Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
Driving forces behind the sectoral wage costs differentials in
Europe∗
Camille Logeay∗∗
Macroeconomic Policy Institute(IMK), Duesseldorf
Sabine StephanMacroeconomic Policy Institute
(IMK), [email protected]
Rudolf ZwienerMacroeconomic Policy Institute
(IMK), [email protected]
This version: July 18, 2008
∗We thank Simon Sturn for helpful comments, the usual disclaimer applies.∗∗Corresponding author, Camille Logeay, IMK, Hans-Boeckler-Str. 39, D-40 476 Duesseldorf, Ger-
many; phone +49/211/7778 334, fax +49/211/4 7778 334.
1
Abstract
In 2004, Eurostat starts publishing new figures on hourly wage costs for all Eu-ropean countries. These figures are new in several respects: It is the first time thatinternationally comparable hourly figures on wage costs are available covering a quiteimportant time period (1995-2005), so that not only cross-country comparisons butalso dynamic analyses are possible. Furthermore, these figures are fairly detailedat the sectoral level, therefore allowing for inter-sectoral comparisons. ConcerningGermany, the Eurostat statistics provide quite unexpected insights; the gap betweenwage costs in the manufacturing sector and the (private and business) services sectoris much larger than in other countries. This study aims at giving some explanations.According to theory, various explanations are possible. First, the neo-classical the-ory emphasises factors affecting or indicating the level of individual productivity,as well as firm or sectoral productivity; indicators corresponding to this approachare tested. Second, dropping the assumption of perfect competition on both labourand goods markets allows for other factors (mark-up, market power) to influencethe wage costs levels; these potential determinants are also tested. Finally, we thinkthat the structure of demand (driven by domestic or foreign demand) could alsohave a major impact on wages in the industry and the services sector and indeedthis factor seems to play an important role. This paper is structured as follows:First, the new Eurostat statistics is presented focussing on some interesting descrip-tive results. In the second section, we present a list of potential determinants ofwage differentials between the industry and the services sector derived from theoryand literature. A bivariate analysis (correlation) is then performed and conclusionsare drawn. In a third step, a multivariate analysis (panel estimation) is performed.The final section concludes.
JEL: J31, C23, E24
Keywords: Wage differentials, Europe, sectoral level, macroeconomic panel.
2
Contents
1 Introduction 4
2 Overview of the literature 5
2.1 Theory: several competing explanations . . . . . . . . . . . . . . . . . . 5
2.2 Empirics: Strong evidence for productivity and rent-sharing factors . . . 6
3 Bivariate analysis: A correlation assessment 9
3.1 The data to explain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2 Individual factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.3 Firm and sectoral factors . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.4 Country specific factors: institutional settings . . . . . . . . . . . . . . . 18
3.5 Growth composition indicators . . . . . . . . . . . . . . . . . . . . . . . 20
4 Multivariate analysis 23
4.1 Static regressions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
5 Conclusions 26
References 27
6 Annex: Overview of the estimators used 29
6.1 Pooled OLS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
6.2 Static Fixed Effects (SFE) . . . . . . . . . . . . . . . . . . . . . . . . . . 29
6.3 Static Random Effects (SRE) . . . . . . . . . . . . . . . . . . . . . . . . 30
7 Annex: Scatter diagrams of the data 32
3
1 Introduction
In 2004, Eurostat starts publishing new figures on hourly wage costs for all Europeancountries. These figures are new in several respects: It is the first time that internation-ally comparable hourly figures on wage costs are available covering a quite importanttime period (1995-2005), so that not only cross-country comparisons but also dynamicanalyses are possible. Furthermore, these figures are fairly detailed at the sectoral level,therefore allowing for inter-sectoral comparisons.
Concerning Germany, the Eurostat statistics provide quite unexpected insights; thegap between wage costs in the manufacturing sector and the (private and business)services sector is much larger than in other countries. This study aims at giving someexplanations. According to theory, various explanations are possible. First, the neo-classical theory emphasises factors affecting or indicating the level of individual produc-tivity, as well as firm or sectoral productivity; indicators corresponding to this approachare tested. Second, dropping the assumption of perfect competition on both labour andgoods markets allows for other factors (mark-up, market power) to influence the wagecosts levels; these potential determinants are also tested. Finally, we think that thestructure of demand (driven by domestic or foreign demand) could also have a majorimpact on wages in the industry and the services sector and indeed this factor seems toplay an important role.
This paper is structured as follows: First, the new Eurostat statistics is presentedfocussing on some interesting descriptive results. In the second section, we present alist of potential determinants of wage differentials between the industry and the servicessector derived from theory and literature. A bivariate analysis (correlation) is thenperformed and conclusions are drawn. In a third step, a multivariate analysis (panelestimation) is performed. The final section concludes.
4
2 Overview of the literature
2.1 Theory: several competing explanations
There are different theories explaining persisting wage differentials. The neo-classicaltradition focuses on factors affecting or indicating the level of individual productivity aswell as firm and sectoral productivity respectively. Theories dropping the assumption ofperfect competition on both labour and goods markets consider additional determinants(mark-up, market power) influencing wage cost levels. We think, however, that thestructure of demand (domestic versus foreign demand) could also play an importantrole in determining wages in specific sectors.
Schramm (2004, chap. 3) distinguishes three different explanations for cross-sectoralwage differentials: First, the neo-classical approach assuming perfect competition onthe labour market and explaining wage diffrentials due to different labour/job qualityor compensation for non-monetary remunerations. Second, the efficiency wage theorybroadens the neo-classical view assuming that wages above the neo-classical equilibriumgive workers an incentives to work in a more efficient manner. In doing so they producea rent which can be divided among employers and employees (rent-sharing). More-over, employers act reasonable if they pay more than the equilibrium wage because ofmoral hazard problems. The third approach focuses on non-competitive features on thelabour and goods markets due to institutional set-ups that give rise to rent-sharing forboth sides, independently from the productivity level. Sociological aspects are also putforward to explain the persistence of the differences over the business cycles.
Schettkat (2006, chap. 2) quotes the same explanations and puts more emphasison the assumption of a monopsonistic labour market (Manning 2003). In this model(contrary to the efficiency wage) the actual wage is fixed below the optimal level. Alot of empirical case studies, esp. in relation to the introduction of a minimum wage inthe UK (Metcalf 2007), show that several sub-branches of the service sector are bettermodeled as a monopsony rather than within a perfect competition framework (Card andKrueger 1994, Dickens and Katz 1986, Card 1996, Machin and Manning 2002). Thus notonly productivity differences but also the structure of the labour market could explainwage differentials across sectors.
The efficiency wage theory assumes that the wage level directly affects workers’ ef-
5
forts and that the information about the strength of this effort is asymmetric; workersknow, employers can only guess (Shapiro and Stiglitz 1984, Krueger and Summers 1988).Consequently, moral hazard problems occur. This is the reason why employers have in-centives to take workers’ interests into account. Wages above the optimal wage level(without efficiency wage consideration) produce involuntary unemployment but reduceturn-over (hiring/firing costs) and provide incentives for high-productive workers to ap-ply and stay in the firm and to increase their work-effort. Therefore, cross-sectionalwage differences can be due to different productivity/qualifications of workers as statedin the neo-classical theory; but they can be modified according to the social consensus– if society prefers a more equal income distribution, we would face a smaller wagedispersion (due to a higher wage compression) than the one occuring according to pro-ductivity/quality differences in the specific sectors.
Nowadays, institutional factors are also put forward; especially the presence of acoordinated wage bargaining system should reduce wage dispersion across sectors andqualifications. Thus, a high union density or a high coverage of union agreements shouldgo hand in hand with a smaller wage differential (Layard, Nickell, and Jackman 1991,“the battle of markups”). The market power of trade unions, however, probably dependson the firm size; Very small firms could be able to resist trade unions’ demands moreeasily than bigger ones (where strikes could be more likely and costly)1. Not only thesize of the firms may play a role but also the amount of the rent the owner receives.A higher rent could encourage workers to organize in a trade unions in order to forcethe owner to share it. In the insider-outsider theory the more institutions protect theinsider, the higher the wage they can achieve (Lindbeck and Snower 1988, Lindbeck andSnower 2002).
