Journal Issue
Share
Article

The Unbearable Stability of the German Wage Structure: Evidence and Interpretation

Author(s):
International Monetary Fund. Research Dept.
Published Date:
July 2004
Share
  • ShareShare
Show Summary Details

The Unbearable Stability of the German Wage Structure: Evidence and Interpretation

ESWAR S. PRASAD*

The relationship between labor market institutions and overall labor market performance has been the subject of considerable interest among academics and policymakers in recent years. In particular, a vigorous debate has emerged on the relationship between wage inequality and employment growth. This debate has been fueled by the large disparities in employment growth and unemployment rates between the United States and the United Kingdom, on the one hand, and the major continental European economies on the other.

Some authors have argued that labor market institutions that constrain changes in wage inequality despite shifts in the relative demand for different types of labor have resulted in rising aggregate unemployment rates in Europe (e.g., Siebert, 1997). Other authors have contested this view and find at best weak cross-country evidence that wage inequality is correlated with employment growth (e.g., Nickell and Bell, 1996; Card, Kramarz, and Lemieux, 1999). Gregg and Manning (1997) also find only weak evidence to support this proposition and argue that making relative wages an argument of the labor supply function is necessary to reconcile observed patterns of wage inequality and unemployment. More recently, Freeman and Schettkat (2000) have characterized the German unemployment problem as being attributable largely to deficient aggregate demand.

This paper provides some new perspectives on this debate by providing a detailed examination of the (West) German wage structure. This case study may be viewed as a complement to cross-country work examining the empirical relevance of the wage inequality-unemployment trade-off. The paper also investigates the roles of various factors that could have influenced patterns of changes in the wage structure. While a documentation of factors underlying the evolution of the wage structure in Germany is interesting in its own right, the analysis in this paper, by facilitating comparisons with changes in the wage structures of other industrial countries, could potentially provide important clues to understanding the poor functioning of the German labor market in recent years. In particular, the analysis attempts to shed light on the reasons behind and possible solutions for a particularly troubling problem, the high and rising rate of nonemployment among low-skilled workers.

The first part of the paper uses data from the German Socio-Economic Panel (GSOEP) over the period 1984–97 to characterize the key features of and changes in the West German wage structure. Over this period, the wage structure in Germany remained remarkably stable, with little change in inequality within or between groups. Returns to observed skill attributes such as education and experience remained essentially unchanged and, if anything, declined marginally during the 1980s. There is, however, some evidence of a modest increase in wage inequality during the mid-1990s. These results stand in stark contrast to the evolution of wage inequality in the United States, where inequality has risen sharply over the last three decades.

The second part of the paper examines a number of factors that could explain the stability of the German wage structure. These include shifts in the relative supplies of skilled and unskilled workers and changes in the sectoral composition of employment. None of these “market forces” appears capable of explaining developments in the wage structure. I also exploit certain unique features of the GSOEP data set to control for the effects of nonwage compensation, as well as selection and cohort effects. These factors turn out not to play a significant role in explaining the apparent stability of the wage structure.

That leaves “institutional factors” as the residual claimant. Indeed, for Germany, anecdotal and more formal evidence abounds that the wage bargaining system is the proximate cause for the rigidity of relative wages. Unions have traditionally set effective wage floors (there is no legislated minimum wage in Germany) and have negotiated uniform relative wage increases for workers of all skill levels, thereby constraining the flexibility of the wage structure. While these “solidaristic” policies may have served Germany well in previous decades, they have had a deleterious effect on labor market performance during 1984–97, a period during which the economy has been buffeted by a number of adverse shocks.1

As has been well documented for many other industrial economies, it is plausible and likely that there has been a substantial shift in the relative demand for skilled workers in Germany. Factors that have accentuated this demand shift in other countries include skill-biased technological change, increased openness to international trade, and de-industrialization, all of which are forces that appear to operate in Germany as well. For instance. Machin and Van Reenen (1998), using an industry-level database that is comparable across countries, provide persuasive evidence that skill-biased technological change has resulted in relative demand shifts favoring skilled workers in virtually all Organization for Economic Cooperation and Development (OECD) industrial countries including Germany (Manacorda and Petrongolo, 1999, reach a similar conclusion).

This paper argues that the rigidity of the German wage structure, coupled with these relative demand shifts that have been accentuated by a series of adverse macroeconomic shocks, has resulted in marked increases in unemployment rates and a deterioration of employment prospects for unskilled workers, in other words, given the inflexibility of the relative prices of skills in response to market forces, employers have had to adjust the relative quantities of skilled and unskilled labor that they employ, to the detriment of unskilled workers. Indeed, employment and retention rates for unskilled workers have continued to fail during the recent recovery, in sharp contrast to the rising employment rate for skilled workers.

The third part of the paper provides a synthetic perspective on recent developments in the German labor market. In particular, I argue that it is essential to draw a distinction between skilled and unskilled labor to reconcile the micro evidence on the wage structure presented in this paper with macroeconomic phenomena such as the decline in the wage share, capital-labor substitution, and rising aggregate unemployment.

I. The Wage Structure

The Data Set

The data used in this paper are drawn from the public-use version of the GSOEP for the years 1984–97. This is a representative sample of German households and individuals, including immigrants without German citizenship. The sample was expanded to cover unified Germany in the 1990s. One of the notable features of the data set is its large and relatively stable panel. Nevertheless, the nonresponse rate for repeat interviews is large enough that attrition bias is a serious concern. New individuals from existing households are added to the survey as they enter the labor force. To reduce the effects of attrition bias and to make the results in this paper reasonably representative of the population, the data set is treated here as a set of repeated cross sections rather than as a longitudinal survey. This distinction has the virtues of yielding a larger sample size and keeping the sample size relatively stable over lime. But it should be noted that this approach could mask the effects of compositional changes in the employed workforce.

To maintain a uniform sample and to minimize distortions from sample selection, much of the analysis below, except where explicitly noted otherwise, is limited to full-time male workers from the West German sample. This approach facilitates comparisons with studies for industrial countries that have focused on samples based on similar selection criteria.2

The wage variable used in this paper is the real gross hourly wage (excluding end-of-year bonuses), constructed using reported gross monthly earnings and “usual number of weekly hours” worked, and using the consumer price index for West Germany (1991 = 100) as the price deflator.3 The hourly wage is the appropriate measure of the price of labor inputs that is relevant for the analysis in this paper. Nevertheless, the sensitivity of the results to using a measure of monthly earnings is also examined below.

The GSOEP contains a generated variable on years of education for individuals in the West German sample. However, there is a strong tradition of apprenticeship and vocational training in Germany, and this variable would not adequately capture the returns to such training. Hence, I split the sample into four categories of educational attainment—general schooling, apprenticeship, vocational training, and university degree. Since the focus of this paper is on changes over lime in skill premia rather than their levels, this choice turned out not to matter for any of the results reported below. Nevertheless, this discussion should be kept in mind should the results from this paper be used for cross-country comparisons of the levels of skill premia. Labor market experience, which is to be interpreted as potential rather than actual labor market experience, is defined as age minus years of education minus six.

Finally, it should be noted that noncitizens are overrepresented in the GSOEP sample relative to their share of the West German population. Where appropriate, I use the GSOEP cross-sectional weights that are intended to control for this feature of the sample. In the regression results, I directly include controls for citizenship.

I restricted the sample to workers between the ages of 17 and 65 and excluded workers who are self-employed, report less than 35 or more than 60 weekly hours of work, or report hourly wages of less than 5 deutsche mark at 1991 prices.

The Overall Wage Structure

Figure 1 displays some summary statistics for real wages for all full-time workers, including women. The top panels show that the median wage increased by a total of about 20 percent over the period 1984–97. Although a significant gender gap remains, the relative female-male wage differential narrowed significantly during this period. The bottom panels of this figure show three-year moving averages of two summary measures of wage dispersion—the standard deviation and the coefficient of variation.4 Both measures of dispersion declined slightly during the latter half of the 1980s, stayed flat through about 1994, and then rose slightly after the mid-1990s.

Figure 1.Median and Dispersion Measures for Hourly Wage

The remainder of this paper will focus on results for the male sample, but it should be noted that preliminary results for the sample of women indicate very similar patterns of changes in wage dispersion here and, as described below, for men. Some summary statistics for the final sample used to analyze the wage structure are shown in Table 1. Figure 2 shows the evolutions of the relative sizes of different education and experience groups in the wage analysis sample.

Table 1.Summary Statistics
VariableMeanStandard Deviation
Log hourly real wage (gross)3.000.36
Log monthly real earnings (gross)8.210.37
Age (in years)39.6011.28
Experience (in years)22.3811.48
Tenure (in years)12.059.61
Education dummies:
General schooling0.220.41
Apprenticeship0.410.49
Vocational training0.260.44
University degree0.110.32
Citizenship dummy0.690.46
Weekly hours worked in survey month42.545.41
Notes: The summary statistics reported here are for West German workers with full-time employment and for whom data on all of the variables listed above are available. Nominal wages were deflated by the consumer price index for West Germany (1991 = 100). The total number of observations over the period 1984–1997 is 32,713 (average of about 2,340 per year). The results reported in this table are weighted by cross-sectional sampling weights.
Notes: The summary statistics reported here are for West German workers with full-time employment and for whom data on all of the variables listed above are available. Nominal wages were deflated by the consumer price index for West Germany (1991 = 100). The total number of observations over the period 1984–1997 is 32,713 (average of about 2,340 per year). The results reported in this table are weighted by cross-sectional sampling weights.

