Inter-industry wage differentials in EU countries: what do crosscountry time varying data add to the picture?

Transcription

1 Inter-industry wage differentials in EU countries: what do crosscountry time varying data add to the picture? Philip Du Caju (NBB), Gabor Katay (MNB), Ana Lamo (ECB), Daphne Nicolitsas (Bank of Greece) and Steven Poelhekke (DNB) June 2008 Preliminary and incomplete Abstract This paper empirically investigates the changes in inter-industry wage differentials between 1995 and 2002 across a number of EU countries: Belgium, Germany, Greece, Hungary, Ireland, Italy, Netherlands, and Spain. Our focus is to investigate the extent to which these inter-industry wage differentials and their evolution have been driven by macroeconomic developments such as competitiveness and exposure to international trade or to labour market institutions. We also investigate the importance of sectors ability to pay and unobserved workers ability as potential determinants of inter-industry wage differentials. Wage differentials are estimated using the so called Structure of Earnings Survey data (SES); a unique dataset of matched employer-employee data, collected from a large sample of firms in each country and comparable across countries. The paper provides evidence of the existence and persistence of industry wage differentials in European countries, additionally it does not find any indication of unobserved ability being one of the determinants. The initial regression results attempting to explain the change in the differentials between the two points in time suggest that there is some correlation with labour market developments, but also with the ability of industries to pay higher wages. Keywords: inter-industry wage differentials JEL Classification: J31, J41 * This paper is part of the Wage Dynamic Network (WDN) research. We are very grateful to our WDN colleagues for their comments and support. The support and help of Frank Smets and of our ECB colleagues in DG-Statistics were essential to gain access to five of the SES data sets used in the paper. We are also grateful to Rebekka Christopoulou who provided excellent research assistance and to Andrew McCallum and Ladislav Wintr for initial help with the data. We would also like to thank Eurostat and the National Statistical Institutes of each one of the countries studied in this paper for granting us access to the SES data and assisting with clarifications. Opinions expressed in this article do not necessarily reflect the views of the European Central Bank, NBB, Bank of Greece, MNB or DNB. Responsibility for errors and omissions remains with the authors. 1

2 1. Introduction There is a wide literature on wage structures examining both the returns to workers characteristics (education, tenure, etc.) based on the work by Mincer (1974), and relative wages or wage differentials among industries and occupations in many countries. The latter literature was triggered by Krueger and Summers (1988) who showed that wage differentials among workers with identical features and the same working conditions, but working in different sectors were quite persistent in the US during the 80s, concluding that this was evidence against market-clearing theories of wage determination. These differentials seem to be an accepted fact in the literature. Recent work on wage differentials for European countries includes several papers produced within the Pay Inequality and Economic Performance project which uses 1995 Structure of Earnings Survey (SES) data. However there is no systematic accounting of cross-country differences in changes in relative wages over the past decade. The objective of this empirical project is to understand how relative wages have changed from the mid 1990 s to the start of the current decade in a large number of EU countries (Belgium, Germany, Greece, Hungary, Ireland, Italy, Netherlands and Spain), for which data are available, and to what extent their evolution has been driven by macroeconomic developments such as competitiveness and exposure to international trade or to labour market institutions. We also investigate the importance of sectors ability to pay and unobserved ability of workers as potential determinants of inter-industry wage differentials as well as the impact of incipient labour market reforms which, as argued in many recent studies, might have an effect on the differentials. Two distinctive features of this study are: (i) the period covered ( ) is characterised by significant technological progress and economic globalization of European markets. These trends are also associated with changes in the environment in which European labour markets operate and could have had an impact on the national wage structures, (ii) compares results across 8 EU countries, namely: Belgium, Germany, Greece, Hungary, Ireland, Italy, Netherlands and Spain, which represent a large proportion of the EU and whose labour market institutions vary significantly. An investigation of the determinants of the changes in the differentials in this cross-country comparison can help to identify the role of competitive and other forces in shaping wages. Wage differentials are estimated using the so called Structure of Earnings Survey (SES), which is a dataset of matched employer-employee data, collected from a large sample of firms in each country. The SES contains rich information on the structure and distribution of 2