2.2 Empirics: Strong evidence for productivity and rent-sharing fac-
tors
According to the neo-classical theory, only differences in the labour productivity or inthe job-conditions can explain wage differences. Schramm (2004), however, reports thatempirical studies on job-conditions (health, ecological, hardness, job-security, working-time and vacations, non-monetary rewards, ...) come to the conclusion that the onlyrobust determinant explaining wage differences is a higher death probability. Thus, dif-
1The idea is also that workers in smaller firms may be less attainable by trade-unions.
6
ferent wages in the industrial and the service sectors can only be attributed to differencesin the ability and productivity of workers in these sectors.
Regarding Europe, Genre, Momferatou, and Mourre (2005) give a broad descriptiveanalysis of the wage differentials within the Euro Area. Using essentially OECD data,their wage data differ from our data. Nevertheless, they find similar results: the rankingof the countries did not change much over time. On average higher wages are paidin the industry compared to the service sector. This cannot be fully explained bypart-time, self-employment or age effects nor by qualification effects (that even have anopposite effect). Gender appears to play no role. The individual productivity-approachhowever cannot fully explain the differences. The alternative theories provide variablesas capital intensity, average firm size and sectoral labour productivity. This means thatrent-sharing theories are in line with the observed data.
Haisken-DeNew and Schmidt (1999) confirm with more recent data for Germany andthe US that even after controlling for human capital components, job characteristics,status and geographical factors, the inter-industry wage structure is very persistent onboth sides of the Atlantic.
Freeman and Schettkat (2001) compare the skill distributions among the unemployedand the employment in USA and Germany, as well as the wage distribution in these twocountries and come to the conclusion that the productivity-theory, i.e. the neoclassicalone, is not able to explain the better employment performance in the USA.
In most studies focussing on the USA the inter-industry wage differentials are foundto be stable over time, even after controlling for unionization and observable character-istics of the workers and job characteristics. Since these differences are highly stableover time, they cannot be attributed to transitory supply or demand shocks affecting aspecific industrial sector. Thus, the competitive theories seem to be unable to explainthese differentials. Some authors argue that non-standard theories would do a betterjob; efficiency wage theories for example can explain why the quit rate in high-wage-industries is lower (Krueger and Summers 1988), or rent-sharing theories can explainwhy profitability seems to be higher in high-wage industries (Dickens and Katz 1986).On the other side, some authors argue that non-observed abilities of workers may affectthe inter-industry wage differentials even in the long-run, since these abilities may bejudged or perceived differently across industries (Gibbons and Katz 1992). In this latterexplanation the competitive model is not challenged anymore, however, the empirical
7
evidence is not that clear.
Martins, Scarpetta, and Pilat (1996) find that market power exist in the studiedindustrial sectors and differ across products.
Suedekum and Blien (2007) use a German panel for 326 regional districts and 28industries & services categories between 1993/5-2002 and find that 70% of the wagedifferential across regions and industries can be explained by industries and area typefixed effects as well as time-varying characteristics such as qualification (low, mediumand high), firm size (small, medium and large), age and gender structure. In a secondstep they regress the employment growth rate on the unexplained regional wage differen-tial and find out that the elasticity is strongly negative for export-oriented and exposedsectors such as manufacturing and insignificant (and even positive) for rather domes-tically oriented service sectors (gastronomy and household-related services). Thus, forexport-exposed industries, the supply-side effects of wages are dominant whereas for theservice sector the demand-side aspects of wages have a bigger impact (however the au-thors claim that the supply-side effects dominate for both type of industries). It is alsoremarkable that for Germany the qualification variable is right-signed (branches andregions with higher workers pay higher wages too), the effect of age is slightly positive,as the firm size and the proportion of men.
Erdil and Yetkiner (2001) perform a macroeconomic panel using the STAN-OECDand UNIDO databases. After showing that the wage differentials within the manufac-turing sector is constant over the whole available sample period (1970-1992), explanationfor these differentials are put forward. For the OECD countries only labour productiv-ity and gender have the right sign and robust effects. Profitability and competitivenessseem to play a role only in 1990. However, their dataset is quite restrictive regardingpossible control-variables; working-time, nationality or race, qualification, country sizecannot be accounted for example.
8
3 Bivariate analysis: A correlation assessment
3.1 The data to explain
3.1.1 Sources and definitions
The labour costs analysed in this study are taken from the Labour Cost Statistics(Eurostat)2.
The wage costs per hour include social security contributions of both employees andemployers, as well as the personal income tax paid by the employees. Some additionaland quite negligible costs paid by the employers are also included in the statistics (work-ing clothes, ...).
The countries covered are the EU-27 countries; i.e. EU-25+2: Euro Area(12) + UK,Denmark, Sweden and the new 10 Eastern countries + Romania and Bulgaria3. WhileIreland is completely missing, others have very few data (Hungary, Malta and Italy forexample).
Note, that for the EU-15 only those employees are considered which are workingin firms employing at least 10 employees, whereas for the new member states all areconsidered. This implies that the wage cost levels are probably over-estimated in theEU-15, since small firms pay lower wages on average (see detailed Eurostat statistics forthe NMS-10 grouped by firm size).
In this study we focus on the sectors C to K with CDE = Industry (without con-struction) and GHIJK = business services (without state and near-state services likehealth care or education)4.
The study is based on annual data covering the years 1995 to 2005. For somecountries the data base is imcomplete, i.e. we face an unbalanced panel.
2See: http://epp.eurostat.ec.europa.eu/ → Data → Labour Market → Labour costs → Labour costsannual data → hourly labour costs.
3As the data ends in 2005, Slovenia is not a member of the Euro Area but belongs to the 10 newmember states.
4More and more data for the other sectors (AB and LMNO) are available. However, these sectorsare either negligible in size or cannot be considered as market-driven and are thus not considered in thisstudy.
9
Chart 1: EU-27-Country ranking for the hourly wage costs in e, in 2005 and for themarket economy (C-K) (Data-update: Aug. 2007).
3.1.2 Two interesting results
1. Contrary to the broadly accepted view (Sinn 2007, Schroder 2007) Germany is not thecountry with the highest wage costs. In fact, several Northern and Western Europeancountries form a bulk of high-wage countries, while Germany has a rather middle positionwithin the EU-15. The Southern countries and especially the new member states arewell below the Western average. See Chart 1 and a detailed descriptive analysis inDuthmann, Hohlfeld, Horn, Logeay, Rietzler, Stephan, and Zwiener (2006) and Horn,Logeay, Stephan, and Zwiener (2007).
2. The second interesting result is the fact that in Germany the wage costs in theindustry sector differ largely from those in the service sector. Explaining this puzzlingresult is the aim of this paper.
Thus, the dependent variable in our empirical study is the hourly wage costs inthe industry (CDE) in relation to those in the market service sector (GHIJK). A valueabove 1 indicates that the average wage costs in the industry are higher than those in
10
the service sector. Germany has the highest value with 1.25, meaning that industrysectors pay on average 25% more than service sectors! At the opposite the Portuguesewage costs in service sector are on average about 20 to 30% higher than those in theindustry5. As we can see in Chart 2, this result is stable over the available years. Adistinction between the “old” European countries (EU-15) and the others (EU-27) ismade.
Chart 2: Relative wage costs in Europe: Industry/Services
atat
at at at at atat
atbe be be be
be be bebe be be
de de de de
de de de de de de
es es es es eses es
es es es
fi fi fi fifi fi
fi fi fi fi
fr fr fr fr fr fr fr fr fr fr
gr gr gr gr gr gr gr gr gr
itit it it
it it it itit it
lu
lulu lu
lulu lu lu
lu lu
nl
nlnl nl
nlnl nl nl nl nl
pt pt pt ptpt pt pt pt pt pt
dk dk dkdk
dk
dk dk dk dk
se se sese
se se se se
uk
uk uk uk uk uk uk ukuk
uk
cy cy cycy cy cy
cy cycy
cz cz czcz
cz czcz cz
cz
cz
ee ee ee
eeee
ee
ee eeee
ee
hu
huhu hu
hu huhu hu hu
hu
lt lt
lt ltlt
lt ltlt lt
lt
lv lvlv
lvlv
lv lv lv lv
mt
mt
mt
mt
plpl
pl
pl pl pl pl pl pl pl
si sisi si si
si sisi si
si si
sk
sk sksk
sk
sksk
sk sk
sk
bgbg
bg
bg bgbg bg
bg
.6.8
11.
21.