Figure 2.Means of Skill Proxies: Males With Full-Time Jobs

Which pan of the wage distribution has accounted for the apparent stability of the overall distribution? One way to approach this issue, following Juhn, Murphy, and Pierce (1993; henceforth referred to as JMP), is to examine the cumulative change in real wages across the entire distribution. The first (top left) panel of Figure 3 shows the change in real wages from 1984–97 at each percentile point of the aggregate wage distribution for full-time male workers.5 Over this period, there was a marginal increase in inequality. However, cumulative wage growth at the top part of the distribution appears to have been only about 5 percentage points higher than at the bottom part of the distribution. This is in stark contrast to recent patterns of wage growth across the distribution in countries such as the United Kingdom and the United States that are viewed as having more “flexible” labor markets (Figure 4). Note that, for the United States, the plotted line has a steep positive slope, with cumulative negative real wage increases at the low end of the distribution.6

Figure 3.Changes in Log Wage Across Wage Distribution

Figure 4.Wage Growth at Different Percentiles: Cross-Country Evidence

The remaining panels of Figure 3 break down the total change over the period 1984–97 into three subperiods. Wage growth across the distribution appears to have been flat during the 1980s, followed by a slight compression during 1989–92, and then by a slight widening of the wage structure in the 1990s.

Panel A of Table 2 shows a number of percentile differentials for hourly wages. The 90–10 percentile differential declined marginally during the 1980s before returning to its earlier levels by the mid-1990s. Although the 75–25 percentile differential is essentially flat, it does show a small increase between 1992 and 1996. It is also interesting to note that the contribution of inequality above the median of the distribution to total wage inequality is greater than that of inequality below the median. In other words, the wage structure is more compressed below the median than above. However, the slight increase in wage inequality in the 1990s seems to have occurred both above and below the median of the distribution.

Table 2.Measures of Wage Inequality
Percentile
Differential:90-1090-5050-1075-2575-5050-25
A. Log Hourly Wage
19850.790.430.360.370.190.18
(0.02)(0.02)(0.02)(0.02)(0.01)(0.01)
19890.770.440.330.380.200.17
(0.02)(0.02)(0.01)(0.01)(0.01)(0.01)
19920.760.420.340.370.200.17
(0.03)(0.02)(0.02)(0.02)(0.02)(0.01)
19960.800.440.360.400.220.19
(0.03)(0.03)(0.02)(0.02)(0.02)(0.01)
B. Wage Residuals
19850.610.310.300.310.160.15
(0.02)(0.02)(0.01)(0.01)(0.01)(0.01)
19890.610.320.290.310.160.15
(0.02)(0.01)(0.01)(0.01)(0.01)(0.01)
19920.610.320.290.310.160.15
(0.02)(0.02)(0.01)(0.02)(0.01)(0.01)
19960.630.320.300.320.160.16
(0.02)(0.02)(0.02)(0.02)(0.01)(0.01)
Sources: German Socio-Economic Panel and author’s calculations.Notes: The reported differentials are three-year averages centered on the years shown above. Standard errors are in parentheses. The sample includes West German males with full-lime jobs. Panel B reports differentials based on residuals from annual regressions of log hourly wages on a constant, education dummies, tenure, experience and its square, a dummy for German citizenship, and interactions of this dummy with the education dummies, tenure, experience, and squared experience.
Sources: German Socio-Economic Panel and author’s calculations.Notes: The reported differentials are three-year averages centered on the years shown above. Standard errors are in parentheses. The sample includes West German males with full-lime jobs. Panel B reports differentials based on residuals from annual regressions of log hourly wages on a constant, education dummies, tenure, experience and its square, a dummy for German citizenship, and interactions of this dummy with the education dummies, tenure, experience, and squared experience.

Overall, the analysis so far yields a picture of relative stability in the aggregate German wage structure over the last 15 years. There is little evidence of major increases in wage inequality, let alone increases in inequality of the magnitude seen in the United Kingdom and the United States.

Within-Group Inequality

It is interesting to examine the evolution of wage inequality within skill groups to understand the effects of within- and between-group wage dynamics on overall inequality. In the United States, for instance, JMP have documented that the rise in wage inequality in recent decades has been as dramatic within narrowly defined skill groups as it has been in terms of increases in inequality between these groups. Figure 5 plots cumulative wage changes at different ventiles for specific skill groups. Skill groups are defined on the basis of three skill attributes—education, labor market experience, and tenure on the current job. The pattern of a mild increase in wage inequality across the distribution is consistent across most skill groups.

Figure 5.Within-Group Changes in Log Wage, 1984–97

A more direct approach to control for between-group effects is to regress wages on observed skill attributes and to examine the dispersion of the wage residuals. Inequality measures based on wage residuals from human capital wage equations are reported in Panel B of Table 2. The percentile differentials based on wage residuals are smaller than those based on actual wages but are still quite large, indicating that unobserved attributes constitute an important determinant of the wage distribution. The time profiles of the percentile differentials in this panel are, however, very similar to those in the top panel, indicating that within-group inequality has also been quite stable over the last 15 years.

Relative Prices of Skills

Next is an examination of changes in between-group inequality, based on changes in prices for observed skill attributes. The evolution of skill prices has important implications, since the incentives for acquisition of human capital are determined by the returns to that capital. The general equilibrium effects of inadequate wage differentiation, which typically implies smaller returns to skill attributes, could therefore be quite large. Furthermore, the evolution of the wage structure has implications for the demand for different types of labor.

Examined first is the evolution of skill prices based on estimates of standard human capital wage regressions. The results reported below are based on annual ordinary least squares (OLS) regressions of log hourly wages on education dummies, labor market experience, the square of labor market experience, tenure, a dummy variable for German citizenship, and interactions of this dummy with the other variables.7 Labor market experience may be viewed as a component of general human capital, while the tenure variable would be expected to pick up the returns to firm-specific human capital.

Figure 6 shows the evolutions of three-year moving averages of the estimated (conditional) returns to education, experience, and tenure. The regression coefficients for each year are reported in Table 3. General schooling is the excluded education category. Hence, the returns to the other three categories of education are expressed relative to that category. Since the experience variable enters the regressions as a quadratic, the returns to experience are evaluated at particular levels of experience.

Figure 6.Returns to Education, Experience, and Tenure

Table 3.Skill Premia: OLS Results
EducationExperienceTenure
VocationalUniversityAdjustedNumber of
ApprenticeshipTrainingDegree5 years15 years25 yearsR-squaredObservations
198411.9216.6447.402.581.420.260.530.372,865
(1.38)(1.49)(3.20)(0.15)(0.08)(0.05)(0.07)
198512.3617.5849.432.441.390.340.580.382,550
(1.72)(1.84)(3.47)(0.18)(0.10)(0.06)(0.08)
198611.5814.9445.882.421.370.310.580.342,521
(1.77)(1.86)(3.59)(0.18)(0.10)(0.07)(0.08)
19879.1313.9448.162.341.330.320.520.352,582
(1.63)(1.74)(3.24)(0.18)(0.10)(0.06)(0.08)
19869.8713.0447.772.411.360.310.550.362,447
(1.59)(1.70)(3.05)(0.18)(0.10)(0.06)(0.08)
198910.2512.7345.722.111.210.300.530.352,476
(1.62)(1.74)(3.06)(0.17)(0.10)(0.06)(0.08)
199011.1514.3445.942.281.270.260.470.352,398
(1.61)(1.73)(2.92)(1.18)(0.10)(0.06)(0.08)
19919.8313.9842.512.331.270.220.580.332,332
(1.76)(1.87)(2.87)(0.19)(0.11)(0.07)(0.08)
19929.7813.3144.782.291.320.340.410.342,225
(1.78)(1.88)(2.88)(0.20)(0.11)(0.07)(0.80)
199310.8814.2947.962.461.370.270.440.382,213
(1.68)(1.84)(3.18)(0.19)(0.11)(0.06)(0.08)
19947.8011.0943.822.541.380.220.560.372,132
(1.88)(2.00)(3.03)(0.21)10.12)(0.07)(0.08)
19959.0111.6344.912.551.410.280.500.352,056
(1.85)(1.97)(3.23)(0.21)(0.12)(0.07)(0.08)
199612.1216.4852.212.701.520.430.560.351,985
(2.171(2.30)(3.76)(0.23)(0.13)(0.08)(0.10)
19977.6410.5346.532.691.500.310.720.371,931
(2.10)(2.23)(2.80)(0.22)(0.12)(0.08)(0.09)
Sources: German Socio-Economic Panel and author’s calculations.Notes: The results reported in this table are from OLS regressions of log hourly wages on a constant, education dummies, experience and its square, tenure, a dummy for German citizenship, and a full set of interactions with this dummy. The excluded education category is general schooling. Since experience enters the specification as a quadratic, the returns to experience are evaluated at specific experience levels. All coefficients were multiplied by 100. Robust standard errors are reported in parentheses.
Sources: German Socio-Economic Panel and author’s calculations.Notes: The results reported in this table are from OLS regressions of log hourly wages on a constant, education dummies, experience and its square, tenure, a dummy for German citizenship, and a full set of interactions with this dummy. The excluded education category is general schooling. Since experience enters the specification as a quadratic, the returns to experience are evaluated at specific experience levels. All coefficients were multiplied by 100. Robust standard errors are reported in parentheses.

