Earnings Dynamics and Volatility in Europe

Transcription

1 Earnings Dynamics and Volatility in Europe ** * Aedín Doris*, Donal O Neill & Olive Sweetman Very Preliminary Draft ** Corresponding author: National University of Ireland Maynooth and IZA, Bonn and corresponding author. donal.oneill@nuim.ie, (tel : fax ; address Rhetoric House, NUI Maynooth, Maynooth, Co. Kildare, Ireland. * National University of Ireland, Maynooth. We gratefully acknowledge the financial support provided by the Irish Research Council for the Humanities and Social Sciences. 1

2 1. Introduction In recent years a number of studies, mostly in the US, have documented changes in earnings instability over time (Moffitt and Gottschalk (1995, 2002, 2011), Dickens (2000), Haider (2001), Ramos (2003), Baker and Solon (2003), Capellari (2004), Gustavsson (2004), Kalwij and Alessie (2007), Daly and Valetta (2008), Myck et al (2008), Dynan et al (2008), Jenkins (2011) and Shin and Solon (2011)). As noted by Moffitt and Gottschalk (2011), this interest in instability arises partly because of the different response of consumption to permanent and transitory changes in earnings and also because the welfare losses associated with income changes are likely to differ depending on whether those changes are permanent or transitory. In this paper we contribute to this literature by examining earnings volatility across twelve European countries. Our results in Section 3 show significant differences in volatility across countries. In Section 4 we examine the extent to which differences in labour market conditions, human capital or general macroeconomic conditions can explain the differences in volatility across countries. We find evidence that labour market institutions and the macroeconomic environment can help explain some of the observed differences. The availability of longer panel data for the UK and Germany allows us to analyse long run and more recent trends in volatility in these two countries. We show that, despite significant increases in total inequality in these two countries, volatility remained relatively stable in recent decades. In the final section, we exploit the length of the German panel to compare two popular measures of earnings instability. The results show that the conclusions can differ substantially depending on the measure of instability adopted. 2

3 2. Measuring Earnings Instability In this paper, we focus on two popular approaches to measuring instability in earnings. The first approach was made popular by Moffitt and Gottschalk (1995, 2002, 2008) and used in many studies, including Dickens (2000), Haider (2001), Ramos (2003), Baker and Solon (2003), Capellari (2004), Gustavsson (2004), Kalwij and Alessie (2007) and Daly and Valetta (2008). This method decomposes total inequality into its permanent and transitory components using parametric models of earnings dynamics. This approach associates earnings instability with transitory shocks, where a shock is defined as transitory if it is mean-reverting. We refer to this approach as the MG approach. In a recent paper, Doris et al. (2011) discuss the identification of these models for a variety of data structures typically encountered in practice. Monte Carlo results show that the Generalized Method of Moments (GMM) estimator used to estimate these models often performs poorly in short panels, particularly when the persistence of transitory shocks is strong. Large biases arise in key parameters and predicted trends in permanent and transitory variances differ substantially from the true patterns. Furthermore, the standard errors on the estimated parameters are very large and are not well estimated by the asymptotic standard errors, making hypothesis tests unreliable. The paper concludes that for household panel data sets of typical sample size, panel lengths of the order of 20 years are needed to correctly identify the model. The second approach associates instability with earnings volatility, measured as the dispersion in year-to-year earnings changes (Dynan et al. (1997), Cameron and Tracy (1998), Dynarksi and Gruber (2007) and Shin and Solon (2011)). For brevity, we refer to this approach as the SS approach. An advantage of this approach is that only two years of data are needed to calculate this volatility measure. Shin and Solon 3