4lk
_neu
: rel
ativ
e ho
urly
wag
es (
indu
stry
/ se
rvic
es)
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
Source: Eurostat, own calculations. Some countries are missing (Ireland for example).
3.2 Individual factors
As introduced above, we focus on different categories of explanatory variables. Thefirst set which is in line with all theories are the individual features. These variables
5Interestingly the contribution of domestic demand to GDP growth was 0.8%-points in Germany and3.7%-points in Portugal between 1995-2005!
11
are indicators for the individual productivity (stemming from working experience oreducation), or for some inherent social discrimination (that per se says nothing aboutthe individual productivity but rather how high social discrimination is or may indicatesome hidden specific job characteristics). The correlations are shown in Table 1 (p. 14).
3.2.1 Age/Experience
Data for the age composition of the sectoral labour force is available from the LabourForce Survey (Eurostat). Two variables are constructed: a variable for young employ-ees (15-24 years old: alter 1524) and one for elder employees (more than 50 years:alter 50+). The variables are constructed similar to the wage costs: a value above onemeans that the proportion of younger resp. elder employees in the industry is higherthan in the service sector.
alter 1524 =Young employees in the industry/All employees in the industryYoung employees in the services/All employees in the services
(1)
alter 50+ =Elder employees in the industry/All employees in the industryElder employees in the services/All employees in the services
(2)
The age can be regarded as a proxy for the working experience and thus the ac-quired qualification of the employee. Usually, a hump-shaped productivity-age-curve isassumed. However, in empirical studies the curve is rather flat at the end, so that areal decline at higher ages is probably not significant. Therefore, a negative sign for the15-24-variable is expected and a presumably negative albeit not significant correlationfor the 50+-variable.
Actually, the correlations are not significant, an indication that either the age playsno role, or it is a rather poor indicator for acquired qualification or of course othercounter-acting effects linked to the age are offsetting the presumed effects. However, itis worth noting that from the scatter-plots in the annex (p. 33), Luxembourg seems tofollow another logic than the rest of the EU. If we exclude this country, the correlationbetween alter 50+ and the relative wage costs is significantly negative for both theEU-15 and EU-27 with ca. -35%. Therefore, the hypothesis of an hump-shaped age-productivity-curve is not true or there are strong institutional features (seniority clauses)of the wage bargaining inducing higher wages for elder employees, independently fromtheir productivity.
12
3.2.2 Gender
Data for the gender composition of the sectoral labour force is available from the LabourForce Survey (Eurostat). The variable (gender) is constructed similar to the wage costs:a value above one implies that the proportion of men in the industry is higher than theproportion of men in the service sector.
gender =Male employees in the industry/All employees in the industryMale employees in the services/All employees in the services
(3)
There is no “theoretical” explanation – along the productivity-theories – why womenshould earn less than men. But it is a stylized fact reported in various and numerousgender studies, that women are discriminated with respect to wages. Thus, omitting thisvariable could yield to a severe bias in quantifying the effects of the other variables. Thisvariable is correlated with the part-time proportion, probably indicating a difference injob quality between men and women.
The sign of the expected effect is positive. And indeed the correlation is strong andpositive.
3.2.3 Qualification
Data for the formal qualification composition of the sectoral labour force are availablefrom the Labour Force Survey (Eurostat), according to the ISCED classification. Theemployed persons are divided into three groups according to the highest level of edu-cation attained: low, medium and high qualifications6. For each sub-sectoral group,a weighted mean was calculated. Again, if the variable (qualification) is above one itmeans that on average employees in the industry are (formally) better qualified thanthose in the service sector.
qualification =Weighted mean of the qualification categorical variable in the industryWeighted mean of the qualification categorical variable in the services
(4)6Low: At most lower secondary (ISCED 0-2); Medium: Upper secondary (ISCED 3-4); High: Tertiary
(ISCED 5-6). See:http://forum.europa.eu.int/irc/dsis/employment/info/data/eu lfs/f lfs statistical classifications.htm.
13
As for the acquired qualification, a positive effect (more qualification should go alongwith higher wages and thus higher wage costs) is expected here. On the other side, as forthe acquired qualification, only the individual qualification is taken into account here.Labour productivity gains induced by useing specific technologies or capital inputs in abroad sense may not be well reflected in this variable.
Actually, the correlations are significantly negative, which is in contrast to standardtheory. This result was also found by the ECB (Genre, Momferatou, and Mourre 2005).We interpret this result in a way that either this variable is a poor indicator for individualand formal qualification, or that this variable is correlated with some sectoral specificfeatures that distroy the true (positive) presumed correlation. In any case, this resultremains puzzling because one would expect that the necessary bias induced by thesebivariate considerations should not be so severe as to even produce a counter-intuitivesign.
Table 1: Correlation between individual factors and the wage costs differential betweenIndustry and Services for the EU-27 and the EU-15.
EU-27 EU-151995-2005
Relative wage costsALTER 1524 -4.2% -2.2%(15-24 years/all employees) 53.5% 80.2%
219 135ALTER 50+* 28.2% 31.8%(50+/all employees) 0.0% 0.0%
209 125GENDER 42.0% 59.0%(Men/all employees) 0.0% 0.0%
225 135QUALIFICATION -33.5% -27.9%(weighted mean: 1-low, 2-mid, 3-high) 0.0% 0.1%
216 132
Bold figures are significant at the 1%-level. The first number is the correlation coefficient (ρ),the second the probability associated with H0: ρ = 0 and the last number is the number ofobservations.*: w/0 Luxembourg due to big outlier.
14
3.3 Firm and sectoral factors
The second set of possible explanatory variables are the firm and sector-specific variables.They are also indicators for productivity in the sense that they are in fact not linked tothe personal abilities of the workers, but through capital-intensity still affect the overalllabour productivity. They can also be interpreted in line with the rent-sharing theories;the bigger the firm, the more likely it possesses some market power and therefore receivesa monopolistic rent that can be shared with the workers, leading to wage differences thatcannot be explained due to productivity differences. The correlations are shown in Table2 on p. 17.
3.3.1 Firm size/capital intensity
Data on firm size are very difficult to obtain and are even more difficult to get for a longtime period. Data from Eurostat could be collected essentially for the middle of thesample (1999-2001; Eurostat, annual Structural Business Statistics – SBS). Because of alack of data, some calculations based on the national accounts statistics were performed.Therefore, one should not rely too much on this variable since the measurement erroris likely to be important. The variable (firmsize) is also constructed similar to the wagecosts variable. A value above one means that firms in the industry are larger on averagethan those in service sector.
firmsize =Average number of employees per firm in industryAverage number of employees per firm in services
(5)
This variable is a proxy for capital intensity or the degree of market power in thesector: the larger a firm, the bigger the economies of scale, and thus the larger theamount of capital per head but also the more likely the fact that the firm possessesmarket power. The first explanation yields a positive effect on labour productivity andhence on wages, the second through the rent-sharing among employers and workers apositive effect on wages too. Thus, in both cases, a positive correlation is expected butcannot be shown for the EU-15. Due to the small number of observations, it is difficultto state significance at all.
For Germany, however, the Federal Statistical Office (Destatis 2006, p. 15, Chart 4)shows that the size of firms and the wage costs are positively correlated and that this
15
is a long-term feature (1992-2004). Capital intensity (measured as GCF/head) is alsopositively correlated through the German branches in 2004 with the amount of the wagecosts.
3.3.2 Self-employment ratio/capital intensity
Data on self-employment and employment in a sectoral disaggregation can be takenfrom the national account statistics (Eurostat) but we choose the labour force survey asit is more complete (at least for the EU-15 countries). The variable (selfemploy) takesa value above one if the proportion of self-employed in the industry is larger than thosein the service sector.
selfemploy =Self-employed/All employment, in the industrySelf-employed/All employment, in the services
(6)
This variable is also a proxy for the firm size (the correlation between the two vari-ables is indeed significantly negative with about -65%) and consequently for the capitalintensity and/or for the market power of the firms. Because the link is reversed (a largerproportion of self-employed goes along with a smaller average firm size), we expect anegative sign. Indeed the correlation is negative, however it is not significant.