3 earnings and individual characteristics of employers and employees on a comparable basis across EU countries, and thus provides a unique opportunity to estimate inter-industry wage differentials where both the effects of workers and job characteristics have been conditioned out and are fully comparable across countries. In addition the fact that these data are available for two points in time, in general 1995 and 2002, allows to empirically test the relationship between relative wage adjustments and demographic trends, technological changes and institutional features across EU countries over the last decade, as described above. The rest of the paper is structured as follows: the next section provides a selective overview of related literature. Section 3 briefly describes the data sets used, Section 4 discusses the observed wage differentials in the 8 countries, and Section 5 presents and discusses the conditional ones, after controlling for individual, job and firms characteristics. Section 6 investigates the importance of unobserved workers ability as a potential determinant of interindustry wage differentials. Section 7 investigates to what extent these inter-industry wage differentials and their evolution have been driven by macroeconomic developments such as competitiveness and exposure to international trade or by labour market institutions. Section 8 concludes. 2. Literature review The existence of industry wage differentials has been extensively documented in the economics literature. One of the earliest and best known pieces of documentation is the evidence provided by Slichter (1950). In more recent times, the interest in the topic was stimulated in a series of papers by Dickens, Gibbons, Katz, Krueger, Murphy, Topel and Summers in the late 1980s. The papers focused on the US, and each investigated a different facet of the issue: the role of unobserved quality in explaining inter-industry wage differentials (Krueger and Summers, 1987 and 1988; Murphy and Topel, 1987; Gibbons and Katz, 1992), the influence of institutional factors (Krueger and Summers, 1987), the persistence of these differentials over time (Krueger and Summers, 1987), the similarity of the structure of wages across countries (Krueger and Summers, 1987), and across occupations (Dickens and Katz, 1987). All of the above studies conclude that industry wage differentials persist over time, but the explanations for this phenomenon differ. The aforementioned papers, with the exception of that of Murphy and Topel, appear to hold the view that interindustry wage differentials cannot be explained by competitive labour market theories since individuals of the same ability (observed and unobserved) appear to gain more in certain industries, and it is the same industries that pay higher wages both over time and across countries. Non-competitive explanations along the lines of a combination of efficiency wage theories and rent sharing seem to fit better with these facts. Murphy and Topel (1987) on the 3

4 other hand challenge this view by applying a different methodology which involves taking into account the fact that the sorting of abilities across industries is correlated with job attributes. They conclude that about two-thirds of industry wage differentials are due to unmeasured worker characteristics while the remaining third is ascribed to compensation for the instability of jobs within certain industries. The results of the ensuing literature would suggest that unobserved labour quality might be more important than found in the previous literature, and the similarity of the differentials across countries is not as great as claimed until then. One could say that since the late 1980s the subsequent literature on the issue has followed mainly three directions in trying to resolve the controversy between competitive and non-competitive explanations. The first, includes international comparisons of inter-industry wage differentials (see, inter alia, the list of studies presented in Table 1), the second route consists of the application of different methodologies in measuring the magnitude of the differentials and differences in the assumptions made about the endogeneity or otherwise of occupational and industry choice, while the third direction pertains to the exploration of longer panels of individuals (see, inter alia, Carruth and Dickerson, 2004). As already mentioned above, the verdict of the earlier literature regarding international comparisons was that despite certain differences in the magnitude of the inter-industry variation of wages, the rankings of industries remained relatively consistent across countries a fact that was difficult to reconcile with an institutional factors interpretation. Edin and Zetterberg (1992) is one of the first papers to question this conclusion; using micro data for Sweden, the authors illustrate that the raw and conditional dispersion of wages in Sweden is narrower than in the US. They ascribe the difference in their findings with those reached in the earlier literature to the use of micro-level data, while the earlier conclusions on cross-country comparisons were based on aggregate data. Another paper also making international comparisons on the basis of individual-level data is that of Zanchi (1995) which uses data for 5 countries (US, Canada, Australia, Germany and the Netherlands) and again seems to find that there is not much similarity in the wage structure across countries. Both the Edin and Zetterberg and the Zanchi papers attribute the differences in the wage structure between countries to divergence in institutions. As for the methodological differences, the 1990s literature on the topic makes use of more disaggregated industry information, uses individual level longitudinal data for long periods of time, explores the hypothesis that firm (rather than industry) wage policies are prevalent, and further examines the wage distribution within industries in order to test for evidence of differences in qualities which are difficult to measure between workers. Abowd, Kramarz and Margolis (1999) for example are able to separate worker and firm fixed effects and conclude 4