As noted earlier, differences in definitions of the education variables make it difficult to compare the levels of education premia across countries. It is striking, however, that the education premia for all three categories are relatively flat over the entire sample. This is, again, in sharp contrast to the U.K. and U.S. experiences, where education premia rose sharply during the 1980s and 1990s. In the United States, for instance, the college-high school differential more than doubled over a similar period, from 25 percent in 1980 to over 50 percent by 1995.

For workers with low and medium levels of experience (5 and 15 years of experience, respectively), the returns to an additional year of experience declined marginally during the 1980s, followed by a reversal of this decline during the 1990s. As in other industrial countries, the marginal returns to experience tend to be lower at higher levels of experience. One interesting finding is that, compared with some other industrial countries, experience premia are lower in Germany at all experience levels. For the United States, for instance, Buchinsky (1994) reports average returns to experience of about 5 percent and 3 percent when evaluated at 5 and 15 years of experience, respectively, in the 1980s. Returns to experience appear to have been lower but also relatively more stable in Germany over the last 15 years.

The marginal returns to tenure, after controlling for attributes that would be expected to reflect general human capital, are quite small. An additional year of tenure adds about 0.6 percent to the hourly wage, and, apart from a slight increase toward the end of the sample, this coefficient appears not to have changed much.

An interesting question that arises at this juncture is how recent labor inflows into West Germany have affected the wage distribution. To examine this issue, the sample was extended for 1991–97 to include migrants and commuters from East Germany. Starting in 1994, the GSOEP includes an additional sample of immigrants. For the period 1994–97, full-time male workers from this sample were included and then the OLS wage regressions (using sample weights to correct for the consequent overrepresentation of immigrants) were recomputed. Given the small number of migrants and commuters in the sample for 1991-93, the results for these years hardly changed. For 1994–97, estimates of OLS wage regressions indicated that the slight increase in skill premia in the mid-1990s apparent in the West German sample is in fact slightly attenuated in this broader sample (Prasad, 2000, Table 3A). Overall, the results remained unchanged. To maintain a homogeneous sample, the remainder of the analysis in this paper is limited to the basic West German sample.

Quantile Regressions

How have skill prices changed at different parts of the wage distribution? The OLS regressions provided estimates of the marginal returns to human capital attributes only at the conditional mean of the data. Quantile regressions can be used to provide a parsimonious characterization of the entire conditional wage distribution. This technique can be used to estimate the marginal return to an observed skill attribute at any specific quantile point of the aggregate distribution, I estimated a set of quantile wage regressions, keeping the independent variable and the dependent variables the same as in the OLS regressions discussed above.

Table 4 shows that, as in other countries, the returns to education tend to be higher at the upper quantiles of the distribution. For instance, in 1984, the marginal return to a university degree (relative to general schooling) was about 38 percent (exp(32.17/100)) at the 0.10 quantile compared with 72 percent at the 0.90 quantile. There is a fair amount of year-to-year variation in the estimated conditional returns to education at different quantile points. Overall, it is difficult to detect any systematic patterns of changes in wage inequality.

Table 4.Returns to Education: Quantile Regressions
ApprenticeshipVocational TrainingUniversity Degree
Education level:
Quantile point:0.100.250.500.750.900.100.250.500.750.900.100.250.500.750.90
198412.1610.3910.9411.2514.8915.6414.3015.2617.3621.4232.1741.4551.4355.2253.97
(3.64)(1.65)(1.70)(1.86)(2.38)(3.65)(1.98)(1.52)(1.99)(2.31)(7.77)(3.90)(6.74)(2.35)(2.86)
198511.919.8910.2910.8515.1315.5916.0214.4318.1221.9246.3641.6453.8155.1656.24
(4.07)(3.08)(1.76)(2.56)(3.46)(3.52)(3.54)(1.66)(2.55)(3.62)(6.18)(3.78)(6.27)(3.74)(5.14)
198613.9712.5310.8211.2514.0917.9512.3012.4417.4418.8540.8641.5346.3447.0257.52
(2.75)(2.25)(1.64)(1.82)(3.22)(3.15)(2.50)(2.19)(2.51)(3.69)(4.49)(4.97)(4.30)(3.42)(6.97)
198710.2510.399.079.9310.4613.5313.2214.3815.9117.4941.0443.2246.7152.2757.37
(3.05)(1.69)(1.10)(1.34)(4.73)(2.85)(2.34)(1.73)(1.74)(4.75)(4.01)(5.39)(4.00)(4.39)(6.90)
19886.607.3911.7510.3911.038.5010.2415.7615.0515.8642.5142.2446.6549.0655.83
(2.34)(1.94)(2.25)(2.23)(3.31)(3.19)(2.01)(2.04)(2.33)(4.90)(3.21)(2.65)(4.S8)(4.88)(3.52)
198913.1410.038.819.668.5116.4210.6511.9513.5511.6941.9240.7143.7345.8150.58
(3.80)(1.43)(1.90)(2.87)(4.64)(3.53)(1.62)(1.96)(2.93)(4.21)(7.20)(3.97)(3.07)(3.90)(6.98)
199010.448.6710.4711.0114.0914.9811.6813.3415.8018.0539.5439.4747.1146.2355.47
(3.45)(2.70)(1.61)(1.97)(3.21)(3.63)(3.00)(2.13)(2.22)(3.80)(6.39)(6.32)(3.15)(2.90)(6.97)
19919.6911.269.8610.147.0713.6613.4014.7215.3614.6338.1641.1941.2945.8743.37
(1.93)(2.24)(1.80)(1.55)(5.63)(2.58)(2.32)(2.25)(2.70)(5.54)(4.50)(3.34)(3.75)(4.75)(6.12)
19927.627.299.449.1411.6210.8711.0713.6413.8316.1239.4438.3846.6645.6049.38
(3.77)(1.42)(2.15)(2.38)(2.84)(3.64)(2.20)(2.54)(2.79)(3.92)(6.36)(3.55)(3.65)(3.87)(6.43)
199313.3811.5610.7411.2112.1714.7813.4415.5916.3314.3039.8841.6448.0454.8157.43
(2.97)(1.70)(2.63)(2.44)(3.11)(3.99)(2.16)(2.54)(3.18)(4.37)(5.68)(3.75)(4.79)(3.28)(7.33)
19948.6810.748.649.569.7111.0612.1010.8214.5515.0937.8641.6944.4649.6844.81
(3.99)(2.38)(2.27)(2.41)(4.32)(3.23)(2.87)(2.11)(3.04)(4.94)(3.76)(4.30)(4.96)(2.73)(7.98)
199513.098.058.139.745.4410.949.4011.7914.637.6439.3639.0947.4050.2251.08
(3.36](2.20)(1.69)(2.65)(3.65)(3.59)(2.66)(1.75)(2.58)(3.26)(5.57)(4.44)(2.93)(4.65)(6.44)
199615.389.4411.2711.108.7418.6314.5416.8518.4915.5845.3248.0254.7255.0462.66
(5.51)(2.77)(1.97)(3.08)(4.04)(5.90)(2.22)(1.80)(2.46)(4.06)(6.86)(4.27)(3.98)(2.97)(8.69)
19979.459.246.846.634.8310.6511.7610.2011.259.8142.5744.4747.1549.7748.81
(4.64)(1.80)(2.80)(3.37)(3.24)(4.77)(2.36)(2.29)(3.67)(3.68)(5.51)(3.39)(2.31)(4.26)(3.73)
Sources: German Socio-Economic Panel and author’s calculations.Notes: All coefficients were multiplied by 100. See notes to Table 3 for list of repressors. Bootstrapped standard errors are reported in parentheses.
Sources: German Socio-Economic Panel and author’s calculations.Notes: All coefficients were multiplied by 100. See notes to Table 3 for list of repressors. Bootstrapped standard errors are reported in parentheses.