4 (2011) and Moffitt and Gottschalk (2011) provide very useful discussions of the conceptual differences between this approach and the parametric approach discussed above. One important difference is that earnings volatility as measured by the SS approach includes permanent shocks in addition to transitory shocks, whereas instability as measured by the MG approach, includes only the latter. Furthermore, an increase in the persistence of transitory shocks increases the MG measure, but reduces the SS measure. This has implications for comparisons of the two measures, as discussed in Section Data To compare earnings instability across European countries, we use the eight waves of the European Community Household Panel Data (ECHP) for Belgium, Denmark, France, Germany, Ireland, Italy, Netherlands, Portugal, Spain, and the UK, the seven waves available for Austria and the six waves available for Finland. 1 These data are the only available panels including comparable earnings variables across a range of European countries. The years covered by the survey are For each of the twelve countries, we construct unbalanced panels from the initial sample. These comprise all males aged who are neither in full-time education nor retired and who report earnings in any year. Our measure of earnings is total labour market earnings in the previous month, deflated using the OECD Consumer Price Index. Observations with earnings in the top and bottom 1% of the 1 We did not use data for Sweden or Greece. The Swedish component of the ECHP is not a panel and therefore is not suitable for this type of analysis, while the scheduling of interviews in the Greek surveys led to concerns about comparability with other countries. 4

5 sample are excluded. Table A1 of the Appendix gives actual sample sizes used in each year for each country in the ECHP analysis. Figure 1 shows the variance of log earnings in the samples used for our twelve ECHP countries. Ireland, Spain, the UK and Portugal have the highest inequality, while Denmark and Italy have the lowest levels. The only country showing a significant trend in inequality over this time period is Germany, where the variance of log earnings increased by 47%. These twelve countries differ in terms of labour market institutions and educational composition of the workforce, as well as in the macroeconomic environment. We investigate the impact of these variables on the volatility of earnings in Section 4. In particular we consider three indicators of the strength of labour market institutions: union coverage, 2 the degree of centralization of bargaining and gross benefit replacement rates in the first year of unemployment. These three variables are taken from the institutional data set developed by Nickell (2008). To measure macroeconomic conditions, we use the total unemployment rate, again taken from the Nickell data set. Since more highly educated workers may have a comparative advantage in implementing new technology (Bartel and Lichtenberg (1987)), education may reduce an individual s exposure to technology shocks, thereby reducing volatility in earnings. To examine this, we use the proportion of the population aged with third level education, taken from the OECD factbook, as 2 Union coverage is not available for any country for For Ireland, union coverage is not officially measured, so we approximated it as a fixed multiple of union density, using National Centre for Partnership and Performance survey data to calculate the appropriate scalar multiple. 5

6 3 our measure of human capital. Summary statistics and time series plots for each of these variables are given in Table A2 and Figure A1-A5 of the Appendix. With at most eight years of data in the ECHP, we are able to calculate the SS measure of earnings instability for the twelve European countries, but not the MG measure. For the UK and Germany, longer panel surveys are also available. For the UK, we use data from the British Household Panel Survey (BHPS), which covers the 17 years from ; for Germany, we use the German Socio-Economic Panel (GSOEP), which covers the 27 years from The BHPS and GSOEP are the parent surveys on which the ECHP was based. When using these longer surveys, we follow the same sample selection criteria as for the ECHP, described above. The German data cover the period during which West and East Germany were reunified; although the GSOEP was expanded to include East Germany in 1990, we use only individuals present in the survey prior to reunification. 4 The availability of these data allows us to compare long-run trends in volatility in two large European countries. For the BHPS, we cannot calculate the MG measure because the panel is still not long enough to reliably estimate the parameters of a model that incorporates persistence of transitory shocks, time varying variances and heterogeneous growth in the permanent component, features of earnings dynamics that have been found to be important in previous studies. However, the German dataset has the advantage that the panel is long enough to estimate the MG as well as the SS measure of earnings instability. 3 A similar measure was used by Checchi and Garcia-Penalosa (2008) in their study of aggregate inequality. 4 These correspond to Sample A, which includes West German residents and Sample B, which includes foreigners living in West Germany, 6