5 that a large portion of wage variation in France is due to unobserved person fixed effects. 1 While another study for France, that by Goux and Maurin (1999), who estimate inter-industry wage differentials from a panel data set of individuals in France over the period finds that (a) the extent of evidence in favour of the unobserved quality hypothesis depends on the level of industry wage disaggregation used, (b) the most important factor in determining individuals wages is not in which industry, but in which firm, the individual works in. Martins (2004), on the other hand, investigates whether inter-industry wage differences hold for different quantiles of the distribution. The reasoning behind this line of investigation is that if unobserved ability is significant in explaining the industry wage structure, industry wage differentials would be wider at the top quantile of the wage distribution. Using micro-level data for Portugal Martins is unable, however, to find evidence in favour of the unobserved quality hypothesis. More recently, Gibbons, Katz, Lemieux and Parent (2005) develop a model in which wage changes and sector mobility are endogenous. The model is estimated using US longitudinal data for a large number of individuals over 17 years. The results suggest that the unobserved quality hypothesis could be driving the higher wages paid in certain industries, such as finance and professional and business services, but do not accord with the industry wage differences in mining, manufacturing and construction. Table 1: Some indicative studies on cross-country inter-industry wage differentials Authors Data Countries and period Main conclusions covered Edin & Zetterberg (1992) Micro-level data Sweden and US for 1984 Magnitude of conditional interindustry wage differentials significantly smaller in Sweden than in the US. Correlations of wage structures across countries estimated on the basis of wage differentials arising from aggregate data overestimate the similarities. Gittleman & Wolff Industry-level data 14 OECD countries, Ranking of industries within each country shows little (1993) variation over time. Size of differentials depends positively on productivity growth, output growth and capital intensity and negatively on the degree of import penetration. Zanchi (1995) Micro level data from the US (1986), Canada Little similarity of conditional wage differentials Luxembourg Income Study (1987), Australia (1986), across countries. Importance of demographic, human Databank Germany (1985), capital and socio-economic characteristics in Netherlands (1987) explaining inter-industry differentials varies across countries. Importance of institutional factors (e.g. degree of centralization of negotiations) in explaining cross-country differences in wage structure. Erdil & Yetkiner (2001) Industry-level data 20 countries, Wage structure similar across developed and 1 Postel-Vinay and Robin (2002) allow for endogenous job mobility (on-the-job search) and search frictions. This creates heterogeneous bargaining power among firms and therefore different degrees of rent extraction. In that case the unobserved person effect is much smaller. It decreases with the observed skill level of employees. 5

6 Hartog, Opstal, & Teulings (1987) Rycx, Tojerow & Valsamis (2008) developing world but explanations for this might differ across these two groups of countries. Micro-level data Netherlands and the US Dutch and US wage differentials correlate strongly, but the standard deviation in The Netherlands is up to a half smaller. Tenure profiles are much less steep in The Netherlands and firm size matters much less. The difference may be partly due to more centralized bargaining) in The Netherlands. Micro-level data 6 West-European and 4 The ranking of sectors in terms of wages, even after East-European countries, controlling for characteristics, is quite similar in Eastern and Western European countries. A negative 2002 correlation between the dispersion of inter-industry wage differentials and the degree of corporatism across countries is found. 3. Data The present study is based on micro data from the Structure of Earnings Surveys (SES) carried out by the national statistical offices of Belgium, Germany, Greece, Hungary, Ireland, Italy, the Netherlands and Spain. The SES is a standardized survey conducted in 20 european countries. It was conducted for the first time in In 2002, the survey was repeated and it was then decided (Council Regulation (EC) No 530/1999) that the survey will, starting from 2002, be conducted every 4 years although at the moment only two waves are available. 2 The surveys are carried out on a sample of plants selected by stratified random sampling (stratification is done by economic activity, size and for certain countries region), while within plants a random sample of employees was chosen. 3 The SES provides individual earnings data for employees together with information on their employer. The three main advantages of the data are: (a) it makes available information on earnings in a standardised fashion across countries, (b) this information is repeated over time and (c) since the data are collected through the employer, the measurement error usually associated with household data is much smaller. The samples include, in most countries establishments with at least 10 employees, those active in industry (including construction) and services. 4 The industries, at the 2-digit NACE level, covered for each country are presented in Table A6, while a list with the description of each two-digit industry is given in Table A7. The sample of employees includes both fulltime and part-time employees, but interim and occasional workers with the exception of apprentices are not sampled. The survey provides detailed information on monthly and annual earnings. The number of hours worked both normal and overtime, is also recorded. Employee characteristics include age, education, gender, citizenship, occupation, type of contract (fixed term or indefinite length), management/supervisory position and length of tenure with the 2 The first wave for Hungary refers to 1996 instead of 1995 and the second wave for Germany refers to 2001 instead of The first wave used for Belgium in this paper is that for In Italy and the Netherlands the employer information refers to the firm rather than the plant. The same is true for Belgium if the firm has several plants within the same municipality. For Hungary the sectoral classification of the unit of observation refers to the activity of the firm rather than the plant. 4 This includes in general sections C to K of the economic activity classification scheme NACE 1.1. Agriculture and non-market services were excluded. Some countries included additional sectors and small establishments (establishments with less than ten employees). See Table A2 for details. 6