Table 5 shows that the returns to an additional year of experience are higher at lower experience levels, that is, among younger workers. The returns to experience at different experience levels estimated from the quantile regressions are, however, significantly lower than those estimated for the United States (Buchinsky, 1994). Interestingly, during the 1980s, returns to experience were higher at the lower quantiles of the distribution than at the upper quantiles. Toward the end of the sample, this pattern changes, and the returns to experience become consistently higher at the upper quantiles of the distribution. This is true at all experience levels, although the timing of this switch occurs at different years for different experience levels. Consistent with the aggregate results, returns to experience at all experience levels and at most quantile points are slightly higher by 1997 than in 1989–1990.8

Table 5.Returns to Experience: Quantile Regressions
5 years15 years25 years
Experience level:
Quantile point:0.100.250.500.750.900.100.250.500.750.900.100.250.500.750.90
19842.962.502.372.172.231.631.371.311.171.180.300.240.240.180.14
(0.25)(0.20)(0.23)(0.18)(0.30)(0.14)(0.10)(0.13)(0.11)(0.17)(0.10)(0.06)(0.06)(0.07)(0.08)
19852.982.472.322.001.841.671.371.321.131.070.360.260.330.260.31
(0.29)(0.31)(0.19)(0.17)(0.26)(0.15)(0.17)(0.11)(0.09)(0.13)(0.10)(0.10)(0.07)(0.07)(0.14)
19862.772.332.332.292.391.551.321.281.281.360.330.310.230.27033
(0.30)(0.23)(014)(0.33)(0.44)(0.17)(0.12)(0.08)(0.19)(0.24)(0.12)(0.07)(0.06)(0.09)(0.14)
19872.462.102.112.132.321.381.211.181.181.310.310.320.260.240.30
(0.31)(0.26)(0.20)(0.25)(0.29)(0.18)(0.15)(0.11)(0.14)(0.15)(0.12)(0.06)(0.05)(0.07)(0.10)
19882.942.332.131.992.011.681.301.211.101.150.420.270.290.200.30
(0.28)(0.29)(0.18)(0.26)(0.38)(0.16)(0.15)(0.10)(0.16)(0.20)(0.09)(0.07)(0.07)(0.09)(0.11)
19892.162.001.921.852.091.291.181.101.031.130.430.370.280.220.18
(0.23)(0.22)(0.23)(0.25)(0.37)(0.13)(0.13)(0.13)(0.14)(0.21)(0.08)(0.10)(0.07)(0.08)(0.12)
19902.562.111.932.132.811.371.191.081.191.580.180.270.230.240.34
(0.44)(0.21)(0.20)(0.20)(0.27)(0.23)(0.12)(0.11)(0.12)(0.18)(0.15)(0.07)(0.08)(0.08)(0.13)
19912.652.152.032.062.051.411.171.101.161.200.160.190.170.260.35
(0.45)(0.29)(0.16)(0.22)(0.32)(0.26)(0.18)(0.09)(0.12)(0.19)(0.14)(0.12)(0.08)(0.09)(0.13)
19922.392.182.111.912.631.321.261.211.121.490.240.340.310.340.35
(0 33)(0.27)(0.16)(0.26)(0.32)(0.16)(0.16)(0.10)(0.15)(0.19)(0.15)(0.07)(0.08)(0.10)(0.18)
19932.572.372.122.412.111.441.311.191.381.240.310.240.260.340.37
(0.35)(0.20)(0.21)(0.34)(0.32)(0.18)(0.12)(0.12)(0.16)(0.18)(0.08)(0.08)(0.07)(0.07)(0.12)
19942.952.282.122.311.991.591.201.161.281.110.230.120.200.240.24
(0.42)(0.30)(0.17)(0.29)(0.36)(0.25)(0.18)(0.09)(0.17)(0.22)(0.13)(0.08)(0.08)(0.12)(0.15)
19952.142.302.292.662.741.171.261.261.521.470.210.220.220.380.20
(0.36)(0.26)(0.21)(0.27)(0.34)(0.20)(0.17)(0.14)(0.17)(0.19)(0.09)(0.12)(0.10)(0.09)(0.12)
19962.752.462.212.553.011.491.391.221.451.770.240.320.240.360.53
(0.36)(0.23)(0 28)(0.37)(0.31)(0.22)(0.12)(0.16)(0.22)(0.23)(0.12)(0.07)0.08)(0.18)(0.25)
19972.612.432.652.402.701.431.341.481.401.620.250.250.310.410.53
(0.39)(0.24)(0.22)(0.32)(0.35)(0.23)(0.13)(0.13)(0.19)(0.23)(0.14)(0.06)(0.10)(0.08)(0.20)
Sources: German Socio-Economic Panel and author’s calculations.Notes: Returns to experience are evaluated at specific experience levels. All coefficients were multiplied by 100. See notes to Table 3 for list of regressors. Bootstrapped standard errors are reported in parentheses.
Sources: German Socio-Economic Panel and author’s calculations.Notes: Returns to experience are evaluated at specific experience levels. All coefficients were multiplied by 100. See notes to Table 3 for list of regressors. Bootstrapped standard errors are reported in parentheses.

Although these results indicate some differences in the evolution of skill prices at different parts of the distribution, the overall picture is one of a relatively stable wage structure over the last 15 years, especially compared with changes in the U.S. wage structure. For instance, Buchinsky (1994) estimates average returns to a year of education of about 7 percent in the early 1980s in the United States, rising to about 10 percent by the mid-1980s. He finds a much larger return to education at upper quantiles of the wage distribution than at the lower quantiles and also finds that this disparity has widened significantly during the 1970s and 1980s. This echoes JMP’s findings that both between- and within-group wage inequality have risen in the United States in recent decades. For Germany, abstracting from year-to-year variation, both between- and within-group inequality have changed only very modestly over the last 15 years.9

Effects of Changes in Observed and Unobserved Prices and Quantities

For a more complete description of the effects of changes in skill quantities and prices, I now employ a technique developed by JMP that permits a decomposition of changes in inequality into the components attributable to changes in observed skill quantities, observed skill prices, and unobserved quantities and prices of skills. The main advantage of this framework, compared with a more traditional variance decomposition, is that it facilitates an analysis of how composition and price changes have affected the entire wage distribution.

Consider a wage regression of the form

where wit, is the log wage, Xit, is a vector of observed individual-specific characteristics, uit is the regression residual, and i and t are individual and time subscripts, respectively. The residual is composed of two components; an individual’s percentile in the wage distribution. θit, and the distribution function of the wage residuals, F t(·). In other words,

where Ft1(|Xit) is the inverse cumulative residual distribution for workers with characteristics Xit, in year t. Let ¯² be the set of average prices for observed skill attributes and F¯(|Xit) be the average cumulative distribution. equation (2) can then be rewritten as

Using this formulation, it is straightforward to construct conditional wage distributions that allow one component to vary while keeping the other components fixed. For instance, with fixed observable prices and a fixed residual distribution, equation (3) collapses to

It is then possible to construct wage distributions where the changes over time are attributable solely to changes in observable quantities. Similarly, holding the other components fixed in turn, one can construct wage distributions where the changes in the distributions over time are attributable to changes in observed prices and to changes in unobserved prices and quantities (i.e., the residual), respectively.

Table 6 reports results from this decomposition to examine the changes in wage inequality that can be attributed to these three components. The first column indicates that the total increase in the 90-10 differential over the period 1984–97, small as this increase is, is entirely attributable to changes above the median of the distribution. The compression of the wage distribution below the median in 1985–89 is almost exactly offset by a slight widening of the dispersion in the remaining years of the sample, leaving the dispersion in the lower part of the wage distribution essentially unchanged over the full sample.

Table 6.Decomposition of inequality Changes into Components Attributable to Observed and Unobserved Quantity and Price Changes
PercentileTotalObservedObserved
DifferentialChangeQuantitiesPricesResidual
1985–96
90-100.0140.0010.0010.013
90-500.0150.0070.0010.007
50-10-0.001-0.0060.0000.005
1985–89
90-10-0.019-0.0160.000-0.003
90-500.0100.0090.0000.001
50-10-0.029-0.024-0.001-0.004
1989–92
90-10-0.008-0.016-0.0020.009
90-50-0.018-0.0220.0000.004
50-100.0100.006-0.0020.006
1992–96
90-100.0410.0320.0030.006
90-500.0230.0200.0000.003
50-100.0180.0120.0030.004
Sources; German Socio-Economic Panel and author’s calculations.Notes: The numbers reported above are changes in three-year averages centered on the years shown. See text for details of the decomposition technique.
Sources; German Socio-Economic Panel and author’s calculations.Notes: The numbers reported above are changes in three-year averages centered on the years shown. See text for details of the decomposition technique.

The relative importance of changes in the residual distribution for changes in overall inequality is similar above and below the median during 1985–89 and 1992–96. During 1989–92, changes in observed quantities result in some compression (relative to the 1980s) above the median and a slight widening below the median. It is interesting to interpret this result in the context of German unification. Even though the associated influx of workers into West Germany included workers with relatively high formal educational attainment, these qualifications may have been valued less than equivalent qualifications obtained in West Germany. It is also likely that migrants from the East were viewed as having less favorable work habits (and other unobserved attributes). Thus, although these migrants are not in the sample, the increase in the supply of low-skill workers might account for the slight widening of the wage structure below the median. However, note that even in this period there is no perceptible change in the contribution of observed prices to changes in inequality.