3.3.3 Part-time
Data for part-time jobs on a sectoral level are available from Eurostat only for the year2000 and from the labour costs statistics. Thus this variable is quite restricted. Againthe variable takes values above one if the part-time ratio is higher in the industry thanin the service sector7.
teilzeit =Part-time employees/All employees, in the industryPart-time employees/All employees, in the services
(7)
Similar to gender, there is no “theoretical” justification for an influence of this vari-able, as the labor costs are already corrected for the hours worked. However, like forgender, it is a stylized fact that the hourly wages for part-time employees are lower than
7This was calculated for employees of firms which employ at least ten employees.
16
those of full-time employees (OECD 1999, p. 22-25). Thus, one should take this intoaccount. This variable is negatively correlated with the gender variable (-30%, meaningthat if the fraction of female employees is high, part-time jobs are high too) as men-tioned above, so that this variable may be an indicator for social discrimination and forjob quality. This would yield a negative correlation.
The correlation is indeed negative for the EU-27 (albeit not significant) but is unex-pectedly positive for the EU-15 (but not significant either). These correlations shouldbe interpreted with caution because of the very small sample size.
3.3.4 (Labour) Productivity
Productivity measures are only available on a per-capita basis, because figures for hoursare not available for all countries. Working time reductions have reduced the produc-tivity per head but increased wages per hour in the past. Besides, taking the ratio ofsectoral gross value added to the sectoral employment levels is thought to be highlyendogenous and may yield to strong endogeneity problems. Therefore, productivity percapita cannot be used here.
Table 2: Correlation between firm-specific factors and the wage costs differential betweenIndustry and Services for the EU-27 and the EU-15.
EU-27 EU-151995-2005
Relative wage costsFIRMSIZE 12.0% -13.1%
31.9% 42.8%71 39
SELFEMPLOY* -15.8% -33.9%(Selfemployed/all employment) 2.3% 0.0%
207 125TEILZEIT -12.1% 15.9%(Part-time/all employees) 57.3% 58.7%
24 14
Bold figures are significant at the 1%-level. The first number is the correlation coefficient (ρ),the second the probability associated with H0: ρ = 0 and the last Number is the number ofobservations.*: w/0 Luxembourg due to big outlier.
17
3.4 Country specific factors: institutional settings
The third set of possible explanatory variables are the country-specific variables. Clearly,they are indicators for failures of the neo-classical framework, as they indicate someinstitutional features of the labour market and show whether market power plays asignificant role. The correlations are shown in Table 3.
3.4.1 Unemployment rate/strength of workers
Data on the standardized unemployment rate are provided by Eurostat (ILO-concept).The unemployment rate here is the national-wide rate (no sectoral differences are avail-able). This should be an indicator for the market power or relation between employersand employees as predicted by the insider/outsider theory: A higher unemployment rategoes along with a weaker bargaining position of employees. By which extent this shouldaffect employees in industrial sectors more than in service sectors is an open question.But if no other indicators are included, as in the years observed (1995-2004), the struc-tural change operates in favor of services: A higher unemployment rate should weakenthe industry employees more than those in the service sectors on the one hand (negativeeffect), on the other hand a higher unemployment rate may be rather an indication forpoor macroeconomic performance, which could weaken the service sector more than theindustry8 (positive effect).
The correlation is significantly positive which gives more evidence for the secondexplanation.
3.4.2 Minimum wage/institutional factors
A dummy variable takes value 1 if the country has a minimum wage and 0 otherwise.There are only seven countries without legal national minimum wage throughout thesample (Austria, Germany, Finland, Italy, Denmark, Sweden and Cyprus), the UK has
8Industry-firms are thought to be more likely to export their products than Services in general. This isconfirmed by the Input-Output tables for Belgium, Germany, Ireland, Spain, France, Italy, Netherlands,Austria, Portugal, Finland, Denmark, Sweden, UK, Lithuania, Hungary, Slovenia and Slovakia for theavailable years between 1995 and 2004. By this way, industrial firms may compensate domestic slacknessthrough better foreign demand more significantly than firms in the Services.
18
adopted a minimum wage in 1999.
Another interesting variable would be the proportion of employees covered by min-imum wage(s); as we exclude civil servants and employees of the social services, forcountries where the minimum wage is legal and national, this coverage rate will be100%9. More interesting is the proportion covered in the North-South axis of countriesthat do have minimum wages only at sectoral levels and set by collective agreements(Italy, Germany, Austria, Denmark and the three scandinavian countries). From Funkand Lesch (2005) and Husson (2006) some figures are available for some years. It ap-pears that Austria, Denmark, Finland and Italy have high coverage rates (above 85%)whereas Germany and Norway have low coverage rates (around 70% in 2004, but herewe suspect that this concerns only West-Germany10). A closer look at low-pay sectorsthat are all in the Services (textile/clothing, retail, hotel/restaurants, hairdressing) alsoshows that the coverage of the minimum wages set by the collective agreements aremuch smaller in those two countries than in the others. Due to lack of data however,the aim to construct such a variable was abandoned.
A national-wide minimum wage affects (almost) all employees. It should reduce thewage differentials because it increases the very low wages in all sectors. Thus the absenceof a minimum wage should be associated with more extreme values of our dependentvariable and the presence of a national minimum wage with values concentrated aroundone. This non-linear effect can be catched if we divide the sample in two (countries/yearswith a value of the dependent below one and countries/years with a value above one).In the first group a positive coefficient is expected and in the second a negative one.
For the whole sample, a significant negative correlation for both variables can beobserved: the presence of a minimum wage is significant. The higher its level, the lowerthe industry/services wage gap. For the split sample, the expected non-linear effect isnot met. Even counter-intuitive, the negative correlation in the first sub-sample appearsto be twice to three times higher than in the second sub-sample (see Table 4).
9In Belgium the minimum wage concerns only the employees in the private sector. As we only lookat the private sector, excluding construction and the primary sectors, the coverage rate is 100%. InCyprus, the minimum wage concerns only some professions; sales staff, clerical workers, auxiliary healthcare staff and auxiliary staff in nursery schools, creches and schools. Thus the coverage rate is not 100%.
10From the WSI-Tarifarchiv (2006) figures, the coverage rate by collective agreements, where minimumwages are normally defined, varied from 76% to 67% between 1998 and 2005 in West-Germany and from63% to 53% in the East.
19
3.4.3 Union density rate/institutional factors
The union density rate is taken from the OECD-database (Bassanini and Duval 2006).The data set ends in 2003 and covers only Western European countries.
One would expect that the union density is especially low in service sector, as men-tioned above. Thus, a higher union density would rather go along with a smaller differ-ence between union coverages in the service compared to the industry sector. This wouldsupport a negative effect in the labour cost relation between Industry and Services. Thecorrelation here is negative but not significant.
It is remarkable that a non-linear effect can be found (the correlation coefficient issignificant positive for values of the dependent below one and significant negative forvalues of the dependent above one; See Table 4). The explanation is quite straightfor-ward and follows the arguments for a hump-shaped wage/centralization degree of wagebargaining a la Calmfors and Driffill (1988). The stronger the trade-unions, the morelikely they coordinate their actions and promote wage compression over all sectors lead-ing to a reduction of the wage differential. On the other hand, the less representative,the more firm-specific the trade-unions demands will be, yielding a higher wage disper-sion across sectors. This means that values well above and below one in our dependentvariable should be associated with low values of union density, and values about oneof our dependent variable should be associated with high values of union density. Thismeans that the correlation should be positive for values of the dependent below one andnegative for values above one.
3.5 Growth composition indicators
The idea for this fourth and last set of variables is that the demand addressed to asector plays at least as an important role in the wage determination as the relativecompetitiveness of this sector. Thus, we think that the wage costs in a sector are notonly reflecting the (marginal) productivity of this sector net of the employer’s rent dueto imperfect competition but also the structur of the demand (foreign versus domesticdemand) a specific sector is confronted with.
Here, however, a precision has to be made. As the sectoral wage differentials arepersistent over time i.e. over the business cycles, the state of demand, reflected within
20
Table 3: Correlation between country-specific factors and the wage costs differentialbetween industry and services for the EU-25 and the EU-15.
EU-27 EU-151995-2005
Relative wage costsALQ 14.4% 27.9%(Unemployment rate) 3.0% 0.1%
226 135MINWAGE IS -20.1% -15.7%(0: no minimum wage, 1: national MW) 0.2% 6.9%
242 135UDENS -13.5%(Union Density) 19.2%
95
Bold figures are significant at the 1%-level. The first number is the correlation coefficient (ρ),the second the probability associated with H0: ρ = 0 and the last number is the number ofobservations.