7 4. Results: Earnings Volatility In this section, we compare the SS measure of volatility across twelve European countries, conduct a descriptive analysis of the determinants of the differences found, and use longer panels to examine longer-run trends in Germany and the UK. 4a. Results for the ECHP We first report results on the SS measure of volatility for our twelve ECHP countries. To calculate this measure, we follow Shin and Solon (2011), and remove age effects by regressing the change in log earnings on a quadratic in age. Volatility is then measured as the root mean square error from this regression. The results are given in Figure 2. There are clear differences in the level of volatility in monthly earnings across countries. Ireland, Spain and the UK stand out as having high volatility, with a standard deviation of about Belgium and Denmark, on the other hand, have particularly low levels of volatility, with values of about As a point of comparison, Shin and Solon (2011) calculate US volatility in annual earnings to range from 0.4 to 0.5 over the same period. For most of the EU countries studied, volatility is stable over the period, which mirrors the pattern of stable total inequality shown in Figure 1 and discussed in Section 3 above. Two notable exceptions to the pattern of constant volatility are Austria and France, for which the SS measure declines by 28% and 25% respectively. There is no country for which volatility increases substantially. 7

8 4b. The Determinants of Volatility In attempting to explain differences in inequality across countries, previous studies have mainly focussed on the relationship between institutions and total inequality. 5 Checchi and Garcia-Peñalosa (2008) find that institutions affect inequality, but these effects largely disappear once country effects are included, while Bertola et al. (2002) find that countries with stronger labour market institutions have lower inequality. In this section, we consider the impact of institutions, human capital and macroeconomic conditions on volatility, rather than on total inequality. To estimate the determinants of volatility we consider the following regression: SS = α + β X + e (1) it i it it where the dependent variable is the SS measure of volatility for country i at time t and the X it are the institutional, human capital and macroeconomic variables discussed in Section 3. We estimate Equation 1 using the random effects estimator. Although the assumption implicit in the random effects model of independence of country effects and the X it may not be valid, the alternative of estimating a fixed effects model is not feasible given the limited time variation of the independent variables over the eight years, as shown in Figures A1-A5. Table 1 reports the results of the estimation of Equation 1. Of the regressors considered, two factors appear to be highly correlated with volatility. These are the level of union coverage and the unemployment rate, which are both significant in our 5 Some papers have considered the impact of institutions on permanent and transitory inequality within the MG framework; these include Cervini and Ramos (2008), which examines the role of temporary contracts in Spain, Gustavsson (2007), which considers the role of wage bargaining in Sweden and Sologon and O Donoghue (2009), which assesses the impact of a range of institutional factors. 8

9 estimation. The results show that higher union coverage is negatively associated with volatility. This is to be expected as the rigid wage structure associated with unionnegotiated contracts limits the scope for changes in individual earnings, thereby reducing volatility. On the other hand, there is a positive association between unemployment and volatility. This counter-cyclical pattern in volatility has been noted in previous studies in the US (Moffitt and Gottschalk (2011) and Shin and Solon (2011). Given the short panels available for this analysis and the limited time-variation within countries in many of the institutional variables, we are wary of drawing strong causal interpretations from these results. Nevertheless, as a descriptive analysis, it is indicative of factors that may be important in determining volatility. 4c. Earnings Dynamics During the Crisis To be completed 4d. Results for the BHPS and GSOEP Using the longer panels available in the BHPS and GSOEP allows us to examine volatility over a longer and more recent time period than considered in Section 4a. Although the BHPS began in 1991, after a period of substantial inequality growth in the 1980s, we nevertheless find that in our BHPS sample of year old males, the variance of log monthly earnings still increased by 33% over the 17 years from Figure 3 plots the SS measure of earnings volatility over this period and shows 9