During 1992–96, the wage compression that occurred over the previous decade was largely reversed, with changes in the distribution of skills and increases in within-group inequality (the residual) accounting for much of the increase in overall inequality. This increase was spread in a roughly equal manner above and below the median. In none of the subperiods examined here do changes in the prices of observed skill attributes affect overall inequality significantly. Thus, most of the changes in inequality appear to be attributable to changes in the residual, which captures changes in unobserved prices and quantities and, to a lesser extent, to changes in observed quantities. In other words, changes in the relative prices of observed skill attributes play only a small role in the evolution of wage inequality.

The Structure of Earnings

The discussion thus far has focused on the distribution of hourly wages. The cross-sectional dispersion of hourly wages could differ from monthly (or annual) earnings, depending on the covariance between monthly hours and hourly wages. For instance, high-wage workers might work (and get paid for) more hours per month than low-wage workers. Wage inequality would then be a downward-biased measure of earnings inequality.10 Measurement error in the hours variable is also a potential problem, especially for salaried workers. Therefore, it is useful to examine the evolution of earnings inequality as well.

Figure 7 shows cumulative changes in log gross monthly earnings at different percentiles of the distribution. Similar to the pattern observed for changes in wage inequality, there appears to have been a slight increase in earnings inequality over the period 1984–97, with much of this increase occurring after 1992. A variance decomposition of earnings inequality indicated that the variances of hourly wages and monthly earnings look quite similar, while the variance of hours worked is small and is roughly offset by the covariance component (Figure 8). The data reveal only a small cross-sectional covariance between hourly wages and hours worked. Using an alternative measure of hours, the sum of contracted monthly hours plus overtime, yields similar results.

Figure 7.Changes in Log Monthly Earnings Across Distribution

Figure 8.Variance Decomposition for Earnings

Overall, the basic picture of stability in the wage structure is reinforced by the stability of the structure of earnings.11 The results in this section also indicate that measurement error in the weekly hours variable used to construct the hourly wage measure is unlikely to be driving any of the earlier results.

II. Additional Robustness Tests

Cohort Effects

Cross-sectional measures of wage inequality could be affected by changes over time in the observed and unobserved attributes of cohorts that enter the labor market at different periods. For instance, changes in inequality could be dampened by the increasing equalization of educational opportunities for workers in more recent cohorts. Furthermore, inequality changes over time within cohorts (as employers gain more information about workers based on job histories) could influence measures of overall inequality, especially if cohort sizes change over time. It is difficult to disentangle cohort, experience, and time effects. Nevertheless, by examining changes in inequality over time for different cohorts and different experience groups, it is possible to get an indication of whether cohort and age effects are important for understanding the evolution of overall wage inequality.

For this part of the analysis, I constructed synthetic cohort groups based on the imputed year of market entry for each worker.12Table 7 shows three-year averages of the 90-10 and 75-25 percentile differentials for each cohort, centered on the years 1985, 1991, and 1996. For instance, the cohort with market entry between 1977 and 1981 had a90-10 differential of 0.81 in 1985, which declined to 0.72 in 1991 and then rose to 0.76 by 1996. These results should be interpreted with caution since the samples are relatively small (typical cell size: 250–400; minimum cell size: 100). Nevertheless, despite some small changes over time in inequality within cohorts, there is little evidence that these changes, or the differences in inequality across cohorts, are an important factor in explaining the apparent stability of the wage structure.

Table 7.Wage Inequality Across Cohorts and Experience Groups
90-10 Differential75-25 Differential
Year of Market Entry198519911996198519911996
1987–910.810.41
1982–860.740.780.370.39
1977–810.810.720.760.410.390.40
1972–760.690.750.830.360.370.41
1967–710.740.730.710.360.360.36
1962–660.760.800.840.360.410.44
1957–610.760.790.800.410.450.44
1952–560.700.590.660.330.320.32
1947–510.790.750.370.36
1942–460.700.35
Sources: German Socio-Economic Panel and author’s calculations.Note: The percentile differentials reported above are three-year averages centered on years shown.
Sources: German Socio-Economic Panel and author’s calculations.Note: The percentile differentials reported above are three-year averages centered on years shown.

Note that the evolution of inequality within specific (synthetic) experience groups can be tracked by reading diagonally across this table. For instance, the experience group corresponding to the 1977–81 entry cohort, which has four to eight years of experience in 1985, has a 90-10 differential of 0.81 in 1985,0.74 in 1991, and 0.81 in 1996. Within experience groups, there is a pattern of a small dip in the 90-10 differential in 1991, followed by an uptick in 1996. This is consistent with the pattern detected earlier of marginal wage compression in the 1980s, followed by a slight widening of wage dispersion after 1992.

Thus, changes in inequality within age and cohort groups appear largely consistent with the patterns of overall wage variation. In other words, time effects appear to be more important than age or cohort effects per se in explaining changes in the wage structure.

Supplementary Earnings

Various forms of monetary compensation other than basic wages and salaries constitute an important element of compensation packages in Germany. Such payments could play an important role in differentiating total compensation across workers of different skill levels but would not be picked up in data on monthly wages and salaries. In the GSOEP, workers are asked about the gross amounts of different categories of nonstandard compensation that they received in the previous year. Using these data, 1 constructed for each individual a wage adjustment factor to incorporate supplementary earnings, using the following formula:

The adjustment factor turns out to be quantitatively important. Its distribution over the period 1990–97, shown in the tabulation below, indicates that the median supplementary income amounted to about 8.3 percent of the basic wage. There was no discernible trend over time in this adjustment factor. However, regressions of this factor on skill attributes did indicate a statistically significant positive relationship between the size of this factor and skill level, suggesting that total compensation could be more differentiated than basic wages.

Distribution of Adjustment Factor for Supplementary Income, 1990–97
Percentile point:5102550759095
Adjustment factor:0.0000.0160.0460.0830.0970.1290.167

For each worker for whom the relevant data were available, 1 constructed a new wage variable, where the current year wage was multiplied by (1 plus) this adjustment factor, OLS estimates of the returns to experience and education were generally marginally higher, while returns to tenure were generally marginally lower using (logarithms of) this wage measure compared with the estimates based on the basic wage. However, the differences were quite small in economic terms. More importantly, the time profiles of the returns to skill attributes were not altered when the adjusted wage variable was used. Plots of wage changes at different percentiles over the period 1990–97 (not shown here) were also essentially unaffected by the use of this alternative wage measure. Thus, although there is some evidence that total compensation is more differentiated by skill level than basic wages, the structure of total compensation is essentially as stable as the structure of basic wages over the period 1990–97.

Selection Effects

The sensitivity of the results to sample selection bias was also examined. Since wages are observed only for those workers who are employed, wage regressions could be subject to bias induced by systematic differences in unobserved characteristics of employed versus nonemployed persons. In other words, the observed wage distribution may be a biased measure of the offer wage distribution. Further, the magnitude of selection bias could vary systematically across skill levels, thereby biasing estimated wage differentials and changes over time in these differentials across skill levels (Keane and Prasad, 1996).

Although sample selection effects are likely to be less important for men than for women, I estimated selection-corrected wage equations for the sample of males. To conserve space, the results are summarized only briefly here. The selection-corrected coefficient estimates for the education and experience variables were very similar to those from the OLS regressions. The estimated education premia were slightly higher in the 1980s in the selection-corrected models compared with the OLS results. By the mid- 1990s, however, the effects of this correction are close to zero. The returns to experience are also only marginally affected by the selection correction. The basic story of stable skill premia is thus confirmed by these results.

III. The Role of Market Forces in the Stability of the German Wage Structure

The empirical results thus far have demonstrated the relative stability of the West German wage structure over the period 1984–97. This section explores some possible explanations for this remarkable stability during a period when all major industrial economies have been going through massive shifts in the relative demand for skills resulting from skill-biased technological change, increased openness to external trade, and shifts in employment and output shares from manufacturing toward services. In what follows, particular attention is paid to the roles of shifts in relative supplies of skilled and unskilled workers and in the sectoral composition of employment.

Relative Supply Shifts

Changes in wage inequality that are attributable to changes in skill prices can be analyzed in terms of a supply and demand framework for different skill attributes. For instance, Katz and Murphy (1992) note that, despite an increase in the relative demand for skilled workers, wage inequality did not increase substantially in the United States in the 1970s, since the relative supply of workers with high education levels rose substantially and offset much of the shift in demand. Despite continuing increases in the relative supply of highly educated workers, however, enormous shifts in the relative demand for skilled labor in the 1980s resulted in sharp increases in observed skill premia.13

Is there evidence that shifts in relative skill supplies may have resulted in the stable wage structure observed in Germany? Average education levels in West Germany have indeed been rising over the last two decades, and the relative supply of college graduates, in particular, has increased significantly. In the GSOEP sample used here (the wage analysis sample), for instance, the cross-sectional average of the education variable increases from 10.9 years in 1984 to 11.7years by 1997. Figure 9 shows that this increase was largely the result of a small increase in the proportion of workers with university degrees and a corresponding decline in the proportion of workers with only general schooling. The relative sizes of the other two groups—those with an apprenticeship and with other vocational training—remained fairly stable. In the full GSOEP sample for West Germany that includes employed and nonemployed men and women, the relative proportion of workers with general schooling fell by about 6 percentage points, while the other three groups had increases of 2–3 percentage points each (Figure 9). Could this supply effect explain the absence, in Germany, of the marked increase in the returns to education in the 1980s and 1990s that was witnessed in other industrial countries with more “flexible” labor markets? A cross-country perspective suggests an answer in the negative. The relative supply of more educated and, especially, college-educated workers has been rising at roughly similar, and often higher, rates in most other major industrial countries as well.