7 firm. Firm characteristics include region, industry, firm total employment, type of economic and financial control of the firm (private or public), the principal market for the firm s products, and the level at which wage bargaining takes place. A list of the variables from the SES used in the analysis with short descriptions of these is presented in Table A5. The analysis carried out in this paper has been based on four alternative measures of earnings constructed with the available data: (i) total annual earnings before tax including all regular and irregular pay components (payy), average hourly earnings including regular bonuses and absences paid at full rate but excluding irregular bonuses (payhb) (ii) average hourly earnings including overtime, regular bonuses and absences paid at full rate but excluding irregular bonuses (payh), and (iii) average hourly earnings including overtime, regular and irregular bonuses as well as absences paid at full rate (payhi). The results reported in the paper, however, only refer to the payh variable as it is the one typically used in similar studies and it was possible to construct for every year and country 5. For the years prior to the introduction of the euro for countries belonging to the Euro Area in 2002, monetary variables have been expressed in euros using the irrevocable exchange rates at which countries converted their national currencies to the euro. The samples analysed for almost all countries contain both men and women aged between 16 and 65 years old. 6 In certain countries the sample includes only individuals in the private sector (Belgium, Greece, Hungary), while for the rest of the countries the difference between the two sectors is taken into account by using a dummy in the regressions. The occupational classification used is the single-digit International Standard Classification of Occupations (ISCO) which organises occupations in ten main groups. The regressions contain occupational dummies for 8 groups. 7 Individuals for which earnings information was either not available or which were thought to be outliers on the basis of their earnings information in the sample have been excluded. More specifically, workers with earnings falling below the first and above the 99th percentile within each sector have been excluded. 8 For each country the analysis is restricted to individuals belonging in sectors sampled in both waves. 5 Except for Hungary, for which country we cannot calculate payh for Therefore, we use payhb instead for both waves. We think that this is a good proxy of payh as according to the data used, payh and PayHb in 2002 are pretty similar. 6 In Greece workers younger than 25 and older than 64 were excluded, to increase the homogeneity of the sample in terms of marital status which is a determinant of pay (married individuals receive a benefit equal to 10% of the basic wage). 7 Employees classified as belonging either to the Armed Forces (ISCO group 0) or as Skilled agriculture and fishery workers (ISCO group 6) have been excluded from the sample. 8 In Greece the excluded workers are those with monthly earnings less than 80% of the minimum wage or over 20 times the minimum wage. 7

8 4. Observed inter-industry differentials In this section we look at observed wage differentials across industries at the NACE2 level. 9 By this we mean raw differentials not controlling for worker, job or firm characteristics, calculated as the deviations of (log) mean sectoral wages from a measure of aggregate wages. Table A1 reports observed wage differentials for eight EU countries in the first year of our sample (1995 for all the countries except for Belgium and Hungary where the information refers to 1999 and 1996 respectively). Table A2 shows the same information for 2002, except for Germany for which the information presented refers to Substantial wage differences between workers employed in different sectors are observed in all countries. Sectors paying higher wages to their employees include extraction and mining industries, in those countries in which the sample includes these sectors, the petroleum, nuclear and chemical industries, and the utilities (electricity, gas and hot water supply). In the services sector, higher wages are paid in financial and insurance sectors, as well as in computer activities and research. The size of the differentials for each industry differs across countries. For example in 1995, the observed wage differential for the chemical products industry goes from a mere 5.8% in Italy to 24.2% in Hungary. In the services sector the differential for financial intermediation ranges from 45.9% in Ireland to only 3.9% in Germany. At the lower end of the wage distribution, one finds industries that are classified as old in Europe, namely clothing leather and textiles industries. Traditionally, retail trade and hotels and restaurants are the services sectors where the lowest wages are paid. Also in this part of the distribution, there are differences between countries as to the size of the differentials. In 2002, the negative observed differential for the clothing industry ranges from -13.6% in The Netherlands to -44.2% in Hungary, whereas in retail trade it is between -17.5% in Italy and -44.6% in The Netherlands. Table 2. Spearman rank correlation between observed wage differentials in countries in 2002 BE ES DE 2 GR HU IE IT NL BE 1 ES * 1 DE * * 1 GR * * * 1 HU * * * * 1 IE * * * * * 1 IT * * * * * * 1 NL * * * * * * * 1 2 Germany: 2001 in stead of * Significant at the 5 p.c. level. 9 This level of classification compares to the 43 (2-digit SIC) industry groups used in Krueger and Summers (1988). 8