Table 4: Non-linear correlation between country-specific factors and the wage costsdifferential between Industry and Services for the EU-25 and the EU-15.
EU-27 EU-15All w < 1 w > 1 All w < 1 w > 1
ALQ 14.4% 28.5% -4.8% 27.9% 36.3% -2.4%3.0% 0.2% 62.1% 0.1% 1.0% 82.5%226 114 110 135 50 84
MINWAGE IS -20.1% -41.7% -30.9% -15.7% -75.2% -26.4%0.2% 0.0% 0.1% 6.9% 0.0% 1.5%242 122 118 135 50 84
UDENS -13.5% 66.3% -12.2%19.2% 0.0% 34.8%
95 33 61
Bold figures are significant at the 1%-level. The first number is the correlation coefficient (ρ),the second the probability associated with H0: ρ = 0 and the last number is the number ofobservations.
traditional business cycle indicators like the growth rate, cannot explain the wage gap.Here, we think rather in terms of long-term structure or strategies; Germany for examplehas a long tradition of using external trade to foster growth, whereas other Europeancountries push rather domestic demand. In that respect, it is likely that the participationin the Monetary Union has induced a change in preferences in economic policy in those
21
countries which traditionally focus rather on domestic forces than in countries with along experience of de facto monetary bindings like Germany, Netherlands or Austria.
In these respects, we thought that the indicator should be cumulative and relative,i.e. has a chance to persist over-business cycles, and reflects long-lasting differences inthe demand addressed to the industry and service sector. We look at the composition ofgrowth as an explanatory factor for the wage differential; industry sectors are tradition-ally more export-oriented, whereas the service sector typically depends on the domesticmarket (with some exceptions of course). Growth will not have the same impact on thewage opportunities for workers in the industry and the service sector, depending on thedriving forces; the German model of export-led growth should favour the industry morethan the service sector whereas the Portuguese model of domestic demand-led growthshould be more favourable for the services. The correlations are shown in Table 5.
Along this idea, cumulative and relative growth variables for the exports and domes-tic demand are constructed: the growth rate in % of exports resp. of domestic demandresp. of GDP was cumulated from 1996 onwards, calculating an index with value 100 in1995 (Malta and Romania start only 1999 due to lack of data); This index is then putin relation to the one of the Euro Area. A value above 100 means therefore that theaccumulated growth performance since 1995 is above average.
growth namek2i,t =
∏tτ=1996(
namei,τ
namei,τ−1)
∏tτ=1996(
EUR12i,τ
nameEUR12,τ−1)× 100 (8)
With name=exp, gdp or intdd for resp. exports, GDP and domestic demand in constantterms; i the country index and t the time index. The index starts in 1995 with the value100.
We calculate this indicator for exports, domestic demand and GDP at constantprices. The expected effects are positive for the first one, negative for the second one;for the third one it depends on the fact whether the first or the second effect is dominant.The correlations have the expected signs and are significant for the domestic demandand GDP. The results show that development of the domestic demand is much moreimportant than export demand.
22
Table 5: Correlation between growth structure indicators factors and the wage-costdifferential between Industry and Services for the EU-25 and the EU-15.
EU-27 EU-151995-2005
Relative wage costsGROWTH EXPK2 -1.3% 22.5%(cumulated export-growth relative to EUR-12) 84.3% 0.9%
242 135GROWTH GDPK2 -24.6% -29.7%(cumulated GDP-growth relative to EUR-12) 0.0% 0.1%
242 135GROWTH INTDDK2 -23.7% -41.1%(cumulated domestic demand-growth relative to EUR-12) 0.0% 0.0%
242 135
Bold figures are significant at the 1%-level. The first number is the correlation coefficient (ρ),the second the probability associated with H0: ρ = 0 and the last number is the number ofobservations.
4 Multivariate analysis
The advantage of performing a multivariate analysis which includes both dimensions(time and cross-sections) is twofold: first with an increasing number of observations,we expect to obtain more accurate estimates for the effects of individual variables onthe relative wage differential and second – contrary to the bivariate correlations – ina multivariate analysis, one controls for different effects at the same time and shouldbe able to isolate the genuine effect of a factor, reducing therefore potential omittedvariable bias.
The regressors we use, summarized in Xit, are [alter 1524, alter 50+, gender, qualification,selfemploy, alq, minwage is, growth expk2, growth intdd2k]. We dropped firmsize,udens and teilzeit because these variables had too few data. Still the data set remainsunbalanced. All variables are time-variable, whereas the dummy minwageis actually istime-invariant for all countries but for the UK11.
11It takes the value 0 between 1995 and 1998 and 1 after.
23
4.1 Static regressions
It is not obvious that the model should contain a dynamic term, since the series arequite stable (they are all relative and thus do not have trends). Because our dataset is a macro-panel with countries from the EU-27, it seems also reasonable to ruleout random-effects. At least a discussion about which variable may be correlated withcountry-specific unobserved characteristics should be done.
In principle, all variables may be correlated with the country-effects. But certainlythe unemployment rate, the presence or not of a minimum wage may be very likelycandidates. As we have only the minimum-wage dummy that may be considered asalmost time-invariant, the fixed-effects estimator may be the most appropriate.
lohnkostenit = X ′itβ + λ′DTt + α0 + αi + uit (9)
The results of the standard estimators are reported in Table 6. The Hausman-test (H0:Random Effects are present) is rejected and the Breusch-Pagan test (H0: Fixed effectsare present) is consistently not rejected. This is in line with our intuition that a fixed-effect model would better describe a macro-panel. The pooled estimation is reportedbelow:Nb of obs.: 201. Nb of param.: 20. Using robust standard errorsTime dummies: 10; Nb of individuals: 25longest time series: 11 [1995 - 2005]; shortest: 2 (unbalanced panel)
coefficient H-robust p-valALTER 1524 -4.58E-02 60.2%ALTER 50PLUS -2.51E-03 98.3%GENDER 3.88E-01 1.5%QUALIFICATION -4.70E-02 11.2%SELFEMPLOY -3.21E-02 83.7%ALQ 1.18E-02 3.7%MINWAGE IS -2.60E-03 94.4%GROWTH EXPK2 5.70E-04 45.4%GROWTH INTDDK2 -1.46E-03 19.4%Wald (X): χ2(9) = 34.55 [0.0%] ; Wald (time): χ2(10) = 16.94 [7.6%]AR-1 & 2 rejected at 1%. with PcGive.
24
Tab
le6:
Stat
icre
gres
sion
s:re
sult
sof
the
stan
dard
esti
mat
ors
Use
dpr
ogra
m:
STA
TA
Max
imal
Sam
ple:
26co
untr
ies
ofth
eE
U-2
7av
aila
ble;
1995
-200
5(1
1ye
ars)
End
ogen
ous:
Loh
nkos
ten
xtr
egxtg
eextg
lsxtp
cse
Est
imat
ion
Fix
edR
ando
mR
ando
mR
ando
mf(
gaus
s),i(
id)
pane
ls(h
eter
o)he
tonl
ym
etho
d:eff
ects
effec
tseff
ects
effec
tsro
bust
forc
epa
irw
ise
(wit
hin)
GLS
MLE
GLS?
corr
(exc
h)co
rr(a
r1)
corr
(ar1
)co
rr(p
sar1
)co
rr(a
r1)
corr
(psa
r1)
Alt
er15
2438
.6E
-333
.7E
-333
.5E
-333
.7E
-336
.0E
-354
.2E
-352
.7E
-3∗
25.9
E-3
36.0
E-3
37.1
E-3
Alt
er50
plus
43.2
E-3
39.2
E-3
39.0
E-3
39.1
E-3
41.0
E-3
115.
6E-3
∗∗∗
74.9
E-3
∗∗∗
65.3
E-3
∗∗∗
34.5
E-3
53.1
E-3
∗∗
Gen
der
80.9
E-3
112.
4E-3∗
113.
6E-3∗
112.
9E-3∗
98.3
E-3
117.
0E-3
208.
9E-3
∗∗∗17
3.2E
-3∗∗∗
244.
4E-3
∗∗∗20
8.1E
-3∗∗∗
Qua
lifica
tion
-3.1
E-3
-9.7
E-3
-9.9
E-3
-9.8
E-3
-6.7
E-3
-18.