10 that there is no obvious trend in volatility, despite the increase in total inequality over the same period. 6 For Germany, the variance of log monthly earnings trebled over the sample period, with almost all of the increase occurring after The long-run volatility results, reported in Figure 4 show that volatility was relatively constant up to 1994, then increased from to between 1994 and 2004, with much of this increase reversed by As a result, volatility in 2010 was only 6% higher than in Thus, for Germany, the results for volatility do not mirror the pattern in total inequality. This divergence between total inequality and the SS measure of volatility is seen to some extent in the analysis of the ECHP discussed in Section 4a, but is much clearer in the GSOEP s longer time frame. We return to this issue in Section 5. Figure 5 plots our measure of volatility in Germany against the unemployment rate. The counter-cyclicality of volatility shown in earlier results is even more striking in the GSOEP data. There is some evidence of a rise in volatility after the 2008 financial crisis, but it did not persist in 2010 and it remains to be seen what the longer term consequences of the ongoing crisis will be. 5. Results: Comparing Measures of Instability In this section, we use the long panel available in the GSOEP to compare two measures of earnings instability, the SS and GM measures described in Section 2 6 Jenkins (2011) reports a similar absence of trend in the standard deviation of the change in log earnings for British men in the BHPS, using a balanced panel and men aged See also Blundell and Etheridge (2010). 10

11 above. Previous studies that have examined the instability of earnings in Germany include Myck et al. (2008), Bartels and Bonke (2010) and Fuchs-Schudeln et al. (2010). In contrast to the findings reported in Section 4 above using the SS measure, both Bartels and Bonke (2010) and Myck et al. (2008) find an increase in earnings instability over a similar time period using the MG approach. We now attempt to examine the extent to which the use of different measures of earnings instability accounts for this discrepancy. Previous attempts to compare these measures have been complicated by the fact that the measures have been calculated using different samples. In this paper we overcome this problem by estimating both measures on the same sample. To construct the MG measure, we estimate a version of the model used by Myck et al. (2008), extended to allow permanent inequality to change over the life-cycle. This can be done using either a Random Walk (RW) model (e.g. Dickens (2000)) or a Random Growth (RG) model (e.g. Baker (1997). We report results for the RG model, although the findings for the RW model are very similar. We write earnings for individual i, with x years of experience at time t, y ixt, as y = pα + λ v (2) ixt t ix t it where the first and second terms represent the permanent and transitory components of earnings respectively. It is assumed that E( α ) = E( ν ) = 0. The factor loadings, p t and λ t, allow variances to change over time in a way that is common across individuals. For the RG model, the permanent component is α = α + β (3) ix ix i( x 1) i it 11

12 where E( β i ) = 0. The random growth terms α and β have variances and i0 i 2 σ α 2 σ β respectively and covariance σ αβ. To allow for persistence, the transitory shock, v it, is modelled using an ARMA(1,1) process with AR parameter ρ and MA parameter θ. Specifically, v = ρv + ε + θε (4) it i( t 1) it it 1 where ε it is a random variable with variance σ 2 ε. The recursive nature of the transitory process requires consideration of initial conditions so we follow MaCurdy 2 (1982) and treat the variance of the initial shock, σ, as an additional parameter to be estimated. The model is estimated by GMM. 7 The estimated transitory and permanent components are shown in Figure 6. The results of our model are consistent with those reported by Myck et al. (2008) and show that from 2000 onwards, transitory inequality became an increasingly important element of the growing cross-sectional inequality. This contrasts with the results presented using the SS measure of earnings instability, as can be seen in Figure 7, ν 1 which includes both measures on the same graph for ease of comparison. 8 The divergence in both measures from the mid-1990 s is striking. These results show that, even when using the same sample, these two measures can lead to very different conclusions. As mentioned in Section 2, one possible explanation for differences between the two measures is that the SS measure includes permanent shocks in addition to transitory shocks, whereas the MG measure of instability includes only the latter. However, this cannot explain the divergence between the two measures shown here, 7 The corresponding parameter estimates are given in Table A3 in the Appendix. 8 In this figure, the SS measure is reported as a variance rather than a standard deviation to ensure comparability. 12