9 Despite the differences between countries in the size of the differentials, the ranking of sectors in terms of the differentials is rather similar across the countries we consider. Table 2 reports Spearman rank correlation coefficients between observed industry wage differentials for all countries in The correlations range from between Ireland and Hungary to between Belgium and The Netherlands, and are all significant at the 5 p.c. level. Table 3. Spearman rank correlation of observed wage differentials between 1995 and 2002 BE 1 DE 2 ES GR HU 3 IE IT NL Belgium: 1999 instead of Germany: 2001 instead of Hungary: 1996 instead of All correlations are significant at the 1 p.c. level. Within countries, the ranking of sectors has remained broadly unaffected between 1995 and Table 3 shows highly significant (consistently at the 1 p.c. level) Spearman rank correlation coefficients between the observed wage differentials in 1995 and 2002 in each country, ranging from in Greece to in Spain. However, the dispersion of the observed wage differentials and or other features of their distribution substantially differ across countries. Table 4 shows standard deviations of the observed inter-industry wage differentials, as an overall measure of wage dispersion between sectors. This dispersion is highest in Greece, Ireland, Spain and Hungary and lower in Belgium, Germany and Italy. Table 4. Standard deviations of observed wage differentials in 1995 and Change BE DE ES GR HU IE IT NL Belgium: 1999 instead of Germany: 2001 instead of Hungary: 1996 instead of Conditional inter-industry differentials Estimates of the conditional inter-industry wage differentials for Greece and Spain have been borrowed from the following papers: Izquierdo and Lamo (2008) and Kosma and Nicolistas (2008) which are also WDN research papers and follow the same methodology and use same data and codes as this paper. SES estimations for Italy, Ireland and Spain were done at the Safe Center in Eurostat and those for Germany via remote access at DEstat (Germany). The data for The Netherlands was accessible from Statistics Netherlands through remote access at De Nederlandsche Bank. 9

10 The observed differentials of the average wage across industries summarised in the previous section may obviously be due to differences of worker and or job feature across industries, if an industry employs more skilled and productive workers than others it is also expected to offer higher wages. In this section we try to control for observable productive features of the employees and characteristics of the workplace they are employed in. To this effect we will follow the literature and rely on the estimation of extended Mincer equations for each year for each country. The estimated specification is of the following form = i + β j X ji + γ kyki + j k h ln w α δ Z + η, (1) where w i represents the wage of individual i, X is a vector of workers observable individual and job related features (education, gender, tenure, type of contract, etc.), Y is the vector of employers characteristics (firm size, location, etc.). Finally, Z contains industry dummies. Our parameters of interest are then the δ h where h=1,,h, where H+1 is the number of NACE 2-digit industries in each country sample, δ h measures the wage differential, ceteris paribus, in industry h relative to the omitted industry. Several measures of conditional wage differentials can be defined as functions of δ h. We follow Zanchi (1998) and calculate interindustry wage differentials for all H+1 industries with respect to a weighted (by employment) average as: Where; H h= 1 d d kh = δ π H + 1 h = π ( for h = 1,..., H ) (2) π = p δ ( for h = 1,..., H ) With δ h being the estimated sector coefficient from h h N 1 equation (1) and ph = ph, i ( for h= 1,..., H) is the sectoral employment share in N i = 1 the observed sample. h ih i The standard errors of the industry wage differentials d in equation 1 can be calculated by adjusting those of the original OLS estimate δ h.for that we transform the original variance and covariance matrix following Zanchi (1998): var - cov( δ*) = ( K es ')( var - cov( δ ))( K es ')' where K is a ((H+1) x H) matrix constructed as the stack of an (HxH) identity matrix and a (1xK) row of zeros, e is a ((H+1)x1) vector of ones, s is the vector of employment shares of 10

11 the H first industries, and var - cov( δ ) is the original variance-covariance matrix of the industry dummy coefficients. The standard errors of d are simply the square roots of the diagonal elements of this transformed variance-covariance matrix. Table A3 reports conditional wage premia for eight EU countries in the first year of our sample calculated, following equation (2), from the coefficients on sector dummies estimated through equation (1). Table A4 shows these conditional differentials for the year 2002, except for Germany where the data refer to It is first noted that interindustry wage differences remain significant even after controlling for an important set of worker, job and firm characteristics. A large number of these differentials is significant at the 1 p.c. level and all of them at the 5p.c level. Nevertheless, as expected, conditional wage differentials tend to be smaller in size than the observed ones. In fact, the differential explained through these characteristics can be substantial; as an example, note that in 2002 the highly positive observed wage differential for a worker in the Coke, petroleum production and nuclear fuel industry in Greece of 44.6%, and for a worker in the Electricity, gas and hot water supply in Ireland of 45.2% are reduced to conditional wage premia of 14.4% and 19.1% respectively (see tables A2 and A4). In the low-paying industries in 2002, wage penalties for workers in the Hungarian clothing industry of -44.2% and in the Dutch retail trade of -44.6% are reduced to differentials of -17.7% and -13.1% after conditioning. In all of the countries the hierarchy of sectors is very similar to the one obtained from the observed wage differentials. The Spearman correlation coefficients between observed and conditional wage differentials in 1995 and 2002, reported in Table 5, lies between in Hungary in 1995 and in Ireland in They are all significant at the 1 p.c. level. Table 5. Spearman rank correlation between observed and conditional wage differentials BE DE ES GR HU IE IT NL Belgium: 1999 in stead of , Hungary: 1996 instead of Germany: 2001 in stead of All correlations are significant at the 1 p.c. level. Also, the ranking of sectors in terms of conditional wage premia is very similar across countries. Spearman correlation coefficients of the rankings between countries, presented in Table 6, are mostly significant at the 5 p.c. level. High-wage jobs are still to be found in the extraction and oil and chemical industries, as well as in financial intermediation services. Conditional differentials are mostly negative in traditional textiles, clothing and leather industries, and in retail trade and hotels and restaurants. 11