2E-3
∗∗-4
5.9E
-3∗∗∗
-50.
6E-3
∗∗∗
-31.
1E-3
∗∗-5
5.2E
-3∗∗∗
Selfe
mpl
oy5.
7E-3
-8.2
E-3
-8.7
E-3
-8.4
E-3
-2.1
E-3
-5.8
E-3
-5.7
E-3
15.7
E-3
-10.
4E-3
13.6
E-3
Alq
-3.6
E-3
∗-2
.0E
-3-1
.9E
-3-2
.0E
-3-2
.7E
-31.
6E-3
3.2E
-3∗
1.4E
-37.
2E-3
∗∗∗39
7.7E
-6M
inw
age
is-8
.6E
-3-9
.4E
-3-9
.4E
-3-9
.4E
-3-8
.9E
-366
1.8E
-621
.0E
-344
.5E
-3∗∗∗
-32.
5E-3
∗30
.5E
-3∗∗
Gro
wth
expk
217
7.7E
-621
2.1E
-621
3.3E
-621
2.5E
-619
7.4E
-6-1
10.9
E-6
-40.
6E-6
-60.
8E-6
187.
9E-6
118.
1E-6
Gro
wth
intd
d2-1
.9E
-3∗∗∗
-1.7
E-3
∗∗∗
-1.7
E-3
∗∗∗
-1.7
E-3
∗∗∗
-1.8
E-3
∗∗∗-3
20.0
E-6
-238
.2E
-6-9
98.7
E-6
-583
.7E
-6-5
09.2
E-6
Cou
ntry
effec
tye
sye
sye
sye
sye
sye
sye
sye
sye
sye
sT
ime
effec
tye
sye
sye
sye
sye
sye
sye
sye
sye
sye
snb
.obs
.20
120
120
120
120
116
520
120
120
120
1nb
.par
am.
2020
2020
R2
wit
hin
19.4
%19
.0%
19.0
%R
2be
twee
n3.
5%8.
5%8.
6%R
2ov
eral
l6.
7%13
.6%
13.7
%P
rob>
F1.
2%P
rob>
χ2
0.0%
0.0%
0.0%
0.0%
0.0%
0.0%
0.0%
0.0%
0.0%
Pro
b>F
(TD
=0)
25.3
%0.
0%32
.2%
39.6
%0.
0%25
.1%
67.7
%53
.5%
0.0%
0.0%
xtre
g:O
LS
esti
mat
or;
xtge
e:M
L-e
stim
ator
asin
xtre
gbu
tha
ndle
dun
bala
nced
pane
ldi
ffere
ntly
;xt
gls:
GLS-
esti
mat
orw
ith
assu
med
hete
rosk
edas
tici
tybu
tno
corr
elat
ion
acro
sspa
nels
and
aar
(1)-
stru
ctur
eac
ross
tim
e;xt
pcse
:G
LS-
esti
mat
orw
ith
corr
elat
ion
inbo
thdi
men
sion
s(t
ime
and
cros
s-se
ctio
ns)
and
hete
rosk
edas
tici
tyac
ross
pane
ls;
psar
:th
eas
sum
edco
rrel
atio
nis
decr
easi
ngw
ith
the
tim
edi
stan
ce.
?:w
ith
smal
l-sa
mpl
eco
rrec
tion
for
the
stan
dard
erro
rs(S
wam
y-A
rora
)an
dno
n-co
nsta
ntth
eta.
*,**
,***
:si
gnifi
cant
at10
%,5%
,1%
leve
l.
25
5 Conclusions
In this article we analyze the new data provided by Eurostat on hourly wage costs,which are comparable across all European countries. Among other interesting features,a surprising result is the outlier position of Germany where in the industry sector wagesare on average 25% higher than in the service sector. The aim of this paper is to explainthis puzzle.
In a first and descriptive step, a bivariate correlation analysis was performed and thefollowing results were found: only gender is a significant factor at the individual level.Especially the (formal) qualification shows significant wrong-signed correlation, a con-tradiction with the widely asserted low productivity in services. On the contrary, firmspecific factors indicating the importance of capital intensity in explaining wage differ-entials (self-employment share) do not seem to have a great explanatory power. Factors(country specific and demand indicators) that are in accordance with non-standardwage determination theories are found significant (unemployment rate, minimum wage,growth structure).
In a second step a multivariate analysis is performed, allowing to control for eacheffect. The signs found in bivariate correlations are so far confirmed for the elder-age-variable (+), gender (+), qualification (counter-intuitive -) and the domestic demandgrowth variable (-).
From these preliminary results, it seems that our hypothesis is confirmed; the struc-ture of the demand addressed to the sectors is quite an important factor explaning thedevelopement of wages and therefore the EU sectoral wage differentials.
26
References
Bassanini, A., and R. Duval (2006): “Employment Patterns inOECD Countries: Reassessing the Role of Policies and Institu-tions,” Social, employment and migration working papers 35, OECD,http://www.oecd.org/document/21/0,3343,fr 2649 33927 35427733 1 1 1 1,00.html.
Calmfors, L., and J. Driffill (1988): “Bargaining Structure, Corporatism andMacroeconomic Performance,” Economic Policy, 6, 13–61.
Card, D. E. (1996): “Deregulation and Labor Earnings in the Airline Industry,” Work-ing Paper 5687, NBER.
Card, D. E., and A. B. Krueger (1994): “Minimum Wages and Employment: ACase Study of the Fast-Food Industry in New Jersey and Pennsylvania,” AmericanEconomic Review, 84(4), 772–93.
Destatis (2006): Was kostet Arbeit in Deutschland? Ergebnisse der Arbeitskostener-hebung 2004. Destatis, Wiesbaden.
Dickens, W. T., and L. F. Katz (1986): “Inter-industry Wage Differences and In-dustry Characteristics,” Working Paper 2014, NBER.
Duthmann, A., P. Hohlfeld, G. Horn, C. Logeay, K. Rietzler, S. Stephan,
and R. Zwiener (2006): “Arbeitskosten in Deutschland bisher uberschatzt – Auswer-tung der neuen Eurostat-Statistik,” IMK Report, (11).
Erdil, E., and I. H. Yetkiner (2001): “A Comparative Analysis of Inter-IndustryWage Differentials: Industrialized versus Developing Countries,” Applied Economics,33(13), 1639–1648.
Freeman, R. B., and R. Schettkat (2001): “Skill Compression, wage differentialsand employment: Germany vs. the US,” Oxford Economic Papers, 53(3), 582–603.
Funk, L., and H. Lesch (2005): “Salaires minimums en Europe,” Comparative Studytn0507102s, EIRO,http://eurofound.europa.eu/eiro/2005/07/study/tn0507102s.html.
Genre, V., D. Momferatou, and G. Mourre (2005): “Wage Diversity In TheEuro Area – An Overview Of Labour Cost Differentials Across Industries,” OccasionalPaper 24, European Central Bank.
27
Gibbons, R., and L. F. Katz (1992): “Does unmeasured ability explain inter-industrywage differentials?,” Review of Economic Studies, 59(3), 515–535.
Haisken-DeNew, J., and C. M. Schmidt (1999): “Industry Wage differentials re-visited: A longitudinal comparison of Germany and USA (1984-1996),” DiscussionPaper 98, IZA.
Horn, G., C. Logeay, S. Stephan, and R. Zwiener (2007): “Auswertung derneuen Eurostat-Arbeitskostenstatistik – Billige Arbeit in Deutschland,” IMK Report,(12).
Husson, M. (2006): “Les salaires minima en Europe,” Chronique internationale del’IRES, numero special, Novembre(103), 17–27.
Krueger, A. B., and L. H. Summers (1988): “Efficiency Wages and the Inter-industry Wage Structure,” Econometrica, 56(2), 259–93.
Layard, R., S. Nickell, and R. Jackman (1991): Macroeconomic Performance andthe Labour Market. Oxford University Press.
Lindbeck, A., and D. Snower (2002): “The Insider-Outsider Theory: A Survey,”Working Paper 534, IZA.
Lindbeck, A., and D. J. Snower (1988): “Cooperation, Harrassment, and Invol-untary Unemployment: An Insider-Outsider Approach,” The American EconomicReview, 78(1), 167–188.