Figure 9.Means of Skill Proxies: Full GSOEP Sample

Cross-country comparisons of educational levels are notoriously difficult, but 1 used ostensibly comparable data from the OECD Education Statistics to obtain some suggestive evidence. The tabulation below shows the ratio of (a) graduates of higher education (university and nonuniversity) to the total of (a) plus (b) graduates of upper secondary education (general and vocational/technical) in the population.14 Although the increase in this ratio over the period 1985–97 was 5.4 percentage points in Germany compared with 3.8 percentage points in the United States, this difference seems hardly sufficient to explain the huge disparities in the evolutions of premia for higher education in these two countries. Examinations of other such ratios revealed a very similar picture.

Ratio of Workers with High Relative to Medium Levels of Education
198519901992
Germany0.1750.2160.229
United States0.4060.4280.444

A more direct approach, following Gottschalk and Joyce (1998), is to examine labor market quantities, that is, unemployment and employment of workers of different skill levels. If there were indeed relative shifts in the supplies of workers with different skill levels, this would be reflected in quantities rather than just prices. The first panel of Figure 10 plots unemployment rates for workers with different skill levels.15 Clearly, unemployment rates for workers of different skill levels in West Germany have diverged markedly during the 1980s and 1990s. More strikingly, unemployment rates for unskilled workers have risen sharply during the 1990s, while the increases in unemployment rates have been much smaller for medium-skilled workers and have in fact fallen for highly skilled workers during the recent cyclical recovery that began around 1993.

Figure 10.Unemployment Rates and Employment by Skill Level

One cautionary note about interpreting these unemployment rates is that they could reflect the effects of German unification. From the West German perspective, unification was essentially a labor supply shock that was accentuated in the lower portions of the skill distribution and that may have resulted in the observed increases in unemployment rates for low-skill workers. However, in conjunction with the earlier results on the stability of skill premia, this outcome—persistent increases in absolute and relative unemployment rates for low-skill workers—is precisely what one would expect if a rigid wage structure prevented labor market adjustment through the adjustment of relative prices.

Stronger evidence for this interpretation comes from an examination of employment levels. As shown in the second panel of Figure 10, employment levels for workers of different skill levels in West Germany have diverged steadily since the mid-1970s. During the 1990s, employment levels of high-skill workers have risen sharply even as employment for unskilled workers has actually declined.16 This evidence is difficult to reconcile with a story that relies on changes in the supplies of different skill categories to explain the stability of the wage structure as an equilibrium outcome.

In short, there is little evidence that shifts in relative supplies of workers with different skill levels can explain observed relative wage developments. Furthermore, the evolutions of relative unemployment rates and employment levels are strongly suggestive of the notion that, in the presence of institutional constraints that inhibit relative price adjustment, relative shifts in the demand for skills have resulted in quantity adjustments.

Shifts in Sectoral Employment Shares

As in other industrial economies, in recent decades there has been a secular decline in the employment share of manufacturing and a corresponding increase in the employment share of the service sector in Germany. This and other cyclical shifts in sectoral employment could influence the overall wage structure since average wages and the dispersion of wages are likely to be quite different across sectors. These two channels through which changes in the structure of sectoral employment could affect the wage structure are also likely to be influenced by the effects of changing skill compositions of the workforce in these sectors.

One way to analyze the effects of sectoral shifts on the wage structure is to use a simple variance decomposition. The total variance of wages in a year can be decomposed into within- and between-industry components as follows:

where σt2 is the cross-sectional variance of log hourly wages, sjt, is the employment share of sectorjσjt2 is the within-industry variance of wages, wjt, is the mean sectoral wage, ¯wt is the mean wage in the sample, and the subscript / is a time index. Using this formula, the change in variance over time can be decomposed into changes attributable to within- and between-industry variance as well as composition effects within and between industries. The results of this decomposition are shown in Table 8.

Table 8.Effects of Sectoral Shifts on Changes in Wage Inequality(Variance decomposition)
Within IndustryBetween Industry
PeriodTotal ChangeChange inCompositionChange inComposition
in Variancevarianceeffectvarianceeffect
1985–960.460.350.050.020.05
1985–89-0.01-0.030.06-0.01-0.02
1989–92-0.15-0.140.01-0.01-0.01
1992–960.620.52-0.020.030.09
Sources: German Socio-Economic Panel and author’s calculations.Notes: Workers are classified into 10 broadly defined sectors (agriculture, forestry, and fishing; Utilities and mining; manufacturing; construction; trade; transport and communications; finance and insurance; business and personal services; other basic services; public administration). The numbers reported in this table are changes in three-year averages, centered on the years shown, of total wage variance and its components. All numbers in this table were multiplied by 100.
Sources: German Socio-Economic Panel and author’s calculations.Notes: Workers are classified into 10 broadly defined sectors (agriculture, forestry, and fishing; Utilities and mining; manufacturing; construction; trade; transport and communications; finance and insurance; business and personal services; other basic services; public administration). The numbers reported in this table are changes in three-year averages, centered on the years shown, of total wage variance and its components. All numbers in this table were multiplied by 100.

The total increase of 0.0046 in overall wage variance from 1984 to 1997 is the result of a marginal decline in variance during 1989–92, which is more than offset by an increase during 1992–97. A substantial fraction of the changes in overall wage variance is attributable to changes in within-industry rather than between-industry wage variation. Composition effects, both within and between industries, account for only a small fraction of the changes in variance. The contribution of the between-industry component is also quite small. Thus, shifts in sectoral employment do not seem to have played much of a role in influencing patterns of overall wage dispersion. Within-industry wage variation appears to dominate overall wage variation, and both appear to have evolved in a smaller pattern.

IV. Discussion

The results presented thus far strongly suggest that the stability of the German wage structure is attributable to constraints imposed by institutional factors rather than to market forces. As discussed earlier, it appears that the wage bargaining system and the role of unions have resulted in an inflexible wage structure that does not allow prices for skills to respond to shifts in the demand for and supply of different skills.

As noted by Fortin and Lemieux (1997) among others, it may generally not be appropriate to attribute the behavior of residuals from such an empirical analysis to “institutional factors.” They argue that direct measures of institutional factors and empirical analysis using such measures are required to make such statements. Such an approach is vitiated in this context since there have been few significant institutional changes over the sample period that could help identify the direct effects of these factors. The approach in this paper has instead been to take seriously and analyze the effects of all observable “market forces” that could potentially explain the evolution of the wage structure. Given the results in this paper, the observed price and quantity outcomes in the labor market, and the wealth of anecdotal evidence about the rigidities induced by labor market institutions in Germany, the case for institutional factors playing a dominant role appears quite strong.

Some authors have argued that the relatively narrow dispersion of wages is attributable to the tighter distribution of skills in Germany compared with countries like the United Kingdom or the United States (e.g., Nickell and Bell, 1996). The German wage structure is also viewed by some as providing incentives for firms to provide optimal levels of training to their low-skill workers (e.g., Acemoglu and Pischke, 1999). Nevertheless, the rigidity of the wage structure during a period of massive shifts in demand toward the upper end of the skill distribution has had obvious deleterious consequences, as is evident from the rising nonemployment rates and declining employment levels for unskilled workers, concomitant with declining nonemployment rates and rising employment levels for skilled workers. As discussed earlier, these divergent trends were further accentuated in the mid-1990s. It should be borne in mind that the central argument in this paper is based not so much on the levels of wage differentials as on the inability of the wage structure to adjust to demand shifts over time.

A Synthesis of Microeconomic and Macroeconomic Evidence

This subsection discusses the relationship between the microeconomic evidence presented in this paper and macroeconomic data related to the labor market. As in other continental European economies, the wage share in Germany has declined markedly since the mid-1970s (Figure 11).17 By contrast, in the United Kingdom and the United States, the wage share has remained largely unchanged over the last three decades, abstracting from some year-to-year variation in the data. Also, direct evidence of the substitution of capital for labor can be gleaned from an examination of the capital-output ratio in Figure 11. This ratio trended upward during the 1980s and, after a unification-related spike in 1991, continued to rise, although much more gradually, in the 1990s.18

Figure 11.Aggregate Data

Blanchard (1997) has argued that the pattern of declining wage shares—which he documents for France, Germany, and Italy—can be explained by institutional rigidities that have perpetuated the effects of adverse macroeconomic shocks and that have resulted in the substitution of capital for labor and, consequently, rising aggregate unemployment rates. Blanchard and Wolfers (1999) take this argument further by trying to demonstrate that the interactions of shocks and institutions can explain differences in the evolutions of unemployment rates across industrial countries.