12 Table 6. Spearman rank correlation between conditional wage differentials in countries in 2002 BE DE 1 ES GR HU IE IT NL BE 1 DE 0.712* 1 ES 0.924* 0.748* 1 GR 0.616* 0.414* 0.723* 1 HU 0.740* 0.616* 0.761* 0.456* 1 IE 0.540* * IT 0.901* 0.511* 0.839* 0.660* 0.655* 0.471* 1 NL 0.806* 0.526* 0.741* 0.474* 0.711* 0.424* 0.753* 1 1 Germany: 2001 in stead of * Significant at the 5 p.c. level. Like the observed wage differentials, the ranking of conditional wage premia in sectors has remained rather stable between 1995 and Table 7 shows highly significant (at the 1 p.c. level) rank correlation coefficients between the conditional wage differentials in 1995 and 2002 in each country, ranging from in Hungary to in Germany. Table 7. Spearman rank correlation between conditional wage premia in 1995 and 2002 BE 1 DE 2 ES GR HU IE IT NL Belgium: 1999 in stead of Germany: 2001 in stead of All correlations are significant at the 1 p.c. level. Again, despite these similarities differences across countries in terms of dispersion of these conditional wage differentials do exist. The standard deviations of conditional wage premia in the selected countries in 1995 and 2002, reported in table 8, are, as expected, smaller than those of the observed wage differentials. They are relatively high in Hungary, Spain and Ireland, and relatively low in Belgium and Germany. Between 1995 and 2002, the dispersion of conditional differentials has decreased in Belgium, Greece, Hungary and Ireland; it has risen in Italy, Spain and The Netherlands, while has remained more or less stable in Germany. 11 Table 8. Standard deviations of conditional wage premia in 1995 and Change BE DE ES GR HU IE IT NL Belgium: 1999 in stead of Comparing two points in time does not allow one to draw conclusions about trends in the movement of wage differentials. Du Caju et al. (2008) show that inter-industry wage differentials have decreased in Belgium between 1999 and 2002 and have risen after that and until 2005, more or less in phase with the economic cycle. 12

13 2 Germany: 2001 in stead of Figure 1 shows box plots of the conditional inter-industry wage differentials in each of the eight countries for both of the SES waves, thus providing an overview of the within country distribution of these wage differentials. Again, it appears that the spread is highest for Spain, Hungary and Ireland, and that it is lowest for Belgium and Germany. Figure 1. Distribution of conditional wage differentials in 1995 and 2002 inter-industry conditional wage diffrentials 1995/wave Inter-industry conditioanl wage differentials 2002/wave BE DE ES GR HU IE IT NL BE DE ES GR HU IE IT NL Inter-industry conditional wage dispersion has decreased between 1995 and 2002 in Belgium, Ireland and Italy. Overall conditional wage dispersion across industries has increased between 1995 and 2002 in Hungary, Greece, Spain and The Netherlands, whereas it has remained at a similar and low level in Germany. An analysis of variance (ANOVA) of the conditional wage differentials across sectors, countries and time gives insight on how this variance is distributed between different factors, of course without implying any causality. We define 4 effects in the ANOVA: (1) a "Sector" effect, being simple industry dummies that show common (or average) inter-industry wage differentials; (2) a "Year*Sector" effect, showing common changes in the industry wage structure; (3) a Country" effect, defined as country dummies multiplied by the sign of the differentials; therefore, the coefficient should be interpreted in terms of dispersion. A positive coefficient means that wages are more dispersed, i.e. the magnitude of the corresponding differential is larger (negative or positive) and a negative coefficient means that some country specific factors tend to compress inter-industry wage differentials. 12 (4) a "Country*Sector" effect, that is the interaction between simple country and sector dummies and captures country specific industry wage structures. The residual includes country specific changes in the industry wage structure Conditional wage differentials are expressed as the deviation from the employment-weighted mean wage, thus simple country dummies will capture these country specific employment-weighted mean wage. 13

