriminating among these hypotheses, none appear to suggest that young workers are natural winners from market reforms.

III. Conclusions

This article analyzed the effects of the introduction of economic reform—the transition from a centrally planned to a market economy—on the labor market. It tried to improve on the existing reform literature by focusing on a highly effective reform, defined broadly and covering the periods before, during, and after its implementation. One important result found in the analysis is that returns to schooling are relatively large throughout the transition in Hungary (which is a standard example of gradual reform), at around 10 percent and above since 1995. The returns to a year of schooling increased by 75 percent, from 6.1 in 1986 to 10.7 percent in 2004, according to our preferred (selection-corrected) estimates. Although these returns are larger than those available for other transition economies and for Western Europe, they are credible estimates for several reasons.

Although many Eastern European countries have education levels that are on par with those in Western Europe, average wages continue to be lower. Since estimated returns to schooling are measured as percentage changes in wages, if Western and Eastern European markets were to provide similar returns to schooling in wage levels, this would mean higher returns from estimating Mincer equations such as equation (1). The prior assumption here is that if markets are truly liberalized, rates of return will be higher in transition economies than in Western Europe until there is some convergence in wage levels.

The difference found in estimates may be in part due to some unique characteristics of the survey. The data were collected using the same survey instrument over the years 1986–2004, covering the pre-transition as well as the transition years. Studies that are based on multiple survey instruments for temporal analysis face the difficult question of whether the observed change results from changes in the examined population or changes in the survey instrument. Further, the data were recoded to current standard international classifications to minimize errors in comparisons over time.

The unique characteristics of the Wage and Earnings Survey data allow for the exploration of the important issue of school quality. Because the sample sizes are large, wage equations for a small sub-sample of very young workers in each of the 7 years can be estimated. The analysis reveals that the returns to education for the generation that received much of its schooling after 1989 declined during the mid and late 1990s, while returns for older workers continued to increase.

Wage and Earnings Survey data impose difficult estimation issues, however, that are similar to those that arise with any labor force survey data when estimating returns to schooling. As with essentially all labor force surveys, the researcher has no information on the population that chose not to participate in the wage market and therefore cannot control for sample-selection bias. The approach used in this article to correct for this potential bias has been to exploit the somewhat unusual availability of representative data from 1986, well before the transition began and when workers had very limited choice about whether to participate in the wage sector. The estimation strategy was to use the demographic composition (based on 84 sex-age-schooling classifications) of the 1986 data as a basis for identifying which types of people disproportionately select in or out of the sample in later years.

The analysis showed that the 75 percent increase in returns to a year of schooling between 1986 and 2004 is evidence that the planned economy undervalued education and that liberalization has allowed markets to correct this. Examining returns by type of schooling rather than an aggregate measure of years of schooling sheds further light on whether the type of schooling received in pre-transition Hungary proved to be appropriate for the liberalized, post-1989 market. The common assumption is that socialist economies undervalued and undersupplied general education. The analysis strongly supports this view. Since 1989 an increasing proportion of students have been choosing to attend general school over vocational and technical school. Despite the increasing supply of students in general schooling, the estimated wage premiums to university and college education increased more than 60 percent between 1986 and 2004, whereas the premiums to secondary vocational training decreased 33 percent.

The empirical evidence in this article supports the belief that the liberalized economy has responded to market forces and is providing large returns for human capital investments. The evidence also suggests that wage earners are responding to the changes in the market and making better investment choices. All of this bodes well for future growth. The potential caveat to this conclusion, though, is that the declines in public expenditures on education may have resulted in a decline in the quality of this investment, and the markets seem to have responded to this.

The authors are affiliated with the Institute for the Study of Labour, Bonn, and the William Davidson Institute at the University of Michigan. The authors are grateful to Tim Barmby, Elizabeth Brainerd, John DiNardo, Peter Dolton, Francisco Ferreira, Randall Filer, Stepán Jurajda, Harmut Lehmann, Jaime de Melo, Branko Milanovic, three anonymous referees, and seminar participants at the Universities of Queen's Belfast, Bristol, and Newcastle; Bank of Finland Institute for Economies in Transition; World Bank Inequality and Pro-Poor Growth Conference; and the Northeast University Development Consortium Conference for valuable comments on previous versions. They also thank János Köllo, Gábor Kezdi, Gábor Kõrösi, and György Lázar for the help with the data, and Dana Žlábková and Ján Plánovský for excellent research assistance. The views and opinions expressed in this article do not necessarily reflect the views of the Economic Research Service of the United States Department of Agriculture. Supplemental appendixes to this article are available at http://wber.oxfordjournals.org/. This work benefited from financial support from Phare ACE Grant 97-8085.

References