In Germany, the widening disparities in employment outcomes for skilled and unskilled workers indicate an important additional dimension to this story. A crucial ingredient for a comprehensive explanation is that of capital-skill complementarity.19 Given a rigid wage structure that prevents relative skill prices from adjusting to demand shifts, capital-labor substitution—as has been observed in Germany—is indeed the likely outcome. But, particularly given capital-skill complementarity, this could imply an increase in the demand for skilled relative to unskilled labor and a concomitant net decline in the total demand for labor. Since low-skill labor accounts for a much greater fraction of the total labor force than high-skill labor, the net effect could be to raise aggregate unemployment.

More generally, skill-biased technological change would tend to shift out the relative demand for skilled labor. If the skill premium was constrained to remain unchanged, however, the net effect would be to lead to a substitution of both capital and skilled labor for unskilled labor in the aggregate production function. Capital-skill complementarity would accentuate this effect. A production structure that formalizes this discussion, and shows analytically how capital-skill complementarity can interact with a rigid wage structure to produce these differences in employment effects for skilled and unskilled labor, is presented in the Appendix.

Making the reasonable assumption that industrial economies have been subject to similar shifts in the relative demand for skills in recent decades (Machin and Van Reenen, 1998; Manacorda and Petrongolo, 1999), this discussion suggests that the price and quantity outcomes across different countries can be viewed in part as being influenced by tabor market institutions that affect the evolution of skill prices (Blau and Kahn, 1996, reach a similar conclusion). In the United States, where there are few constraints on wage differentials, relative skill prices have adjusted in response to the asymmetric demand shifts for skills, as evidenced by rising wage inequality and increasing skill premia. In Germany, by contrast, quantities have had to bear the brunt of adjustment, reflected in the rising and persistent disparities in employment and unemployment rates between high-skill and low-skill workers.

Thus, the microeconomic evidence presented in this paper is helpful in understanding the rising disparities in labor market outcomes for workers of different skill levels as well as patterns in macroeconomic data, including developments in the wage share, capital-labor ratios, and aggregate unemployment.

V. Conclusions

This paper has produced two main empirical results. One is that the West German wage structure was quite stable over the period 1984–97, especially in comparison to countries with more “flexible” labor markets such as the United Kingdom and the United States. Inequality both within and between different skill groups declined slightly in the 1980s and increased marginally during the mid-1990s. Returns to skill attributes such as education, experience, and tenure show a similar pattern. The second result is that this stability of the wage structure is not attributable to shifts in the relative supply of skills or other factors such as shifts in the sectoral distribution of employment, or cohort and selection effects.

Taken together, these results suggest that institutional factors, including the wage bargaining system, appear to have fostered a relatively rigid wage structure that has not responded to shifts in the relative demand for skills. As evidenced by patterns of employment growth and evolutions of unemployment rates for different skill groups, unskilled workers essentially appear to have been priced out of their jobs because of the inflexible wage structure that has not accommodated shifts in labor demand toward the upper end of the skill distribution. Furthermore, skill price rigidities appear to have encouraged capital-labor substitution, with detrimental effects on the employment probabilities of unskilled workers.

The paper has also argued that the micro evidence presented here is important for understanding and interpreting patterns in macroeconomic data, including the evolutions of the aggregate unemployment rate, the wage share, and the capital-labor ratio.

APPENDIX

This appendix describes a production structure that formalizes the statements made in the text about the effects of relative demand shifts (for different types of skill) and capital-skill complementarity on the equilibrium skill premium. The potential employment effects of rigidities that prevent adjustment in the skill premium in response to different shocks are also discussed.

Krusell and others (2000) propose the following production function with four inputs—capital structures, capital equipment, skilled labor, and unskilled labor. The production function is Cobb-Douglas over capital structures (Ks) and a constant elasticity of substitution (CES) aggregate of the three remaining inputs. The specification they find to be consistent with U.S. data is as follows:

The parameters σ and ρ determine the elasticities of substitution among capital equipment (Ke.), skilled labor (S), and unskilled labor (U). Inputs of skilled and unskilled labor may be considered as the products of aggregate hours (hi) and an efficiency index (ψr), where i is an index for skill type.

There are a couple of reasons for splitting capital into two types. First, the phenomenon of capital deepening in many industrial economics in recent years is largely attributable to equipment investment (including computers) rather than investment in structures. Second, it is not obvious that skilled and unskilled labor would have different degrees of substitutability with structures, while differences in substiutability with capital equipment are more plausible.

Note that this production function specification implies that the elasticity of substitution between equipment and unskilled labor is the same as that between skilled and unskilled labor. This restriction follows from the symmetry property of the CES aggregation and is consistent with empirical estimates of these elasticities. The elasticity of substitution between capital equipment and unskilled labor is 1/(1 - σ) and that between capital and skilled labor is 1/(1 - ρ). Hence, setting σ > ρ implies capital-skill complementarity.

Under the assumption that factor prices are equal to marginal products (per unit of raw input), the skill premium is as follows:

The key point to note here is that capital equipment is a determinant of the skill premium. In the absence of capital-skill complementarity, σ = ρ and equation (A2) simplifies to

In this case, the skill premium is determined solely by the relative labor inputs, which in turn are a function of raw inputs (i.e., hours) and the respective efficiency indexes. To see the implications of capital-skill complementarity for the skill premium in the event of capital deepening, the skill premium can be differentiated with respect to Ke:

It is apparent that if σ = ρ, the right hand side of equation (A4) is zero and, hence, capital deepening has no implications for the skill premium. For the case where σ > ρ. however, the equilibrium skill premium is increasing in Ke.

If Ke were to increase, the equilibrium skill premium rises to π*. If π* > π but π were constrained to remain unchanged, however, the requirement for equation (A2) to hold would be for S to rise and U to fall (in relative terms), with the relative shifts for the demands of the two types of labor depending upon the model parameters. In either case, there would be an increase in the relative demand for skilled workers at given skill prices.

The phenomenon of skill-biased technological change may also imply an increase in Ϩst relative to ψut20 Even in the absence of capital-skill complementarity, as can be seen from equation (A3), this would imply an increase in the equilibrium skill premium. Again, if the skill premium was constrained to remain unchanged, equation (A3) would be satisfied by reducing hut and/or increasing hst, implying an increase in the relative demand for skilled labor.

REFERENCES

    Abraham,KatharineG., andSusan N.Houseman,1995, “Earnings Inequality in Germany,”Differences and Changes in Wage Structure, ed. byRichard B.Freeman andLawrence F.Katz (Chicago: University of Chicago Press), pp. 371–404.

    • Search Google Scholar
    • Export Citation

    Acemoglu,Daron, andJôrn-SteffenPischke,1999, “Beyond Becker: Training in Imperfect Labour Markets,”Economic Journal, Vol. 109 (February), pp. F112–42.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Blanchard,OlivierJ.,1997, “The Medium Run,”Brookings Papers on Economic Activity.2, Brookings Institution, pp. 89–158.

    Blanchard,OlivierJ., andJustinWolfers,1999, “The Role of Shocks and Institutions in the Rise of European Unemployment: The Aggregate Evidence,”NBER Working Paper No. W7282 (Cambridge, Massachusetts: National Bureau of Economic Research).

    • Search Google Scholar
    • Export Citation

    Blau,FrancincD., andLawrence M.Kahn,1996, “International Differences in Male Wage Inequality: Institutions versus Market Forces,”Journal of Political Economy, Vol. 104 (August), pp. 791–837.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Buchinsky,Moshe,1994, “Changes in the U.S. Wage Structure 1963–1987: Application of Quantile Regression,”Econometrica, Vol. 62 (March), pp. 405–58.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Burkhauser,Richard, andJohn G.Poupore,1997, “A Cross-National Comparison of Permanent Inequality in the United Stales and Germany,”Review of Economics and Statistics, Vol. 79 (February), pp. 10–17.

    • Crossref
    • Search Google Scholar

    Card,David,FrancisKramarz, andThomasLemieux,1999, “Changes in the Relative Structure of Wages and Employment: A Comparison of the United States, Canada, and France,Canadian Journal of Economics, Vol. 32 (August), pp. 843–77.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Fortin,NicoleM., andThomasLemieux,1997, “Institutional Changes and Rising Wage Inequality: Is There a Linkage?”Journal of Economic Perspectives, Vol. 11 (Spring), pp. 75–96.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Freeman,Richard, andRonaldSchettkat,2000, “The Rote of Wage and Skill Differences in US-German Employment Differences,”NBER Working Paper No. 7474 (Cambridge, Massachusetts: National Bureau of Economic Research).

    • Search Google Scholar
    • Export Citation

    Goldin,Claudia, andLawrence F.Katz,1998, “The Origins of Capital-Skill Complementarity,”Quarterly Journal of Economics, Vol. 113 (August), pp. 693–732.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Gottschalk,Peter, andTimothy M.Smeeding,1997, “Cross-National Comparisons of Earnings and Income Inequality,”Journal of Economic Literature, Vol. 35 (June), pp. 633–87.