Machin, S., and A. Manning (2002): “The Structure of Wages in What Should be aCompetitive Labour Market,” Discussion Paper 0532, CEP.
Manning, A. (2003): “The real thin theory: monopsony in modern labour markets,”Labour Economics, 10(2), 105–131.
Martins, J. O., S. Scarpetta, and D. Pilat (1996): “Mark-up Pricing, MarketStructure and the Business Cycle,” Economic Studies 27, OECD.
Metcalf, D. (2007): “Why Has the British National Minimum Wage Had Little orNo Impact on Employment?,” Discussion Paper 781, CEP.
OECD (1999): Employment Outlook. OECD, Paris.
Schettkat, R. (2006): “Inequality and Efficiency Does a Wider Spread of WagesRaise Employment?,” Intervention, 3(2), 209–215.
28
Schramm, M. (2004): Intersektorale Lohndifferenzen in Deutschland; Eine Unter-suchung aus theoretischer und empirischer Sicht, no. 101 in VolkswirtschaftlicheForschungsergebnisse. Verlag Dr. Kovac, Hamburg.
Schroder, C. (2007): “Industrielle Arbeitskosten im internationalen Vergleich,” IW-Trends, (4), 1–20.
Shapiro, C., and J. E. Stiglitz (1984): “Equilibrium Unemployment as a WorkerDiscipline Device,” The American Economic Review, 74(3), 433–444.
Sinn, H.-W. (2007): “Internationaler Vergleich der Arbeitskosten: Warum Deutschlandkeine starke Lohnerhohung vertragt,” ifo Schnelldienst, 60(9), 54–59.
Suedekum, J., and U. Blien (2007): “Stimulating Employment Growth with higherwages? A new approach to addressing an old controversy,” Kyklos, 60(3), 441–464.
6 Annex: Overview of the estimators used
6.1 Pooled OLS
In the pooled model, there is no country effects and the time effects are common toall individuals. The errors are supposed to be uncorrelated in the time dimension andacross countries and homoskedastic in the cross-sectional dimension:
yit = α∗ + X ′itβ + uit (10)
In STATA, the command is xtreg with time dummies.
6.2 Static Fixed Effects (SFE)
In the fixed effects model, the error term is decomposed in a fixed unit-specific com-ponent (a country-specific intercept) and an observational error term. There existsseveral methods to estimate such a model; Either one adds individual dummies (LSDV-estimator):
yit = α′DIi + λ′DTt + X ′itβ + uit (11)
29
Or one can transform the variables (subtract the group mean), so that the time-invariantvariables are dropped from the model (WITHIN-estimator):
yit − yi = (αi − αi) + λ′(DTt −DT ) + (Xit −Xi)′β + (uit − ui) (12)
This is achieved in Stata with the command xtreg and the option “fe”. In both cases(that yield exactly the same estimates for the β, as the OLS method is applied, we needto have homoskedastic and uncorrelated errors, and that they are unrelated to the αi
and the other exogenous factors, for the estimators to be consistent and BLUE.
6.3 Static Random Effects (SRE)
In the fixed effects model, the error term is decomposed in a random unit-specific com-ponent (a country-specific error term) and an observational error term. As now it isclear that the overall error term will not met the assumption of autocorrelation andprobably also not of homoskedasticity, the GLS-estimator is required. One can showfurther that the GLS reduce to OLS applied to the following transformed model:
yit − θiyi = (1− θi)α∗ + λ′(DTt − θiDT ) + (Xit − θiXi)′β + εit − θiεit (13)
θi stands for 1−√
σ2u
σ2u+Tiσ2
αand εit = αi + uit.
We need to estimate θ, and there exists therefore different estimators , as we canonly perform F-GLS.
In Stata the commands [xtreg, re] and [xtreg, re sa] call the GLS estimator. In bothcases the Swamy-Arora variance estimator based on the within and between regressionsis implemented. The second specification has a small sample correction, that differ fromthe first one in unbalanced panels. Important to note here, is that the uit-error termpossess the usual properties: homoskedastic across individual, uncorrelated between theindividuals and in the time-dimension. It is also unrelated to the individual-specific errorαi and to the other exogenous variables. As this is a F-GLS estimation, test-distributionshold only asymptotically (χ2 rather than F).
It is possible to estimate the F-GLS also with maximum-likelihood [xtreg, mle]. Inthis case, the residual are assumed also to be normally distributed and an iteratedregression is performed. In case the total number of observation (here 201) is smaller
30
than 300 and the data unbalanced (our case too) the [mle] and [re] regressions will yielddifferent results.
If the correlation structure of uit is not as simple as assumed above esp. if sometime-autocorrelation is present, then the GLS estimator needs to be adapted. There isseveral ways to take account of richer variance structure.
[xtgee] is the first possibility embraced in the estimation part. It fits a population-averaged panel-data model; the option [f(gauss) i(id)] consider that the errors are nor-mally distributed and the model linear in the coefficients. This estimator consider thenthe within-group correlation structure Ri:
corr(ui1, ui1) corr(ui1, ui2) . . . corr(ui1, uiTi)corr(ui2, ui1) corr(ui2, ui2) . . . corr(ui2, uiTi)
......
. . ....
corr(uiTi , ui1) corr(uiTi , ui2) . . . corr(uiTi , uiTi)
=
r11 r12 . . . r1Ti
r21 r22 . . . r2Ti
......
. . ....
rTi1 rTi2 . . . rTiTi
Then the within-group correlation structure Ri can take two forms:
• corr(exch): rts =
1 if t = s
ρ if t 6= s. This equivalent to the [xtreg, re] and [xtreg, mle]
commands if the data are balanced, with [xtreg, pa] otherwise.
• corr(ar1):rts =
1 if t = s
ρ|t−s| if t 6= s. This equivalent to the [xtreg, re] command.
[xtgls] is another possibility. It is a F-GLS estimator which allows for ar(1)-autocorrelationwithin panel, as [xtgee], and cross-sectional correlation and cross-sectional heteroskedas-ticity. The variance of the overall residuals (αi + uit) cam be written as the Kroneckerproduct of the within-group variance (σ) and the cross-sectional variance (Ω):
E(εε′) = σ ⊗ Ω =
σ11 σ12 . . . σ1N
σ21 σ22 . . . σ12N
......
. . ....
σN1 σN2 . . . σNN
⊗
Ω11 Ω12 . . . Ω1N
Ω21 Ω22 . . . Ω12N
......
. . ....
ΩN1 ΩN2 . . . ΩNN
If the diagonal of the σ-matrix is filled with different numbers, the individuals do not
31
have the same overall variance (panel-heteroskedasticity). If the off-diagonals are notzero, the individuals are correlated with each other (panel-correlation). If the Ω-matrixis not identity, then the residuals are autocorrelated in the time dimension (within-groupcorrelation). With this estimator several options are possible:
• panels(hetero): state that the Ω-matrix is identity (default if no corr-option spec-
ified) and σ =
σ11 0 . . . 00 σ22 . . . 0...
.... . .
...0 0 . . . σNN
• corr(ar1): state that the Ω-matrix is such that the serial correlation structure is
rts =
1 if t = s
ρ|t−s| if t 6= s
• corr(psar1): state that this correlation factor can be different for each individual:
rts,i =
1 if t = s
ρ|t−s|i if t 6= s
The last estimator used in this paper is [xtpcse]; it calculates a panel-correctedstandard errors for the OLS-estimates (actually Prais-estimator that corrects for firstorder autoregression in the time-dimension, i.e. a GLS estimator!). We use the [pairwise]option that specifies that all information available for an individual should be used tocalculate the covariance matrix. This has influence only when the data are unbalanced.The option [hetonly] is also selected implying that panel-heteroskedasticity is allowed(see above). Two sorts of within-group autocorrelation is then allowed as above, acommon ar(1)-structure (ar1) or individual-specific one (psar1). [xtgls] and [xtpcse]are consistent, and under the correct assumption of the error structure F-GLS wouldbe more efficient. But it is argued that with the typical small samples used in socialscience, xtpcse should be preferred, as it is more conservative.