13 since both the permanent and transitory components are trending upwards, as seen in Figure 6. The second possible explanation is that an increase in the persistence of transitory shocks increases the transitory variance, but reduces volatility. To see this, note 9 that the SS volatility measure can be written as: ( ) ( ) ( ) 2 ( ) 2 SS Var yit yit 1 Var yit Var yit C ov y it, yit 1 = + (5) 10 Ignoring heterogeneity in the growth profile and assuming for simplicity that p t pt 1 = pand t t 1 λ λ = λ SS = p σα + λ σ v p 2 t + σα + λ σ v p σ t 1 α + λ σ v t,vt 1 (6) SS = λ σ 2 v + σ t v λ σ t 1 v t,vt 1 (7) The first term of Equation 7 corresponds to the transitory component of the MG approach. However, we see that the SS volatility measure differs from the MG measure by a factor that reflects one-period autocovariances in transitory shocks. This term will also affect comparisons in trends if these autocovariances are changing over time. In particular rising (falling) one-period autocovariances will lead to the trend in SS volatility being muted (magnified) relative to the trend in the MG measure. To see the implications of this for the German data, Figure 8 plots the trend in the one-period autocovariance of transitory shocks along with the SS and MG measures of earnings instability. We see that from , during which time these autocovariances remained constant, both measures exhibited similar trends. However, starting in 1994, the one-period autocovariances began to increase dramatically. This in turn led to a rise in the MG transitory variance, but dampened 9 See also Moffitt and Gottschalk (2011) and Shin and Solon (2011) for discussion of this point. 10 Heterogeneity of a plausible magnitude is unlikely to affect this comparison (Baker (1997); Shin and Solon (2011)). 13

14 the trend calculated using the SS approach, since the term being subtracted in Equation 7 was growing. This accounts for the substantial divergence in the trends in the two measures from 1994 on. The results in this section support the advice given by both Moffitt and Gottschalk (2011) and Shin and Solon (2011); when measuring earnings instability, researchers need to be explicit about what they mean by instability. Moffitt and Gottschalk are very clear in associating transitory inequality with mean-reverting shocks; in this case, a correctly-specified parametric model is preferable. However, a criticism of this approach is that shocks that are highly persistent but mean-reverting will be classified as transitory. This criticism is related to our discussion of the trends in earnings instability: if the persistence of transitory shocks is rising, earnings instability as measured by the MG approach will be rising, which seems counterintuitive. The SS measure will, all other things being constant, calculate instability to be falling in this case, as year-to-year differences in each individual s earnings become smaller over time. 6. Conclusions In this paper we examine the nature and determinants of earnings instability across twelve European countries for the period , using the ECHP. Our findings show clear differences in the level of volatility across countries, with Ireland, Spain and the UK having high levels of volatility and Belgium and Denmark having particularly low levels of volatility. For most of the countries analysed, volatility is stable and no country exhibited a significant increase in volatility over this period. 14

15 Our examination of the determinants of volatility show that countries with higher union coverage exhibit lower levels of volatility, while there is a positive relationship between volatility and unemployment. The counter-cyclical pattern in volatility is also evident when we conduct a longer run analysis for the UK and Germany using the BHPS and GSOEP respectively. In the final section of the paper we use the GSOEP to compare two measures of earnings instability often used in the literature. The results show that, even when using the same sample, these two measures can lead to very different conclusions. Our analysis illustrates how the relationship between these alternative measures is affected by changing patterns of short-run persistence in earnings. This should be borne in mind when choosing how to document changes in instability over time. 15