14 Table 9 shows the results of the ANOVA. Sequential sums of squares (SS) are presented instead of partial SS, which means that each effect is adjusted for all other effects that appear earlier in the model, but not for any effects that appear later in the model. 13 Table 9 shows that more than 50% of the estimated wage differentials are sector specific. Common changes in the industry wage structure are statistically significant but less important. Of course, as already mentioned above, we only have two years which is not enough to truly capture common trends. Table 1: ANOVA of conditional wage premia df Seq. SS % F Model Sector Period*Sector Country Country*Sector Residual Country specific factors influencing the extent of wage dispersion appear to explain a considerable part of the variation (about 15%). This is not surprising as differences in wage bargaining institutions between countries affect wage dispersion. As an example, Rycx et al. (2008) show that the dispersion of inter-industry wage differentials decreases with collective bargaining indicators, like the degree of centralisation, the degree of bargaining coordination, the collective bargaining coverage rate, and union density. The remainder of the variation is either explained by differences in the industry wage structure across countries (22%) or its change over time (5%). 6 Testing the unobserved quality hypothesis Having seen that for the countries in our sample wage differentials across industries are not fully explained by workers, job and firms characteristics, i.e. conditional wage differentials are still significant and show similar patterns to the observed ones, we now try to gather some evidence on whether unobserved quality of workers could be a factor behind these differentials. For that we following Martins (2004) and look at wage differentials across the 13 This is needed because of the correlation of the sector effect and country effect: first, we want to control for the sector effect, and then, add the country effect conditional on common inter-industry wage differentials. 14

15 wage distribution. Martins (2004) argues that if conditional wage differentials reflect compensation for unobservable labour quality one would expect that the wage premia or wage differentials would be higher at the top end of the distribution. In other words, if this was true then in industries with high average wage premia, these would be even higher at the top end of the distribution. We first test, for every industry, whether workers at the 90 th percentile of the wage distribution receive on average higher wage premia (higher wage differential) that those at the 10 th percentile. Table A8 shows the difference between the conditional differentials for each country and year, arising from the estimation of extended Mincer regressions at the 10 th and the 90 th percentiles. It also shows the p-value from an F-test of the hypothesis that the differentials at the 10 th and the 90 th percentiles are equal. Industries are ordered from lower to higher conditional wage differential as estimated over the whole distribution (i.e. following the estimates presented in Tables A3 and A4). Finally, Table A9 presents the average difference in the differentials for lower-wage industries and the average difference in the differentials for higher-wage industries. The evidence from Table A8 reveals that, in most countries and industries, the differences between 90th and 10th percentile are significant, but in most of the cases the wage differentials are higher in the lower end of the distribution (10 th ) than in the higher (90 th ), which goes against the Martins hypothesis. Table A9 summarises this point by showing that the differences in the returns between the 90 th and the 10 th percentile of the distribution in most countries, especially in the second wave, are not higher for the industries classified as better payers. We therefore do not find evidence in support of the unobservable quality hypothesis to explain industry wage differentials, but rather suggest that unobservable quality is not a crucial factor determining these differentials. 7 Explaining variations in the conditional wage differentials through aggregate level variables [very preliminary] The evidence provided in the previous sections (inability to find support for the unobserved quality hypothesis, persistence of differentials over time) points to non-competitive forces as potential determinants of industry wage differentials. In this section we investigate the role aggregate level economic developments play in explaining the conditional industry wage differentials and their changes over time. In terms of the equations presented in Section 5 the object of interest in this section (the dependent variable) is variable d h from equation 2. Given that the estimated differentials are available for a number of countries and for two points in time we exploit the extra variance. 15

16 As regards variations over time in the conditional wage differentials, despite the strong correlation of these between the two points in time (Table 7, Section5), changes have occurred both regarding the size and sign of the differentials, as well as in terms of industry rankings. The idea is here to compare the changes in the differentials in order to find out more about the role of institutional factors: the rigidities of labour markets in Europe. The exact time points for which the evidence is available makes this exercise even more interesting given that significant changes took place in the operation of European product and labour markets which are expected to have had an impact on national wage structures. Wage structures are anyway expected to show some link with the business cycle although the direction of this remains a controversial issue. On the one hand, Lewis (1963) illustrates that while there is no long-run trend in wage dispersion across industries in the US, in the shortrun dispersion is counter-cyclical. 14 On the other hand, Okun (1973) argues that industry wage dispersion is cyclical since industries paying the highest wages also create more jobs and thus in good times wages in these industries increase further. Notwithstanding the difficulty in clearly identifying cycles, the period contains some cyclical variation (see Figure 1). Furthermore, the number of people employed increased substantially: the employment rate in the EU-15 increased by around 4 percentage points in the period , the unemployment rate was on a declining trend (see Figure 2) despite a significant increase in the labour force due to increased inward migration flows in most of the countries we are looking at. In addition, certain structural changes took place in the labour market. Firstly, competition in the labour market intensified in this period due to, on the one hand, the aforementioned migration flows and, on the other hand, the gradual implementation of less restrictive employment legislation in some countries together with an increased social consensus for restrained wage increases in others (e.g. Ireland and Germany) driven mainly by the realization of the growing potential for production outsourcing. Secondly, the number of individuals working part-time or on fixed-term contracts increased significantly (Figure 3), and finally, the shares of different sectors changed with clear gains for the service sector (see Table 10). In the product market, the barriers to entry appeared to fall in many sectors as exemplified by inter alia the OECD indices of product market regulation (see, inter alia, Conway et al and 2006), despite the fact that profit shares have not in general narrowed. The above changes took place to varying degrees in different countries and industries thus contributing to the variation of the industry wage differentials.. 14 A conclusion confirmed by Wachter (1970). 16