    • Search Google Scholar
    • Export Citation

    Gottschalk,Peter, andMaryJoyce,1998, “Cross-National Differences in the Rise in Earnings Inequality: Market and Institutional Factors,”Review of Economics and Statistics, Vol. 80 (November), pp. 489–502.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Gottschalk,Peter, andSheldonDanziger,2003, “Wage Inequality, Earnings Inequality and Poverty in the U.S. Over the Last Quarter of the Twentieth Century” (unpublished; Boston, Massachusetts: Boston College).

    • Search Google Scholar
    • Export Citation

    Gregg,Paul, andAlanManning,1997, “Skill-Biased Change, Unemployment and Wage Inequality,”European Economic Review, Vol. 41 (June), pp. 1173–1200.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Griliches,Zvi,1969, “Capital-Skill Complementarity,”Review of Economics and Statistics, Vol. 51, No. 4, pp. 465–68.

    Hunt,Jennifer,1999, “Has Work-Sharing Worked in Germany,”Quarterly Journal of Economics, Vol. 114 (February), pp. 117–48.

    Jaeger,Albert,1999, “Institutional Change and Economic Performance in Germany: A Fifty-Year Perspective” (unpublished; Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    Juhn,Chinhui,Kevin M.Murphy, andBrooksPierce,1993, “Wage Inequality and the Rise in Returns to Skill,”Journal of Political Economy, Vol. 101 (June), pp. 410–42.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Katz,LawrenceF., andKevin M.Murphy,1992, “Changes in the Wage Structure: Supply and Demand Factors,”Quarterly Journal of Economics, Vol. 107 (February), pp. 35–78.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Keane,MichaelP., andEswar S.Prasad,1996, “The Employment and Wage Effects of Oil Price Changes: A Sectoral Analysis,”Review of Economies and Statistics, Vol. 78 (August), pp. 389–400.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Krusell,Per,LeeE. Ohaman,José-VíctorRíos-Rull, andGiovanniViolante,2000, “Capital-Skill Complementarity and Inequality; A Macroeconomic Analysis,”Econometrica, Vol. 68, No. 5 (September), pp. 981–95.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Machin,Stephen, andJohnVan Reenen,1998, “Technology and Changes in Skill Structure: Evidence from Seven OECD Countries,”Quarterly Journal of Economics, Vol. 113 (November), pp. 1215–44.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Manacorda,Marco, andBarbaraPetrongolo,1999.“Skill Mismatch and Unemployment in OECD Countries,”Economica, Vol. 66 (May), pp. 181–207.

    Nickell,Stephen, andBrianBell,1996, “Changes in the Distribution of Wages and Unemployment in OECD Countries,”American Economic Review, Vol. 86 (May), pp. 302–14.

    • Search Google Scholar
    • Export Citation

    Organization for Economic Cooperation and Development, OECD Education Statistics, various years (Paris).

    • Search Google Scholar

    Prasad,EswarS.,2000, “The Unbearable Stability of the German Wage Structure: Evidence and Interpretation,”IMF Working Paper 00/22 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    Prasad,EswarS.,2002, “Wage Inequality in the United Kingdom: 1975–99,”IMF Staff Papers, Vol. 49, No. 3, pp. 339–63.

    Reinberg,Alexander, andAngelaRauch,1998, “Bildung und Arbeitsmarkt: Der Trend zur Hoheren Quaiifikation ist Ungebrochen.”IABWerkstattbericht, No. 15, pp. 1–15.

    • Search Google Scholar
    • Export Citation

    Siebert,Horst,1997, “Labor Market Rigidities: At the Root of Unemployment in Europe,”Journal of Economics Perspectives, Vol. 11, No. 3, pp. 37–54.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Steiner,Viktor, andKerstenWagner,1997, “Relative Earnings and the Demand for Unskilled Labor in West German Manufacturing,”ZEW Discussion Paper No. 97–17 (Mannheim: Centre for European Economic Research).

    • Search Google Scholar

    Steiner,Viktor,1998, “Has Earnings Inequality in Germany Changed in the 1980s?”Zeitschrift für Wirtschafts and Soziahvissenschaften, Vol. 118, No. 1, pp. 29–59.

    • Search Google Scholar
    • Export Citation

    van derWiiligen,Tessa,1995, “Unemployment, Wages, and the Wage Structure,”United Germany: The First Five Years, IMF Occasional Paper No. 125 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
*

Eswar Prasad is Chief of the China Division in the Asia and Pacific Department of the IMF He was in the Research Department when this paper was written. The author thanks Albert Jaeger for extensive discussions and an anonymous referee whose constructive comments helped sharpen the exposition in the paper. This paper has benefited greatly from the comments of Orazio Attanasio, Robert Flood, Jennifer Hunt, Andrei Kirilenko, Axel Schimmelpfennig, numerous colleagues, and participants at various seminars.

1

See van der Willigen (1995) fora description of the wage bargaining structure and Jaeger (1999) for a discussion of how it may have been well suited to the Wirtschaftswunder era of the 1960s and 1970s.

2

Part-time workers and apprentices account for a relatively small fraction of the sample, and including them did not have much effect on any of the results discussed below.

3

As noted by Hunt (1999), using the sum of the contracted weekly hours and overtime hours variables is problematic. This sum would not capture “under-time,” as only positive overtime hours are reported in the survey.

4

To abstract from any distortions that might arise from year-to-year variation in the results, many of the figures and tables in the remainder of the paper report results for three-year moving averages of the relevant statistics centered on the years shown, although all statistics are first computed using the underlying annual data.

5

The top and bottom 5 percentiles have been trimmed out in the figures. A filled regression line for the cumulative wage changes across percentile points is shown in each panel.

6

Source for U.K. data: Prasad (2002). Source for U.S. data: Gottschalk and Danziger(2003). Note that the data cover a longer time period for the United States than for the other two countries. Much of this uneven growth of real wages across the U.S. wage distribution appears to have taken place in the 1970s and 1980s rather than the 1990s. Also see Gottschalk and Smeeding (1997).

7

The inclusion of higher order polynomials of experience left the results essentially unchanged. Coefficients on polynomials of the tenure variable were also small and statistically insignificant. Industry or occupation dummies are not included in these regressions. Since individuals typically self-select into industries and occupational groups, inclusion of these dummies could induce substantial bias in estimated skill premia.

8

Results for the tenure variable may be found in Prasad (2000, Table 6). Alter declining mildly in the latter half of the 1980s, the returns to lenure at all quantile points began to rise after the rnid-1990s. The returns to tenure appear to have strengthened more at the lower quantiles of the distribution toward the end of the sample period.

9

Using Current Population Survey (CPS)data for the United Stales, Buchinsky (1994) also finds a large amount of year-to-year variation in the returns to education and experience at different quantile points.

10

In addition, the dispersion of annual earnings could differ from that of monthly earnings. However, the GSOEP data set does not contain a variable indicating the number of months that a worker is employed in the survey year.

11

Using GSOEP data. Abraham and Houseman (1995) and Steiner and Wagner (1998) report similar findings of a stable dispersion of gross monthly earnings during the 1980s.

12

Cohorts defined on the basis of birth year yielded similar results.

13

Katz and Murphy (1992) construct proxies for relative demand shifts using shifts in the mix of industry-occupation classifications and the relative proportions of skilled and unskilled workers within these industry-occupation cells. Unfortunately, preliminary calculations indicated that the GSOEP does not have enough data available (as reflected in the cell sizes) for such an exercise to yield reliable results.

14

Source: OECD Education Statistics, 1985–92, Table IV.3. These ratios can also be calculated (1985, 1992) for certain other countries including Canada (0.623. 0.702). Italy (0.191, 0.182), Japan (0.349, 0.348), and the Netherlands (0.280, 0.260). Unfortunately, the relevant data for France and the United Kingdom are not available.

15

The data for this figure (limited to West Germany) are taken from Reinberg and Rauch (1998) and are based on the Mikrozensus, a more comprehensive survey of the German labor force than the GSOEP. Skill levels are defined on the basis of a number of observed characteristics including education levels and occupational categories. The raw data from this survey are not publicly available. GSOEP data revealed very similar patterns.

16

To examine the evolution of group-specific employment rules. I estimated annual probit employment equations for men (extending the sample to include men without a job). The estimated coefficients continued the sharp increase in the employment probabilities of workers with higher levels of education during the 1990s.

17

These data are taken from Statisttsches Tuachenbuch 1998: Arbeits and Sozialstatistik (Bundesministcrium fur Arbeit und Sozialordnung). Note that the numbers refer to United Germany starting in 1991 and to West Germany before that. Assuming plausible elasticities of substitution between capital and labor, the labor supply shift caused hy unification cannot by itself explain anything close to the observed trend decline in the wage share.

18

These data were obtained from the German Ministry of Finance. Since aggregate employment has grown by much iess than output growth in the 1980s and 1990s, the increase in the capital-output ratio is probably a downward-biased measure of the increase in the capital-labor ratio. Recent developments in investment and output suggest that the capital-output ratio shown through 1995 in Figure 11 may have risen further since then.

19

For evidence on capital-skill complementarity, see Griliches (1969) and Goldin and Katz (1998). Krusell and others (2000) argue that capital-skill complementarity is important for understanding changes in U.S. wage inequality.

Other Resources Citing This Publication