16 References Baker, M., (1997), Growth Rate Heterogeneity and Covariance Structure of Life- Cycle Earnings, Journal of Labor Economics, 15(2), pp Baker, M. and G. Solon (2003). Earnings Dynamics and Inequality among Canadian Men, : Evidence from Longitudinal Income Tax Records, Journal of Labor Economics, Vol. 21, No. 2, pp Bartel, A. and Lichtenberg, F., (1987). The Comparative Advantage of Educated Workers in Implementing New Technology: Some Empirical Evidence. Review of Economic and Statistics 69, 1-11 Bartels, C. and T. Bonke (2010), German Male Income Volatility : Trends in Permanent and Transitory components and the role of the welfare state, Freie Universitat Berlin, Discussion paper 2010/18. Bertola, G., F. Blau and L. Kahn (2002). Comparative Analysis of Labor Market Outcomes: Lessons for the US from International Long-Run Evidence pp in A. Krueger and R. Solow (eds.), The Roaring Nineties: Can Full Employment Be Sustained?, Russell Sage and Century Foundations, Blundell, R. and B. Etheridge (2010), Consumption, income and Earnings inequality in Britain, Review of Economics and Dynamics, 13,. pp Cameron, S. and J. Tracy (1998), Earnings Variability in the United States: An Examination using Matched-CPS Data, Mimeo Cappellari, L., (2004), The Dynamics and Inequality of Italian Men s Earnings: Long-Term Changes or Transitory Fluctuations? Journal of Human Resources, Vol. 39(2), pp Cervini, M. and X. Ramos (2008). Long-Term Earnings Inequality, Earnings Instability and Temporary Employment in Spain , IZA WP no Checchi, D. and C. García-Peñalosa (2008). Labour Market Institutions and Income Inequality, Economic Policy, CEPR, CES, MSH, Vol. 23, pp , October. Daly, M. and R. Valletta, 2008, Cross-National Trends in Earnings Inequality and Instability, Economic Letters, Vol. 99(2), pp Dickens, R. (2000). The Evolution of Individual Male Earnings in Great Britain: , Economic Journal, Vol. 110, No. 460, pp Doris, A. D.O Neill and O.Sweetman (2011) Identification of the Covariance Structure of Earnings using the GMM Estimator, forthcoming Journal of Economic Inequality Dynan, K, D. Elmendorf and D. Sichel (2008), The Evolution of Household Income Volatility, Mimeo Brookings Institution 16

17 Dynarksi, S. and J. Gruber (1997) Can Families Smooth Variable Earnings? Brookings Papers on Economic Activity 1, pp Fuchs-Schundeln, D. Krueger and M. Sommer (2010), Inequality Trends for Germany in the Last Two Decades: A tale of Two countries, Review of Economics and Dynamics, 13, pp Gustavsson, M. (2007). The 1990s Rise in Swedish Earnings Inequality Persistent or Transitory?, Applied Economics, Vol. 39, pp Haider, S.,(2001), Earnings Instability and Earnings Inequality of Males in the United States: , Journal of Labor Economics, Vol. 19(4), pp Jenkins, S (2011) Changing Fortunes: Income Mobility and Poverty Dynamics in Britain Oxford University Press, Oxford. Kalwij, A. and R. Alessie (2007), Permanent and Transitory Wages of British Men, : Year, Age and Cohort Effects, Journal of Applied Econometrics, Vol. 22, pp MaCurdy, T. (1982). The Use of Time-Series Processes to Model the Error Structure of Earnings in a Longitudinal Data Analysis, Journal of Econometrics, Vol. 18(1) pp Moffitt, R. and P. Gottschalk (1995), Trends in the Autocovariance Structure of Earnings in the U.S., , mimeo, Johns Hopkins University. Moffitt, R. and P. Gottschalk, (2002), Trends in the Transitory Variance of Earnings in the United States, Economic Journal, Vol. 112, pp Moffitt, R. and P. Gottschalk, (2011), Trends in the Transitory Variance of Male Earnings in the U.S., , forthcoming Journal of Human Resources Myck, M., R. Ochmann and S. Qari (2008), Dynamics of Earnings and Hourly Wages in Germany, IZA DP No Nickell, W (2006). The CEP-OECD Institutions Data Set, CEP Discussion Paper no OECD (2005) Education at a Glance: OECD Indicators 2005 Ramos, X., (2003), The Covariance Structure of Earnings in Great Britain, , Economica, Vol. 70, pp Shin, D. and G. Solon, (2011), Trends in Men s Earnings Volatility: What does the Panel Study of Income Dynamics show? Journal of Public Economics, 95, pp Sologon, D. and C. O Donoghue, (2009), Policy, Institutional Factors and Earnings Mobility, IZA Discussion Paper No