17 Table 10: Share of employment by sector of economic activity in 1995 and 2002 BE '95 BE '02 DE '95 DE '02 ES '95 ES '02 GR '95 GR '02 HU '95 HU '02 IE '95 IE '02 IT '95 IT '02 NL '95 NL '02 Agriculture, hunting, forestry and fishing Mining and quarrying Total manufacturing Electricity, gas and water supply Construction Wholesale and retail trade, restaurants and hotels Transport, storage and communication Finance, insurance, real estate and business services Community, social and personal services Total services Business sector services Source: OECD STAN Database (2005) and EU-KLEMS. 17

18 Figure 1: Real GDP growth (%) Figure 2: Unemployment rate (%)EU-15, Real GDP in the EU-8 countries (% grow th rate) Unemployment rate for EU-15, % 3.50% % % % % 1.00% 0.50% 0.00% % % Source: Eurostat Source: Eurostat Figure 3: % part-time workers and fixed term contracts Percentage of part-tme w orkers and employees on fixed-term contracts, EU-15 ( ) Source: Eurostat % part-time workers in total employment % of employees with fixed term contracts Our specific focus is to investigate the extent to which inter-industry wage differentials are driven by factors such as sectors ability to pay (rent sharing hypothesis) and product market competitiveness more general, exposure to international trade or labour market institutions. Our sample is particularly well suited for this purpose as it covers 8 EU countries with significant differences in the structure of product and labour markets. Table 11 shows some very preliminary results 15 where industry wage differentials are estimated as a function of: (i) Industry performance variables: these include the capital to labour ratio (value of capital services per hour worked) and a measure of labor productivity. (ii) Measures of competition. We include a variable measuring firm concentration as a proxy to competition, namely the share of small firms: share of firms with less than 20 employees divided by the total number of firms. 16 (iii) Institutions: in this first attempt we try the OECD indicator of product market 15 Note: these results are very preliminary and are subject to a number of shortcomings as omitted variable bias, potential endogeneity, that we need to address. 16 Not available for Greece and Hungary 18

19 regulation (PMR) for 1998 and 2003 and Union density 17. In fact PMR could be understood as an additional proxy of the extent of competition; a higher PMR, acts as a barrier to entry thus limiting competition between firms and perhaps leading to higher wage differentials. The capital-labour ratio, the share of small firms and the labour productivity are sectoral or industry specific variables; and we express them as deviation from the country average. PMR and union density are only available at the country level. We introduce these institutional variables in the regression interacted with a dummy that takes value minus one if the sign of the dependent variable is negative and one otherwise. 18 This transformation allows interpreting the coefficient in terms of effect of the original variable on the inter-industry wage dispersion. Accordingly, a positive coefficient means that the variable tends to disperse wage premia and a negative coefficient means that it tends to shrink inter-industry wage differentials in a specific country. Column 1 in Table shows that productivity and the capital to labour ratio can explain more than one fourth of the variation of industry wage differentials. Coefficients are low but statistically significant. Regarding measures of competition, the share of small firms (column 2) has a negative and significant coefficient, this negative relationship could support the idea that more intense competition in the product market leads to lower profits and thus less space for firms to increase wages. PMR turns out significant in the estimation and has a positive coefficient, i.e. inter-industry wage dispersion increases with PMR (column 3). Both the share of small firms and PMR remain significant when added together to the regression (column 4). Specifications in columns 5 to 7 add union density to test whether inter-industry wage dispersion decreases with trade union density, and its interaction with productivity; the coefficient of this interacted measure could suggest whether unions prevent wages from adjusting to productivity. Results seem to indicate that unionization does compress the wage distribution and unions prevent wages from adjusting to productivity. Columns 8 and 9 include industry and country dummies. Both remain significant, suggesting that other sector and country specific terms are omitted from the model. The country dummy does not seem to affect the other coefficients and the significance levels. On the other hand, once we add industry dummies, the capital-labour ratio and the share of small firms are no longer significant OECD data, 2001 data are used for Belgium, Spain, Ireland, Greece and Hungary 18 Similarly to the country effect in the ANOVA exercise. 19 The source for all explanatory variables used in Tables 11 and Table 12 is the OECD and Table A10 in the Appendix provides their definitions.. 20 One possible explanation is that these structural measures do not change significantly over time and including industry dummies does not leave enough variation in the data. 19