18 Table 1: Determinants of Aggregate Inequality and Volatility (Dependent Variable SS measure of instability in country i at time t). Explanatory Variable Coeff (Std.Error) Union Coverage * ( ) Bargaining Centralization ( ) Replacement Ratio ( ) Education Level ( ) Unemployment Rate * ( ) Random Effects YES No. Obs

19 Figure 1: Total Inequality in 12 European Countries: Variance of Log Earnings Austria Belgium Denmark Finland Var Lnwage Var Lnwage Var Lnwage Var Lnwage Year Year Year Year France Germany Ireland Italy Var Lnwage Var Lnwage Var Lnwage Var Lnwage Year Year Year Year Netherlands Portugal Spain UK Var Lnwage Var Lnwage Var Lnwage Var Lnwage Year Year Year Year Figure 2: SS Measure of Volatility in Earnings in 12 European Countries Austria Belgium Denmark Finland stdiffres stdiffres stdiffres stdiffres year1 year1 year year1 France Germany Ireland Italy stdiffres stdiffres stdiffres stdiffres year1 year1 year1 year1 Netherlands Portugal Spain UK stdiffres year1 stdiffres year1 stdiffres year1 stdiffres year1 19

20 Figure 3: SS Measure of Instability in Earnings, UK SS measure of instability year Figure 4: SS Measure of Instability in Earnings in Germany, SS measure of instability year 20

21 Figure 5: SS Measure of Instability and the Unemployment Rate in Germany, SS measure of instability year unemployment rate SS measure of instability unemployment rate Figure 6: Evolution of Inequality in Germany and its Permanent and Transitory Components, Year perm1 predicted_total1 temp1 actual_total1 21

22 Figure 7. Comparison of SS and MG Measures of Instability for Germany, year SS meaure of instability MG meaure of instability Figure 8. SS and MG Measures of Instability and the First-Period Autocovariance of Transitory Earnings for Germany, year MG meaure of volaitlity transcov SS meaure of volaitlity 22

23 Appendix Table A1: Sample Sizes for ECHP Analysis Austria Belgium Denmark Finland France Germany Ireland Italy Netherlands Portugal Spain UK Table A2: Summary Statistics of Institutional Variables Variable Mean Std. Deviation Minimum Maximum Union Coverage (UK 2000) 98.8 (Aus. 1995) Centralized Bargaining (UK) 5 (Finland) Replacement Ratio (Italy 1994) 72 (Denmark 1994) Education Level (Italy 1994) 32 (Finland 2000) Unemployment Rate (Neth. 2000) 23.9 (Spain 1994) 23

24 Table A3: GMM Parameter Estimates from Random Growth Model of Earnings Dynamics for Germany, Coef. Std. Err. z P> z [95% Conf. Interval] σ α ρ σ v σ є λ λ λ λ λ λ λ λ λ λ λ λ λ λ λ λ λ λ λ λ λ λ λ λ λ λ p p p p p p p p p p p p p p p p p p p p p p p p p p σ β σα β θ

25 Figure A1: Union Coverage Across Countries Over Time Austria Belgium Denmark Finland France Germany Ireland Italy Netherlands Portugal Spain United Kingdom Graphs by country Union Coverage Figure A2: Centralised Bargaining Across Countries Over Time Austria Belgium Denmark Finland France Germany Ireland Italy Netherlands Portugal Spain United Kingdom Graphs by country Centralised Bargaining Index 25

26 Figure A3: Gross Replacement Rates Across Countries Over Time Austria Belgium Denmark Finland France Germany Ireland Italy Netherlands Portugal Spain United Kingdom Graphs by country Gross Replacement Ratios Figure A4: Third Level Educational Attainment Across Countries Over Time Austria Belgium Denmark Finland France Germany Ireland Italy Netherlands Portugal Spain United Kingdom Graphs by country Education 26

27 Figure A5: Unemployment Rates Across Countries Over Time Austria Belgium Denmark Finland France Germany Ireland Italy Netherlands Portugal Spain United Kingdom Graphs by country Unemployment Rates 27