7 Annex: Scatter diagrams of the data
32
Chart 3: Scatter diagram: Age (15-24) vs Wages, EU-27 and EU-15
at
atat
atatat
atatat
be
bebe
be
bebebe
be
be
be
dededede dedede
dededees
eses
es
eseseseses
es
fi
fifi
fi fifi
fifi
fifi
frfrfrfr
frfr
fr
frfrfr
grgrgrgrgrgrgrgrgr
ititit
it
itititit
itit
lu
lu
lu
lu
lu
lu
lulu
lu
lu
nl
nlnlnl
nlnlnlnlnlnl
pt
ptpt
pt
pt
pt
ptpt
pt
pt
dk
dk
dk
dkdk
dkdk
dkdk
se
sesese
se
sesese
ukukukuk
ukukukukuk uk
cy
cy
cycy
cy
cyczcz
czczczczczcz
czee
ee
ee
eeee
ee
eeee ee
huhu
huhu
huhu
huhuhu
hultlt
ltlt
lt
ltlt
lt
lv lvlv
lv
lvlvlvlv
mt
mtmt
mt
plplplpl
plpl
si
si
sisisi
si
sisi
sisi
sksk
sk
sk
sksksk
sk
bg
bgbgbg
bgbg
.2.4
.6.8
11.
2al
ter_
1524
: rel
ativ
e pr
opor
tion
of y
oung
wor
kers
(in
dust
ry/s
ervi
ces)
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
at
atat
atatat
atatat
be
bebe
be
bebebe
be
be
be
dededede dedede
dededees
eses
es
eseseseses
es
fi
fifi
fi fifi
fifi
fifi
frfrfrfr
frfr
fr
frfrfr
grgrgrgrgrgrgrgrgr
ititit
it
itititit
itit
lu
lu
lu
lu
lu
lu
lulu
lu
lu
nl
nlnlnl
nlnlnlnlnlnl
pt
ptpt
pt
pt
pt
ptpt
pt
pt
dk
dk
dk
dkdk
dkdk
dkdk
se
sesese
se
sesese
.2.4
.6.8
11.
2al
ter_
1524
: rel
ativ
e pr
opor
tion
of y
oung
wor
kers
(in
dust
ry/s
ervi
ces)
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
Chart 4: Scatter diagram: Age (50+) vs Wages, EU-27 and EU-15
at atatatatat
atatat
bebebe
bebebebebebebe
dededede dededededede
eseseseseseseseseses
fifififi fi fi fififififr
frfrfr
frfrfrfrfrfr
grgrgrgr
grgrgrgrgr
it ititititititit itit
lulu
lulu
lu
lu
lu
lu
lulu
nlnlnlnl
nlnlnlnlnlnl
ptptptpt
pt
ptpt
pt
ptpt
dkdkdkdk
dk dkdkdkdk
sesesesese
seseseuk
ukukukukukukukuk uk
cy
cy
cy cy
cy
cy
czczcz
czczczczczcz
eeeeeeee
ee
ee
eeee
eehuhu
huhu huhu
huhuhuhu
ltlt
lt
lt
ltlt
ltlt
lv
lv
lvlvlv
lv
lv
lv
mt
mt
mt
mt
plplplplpl
plsi si
si
sisisi
sisi
sisi
sksksk
sk
sk sksk
sk bgbgbgbgbgbg
.51
1.5
22.
5al
ter_
50pl
us: r
elat
ive
prop
ortio
n of
eld
er w
orke
rs (
indu
stry
/ser
vice
s)
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
at atatatatat
atatat
bebebe
bebebebebebebe
dededede dededededede
eseseseseseseseseses
fifififi fi fi fififififr
frfrfr
frfrfrfrfrfr
grgrgrgr
grgrgrgrgr
it ititititititit itit
lulu
lulu
lu
lu
lu
lu
lulu
nlnlnlnl
nlnlnlnlnlnl
ptptptpt
pt
ptpt
pt
ptpt
dkdkdkdk
dk dkdkdkdk
sesesesese
sesese
.51
1.5
22.
5al
ter_
50pl
us: r
elat
ive
prop
ortio
n of
eld
er w
orke
rs (
indu
stry
/ser
vice
s)
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
Chart 5: Scatter diagram: Gender vs Wages, EU-27 and EU-15
at atatatatat
at at
at
bebebe
bebebebe
bebebe
dededede dedededede
de
eseseseses
eseseseses
fi
fi
fifi fi fi
fifi
fifi
frfrfrfrfrfrfr
frfrfr
grgrgr
grgrgr
grgrgr
it itititit
ititit
itit
lululu
lu
lulululu
lu
lunlnl
nlnlnlnlnlnlnlnl
ptptptpt ptptptptpt
pt
dkdkdkdk
dkdkdkdkdk
sese
sesesesesese
ukukukukukukukukuk uk
cy
cy
cycy
cy
cy
czcz
czczczczczczcz
ee
eeee
eeeeee
eeee ee huhuhuhu
huhu
huhuhu
hu
lt
ltlt
lt
lt
lt
ltlt
lvlv
lvlv
lv
lv
lv
lv
mt
mtmtmt
plplplplplpl
si sisisi
sisi sisi
sisisk
sksksk
sk sksk
sk
bg
bgbgbg
bgbg
.81
1.2
1.4
1.6
gend
er: r
elat
ive
prop
ortio
n of
men
(in
dust
ry/s
ervi
ces)
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
at atatatatat
at at
at
bebebe
bebebebe
bebebe
dededede dedededede
de
eseseseses
eseseseses
fi
fi
fifi fi fi
fifi
fifi
frfrfrfrfrfrfr
frfrfr
grgrgr
grgrgr
grgrgr
it itititit
ititit
itit
lululu
lu
lulululu
lu
lunlnl
nlnlnlnlnlnlnlnl
ptptptpt ptptptptpt
pt
dkdkdkdk
dkdkdkdkdk
sese
sesesesesese
.81
1.2
1.4
1.6
gend
er: r
elat
ive
prop
ortio
n of
men
(in
dust
ry/s
ervi
ces)
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
33
Chart 6: Scatter diagram: Qualification vs Wages, EU-27 and EU-15
at atatatatatat atat
bebebebebebe
bebebebe
dedede dededededede
eseseseseseseseseses
fififi
fi fi fi fifififi
frfrfrfrfrfrfrfrfrfr
grgrgrgrgrgrgrgrgr
it ititititititititit
lulu
lulu
lu
lulu lulu
nlnlnlnlnlnlnlnlnlnl
ptptptpt ptptptptptptdk
dk
dk
dk
dk
dkdkdkdk
sesesesesesesese
ukukukukukukukuk uk
cycycy
cy
cycy
cz
czczcz
czcz
czcz
ee
ee
ee
ee
ee
ee
ee
ee
eehu
hu
huhuhu
huhuhuhu
lt
lt
lt
lt
lt
lt
lt
lt
lv
lvlv
lvlv
lv
lvlv
mtmtmtmt
pl
pl
si
sisi
si
si
si
si
si
sisi
sk
sk
sk
sk
sksksksk
bgbg
bg
bg
bgbg
11.
52
2.5
33.
5qu
alifi
catio
n: r
elat
ive
qual
ifica
tion
aver
age
(indu
stry
/ser
vice
s)
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
at atatat
at
atat
atat
be
bebe
be
bebe
be
be
be
be
dedede de
dedede
de
de
eseses
eses
eses
eses
es
fi
fi
fi
fi fi
fififi
fifi
frfrfr
frfrfrfrfrfrfr
grgrgrgr
gr
gr
gr
grgr
ititit
ititititit it
it
lu
lu
lulu
lu
lu
lu
lu
lu
nlnl
nl
nl
nlnl
nlnl
nlnl
pt
pt
ptpt ptptpt
ptptpt
dk
dk
dk
dk
dk
dk
dkdkdk
sesesese
sese
se
se
11.
21.
41.
6qu
alifi
catio
n: r
elat
ive
qual
ifica
tion
aver
age
(indu
stry
/ser
vice
s)
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
34
Chart 7: Scatter diagram: Firm size vs Wages, EU-27 and EU-15
atatat
bebebede dede
esesesfi fi fi
frfrfrititit
lululu
nlnlnl
pt ptptdkdkdk
sesese
ukukuk
cycycy
czczczeeeeee
hu huhu
ltltltlvlv lv
plplplsisisi
sk sksk
bg bgbg
02
46
8fir
msi
ze: r
elat
ive
firm
siz
e (in
dust
ry/s
ervi
ces)
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
atatat
bebebe
de dede
eseses
fi fi fi
frfrfr
ititit
lululu
nlnlnl
pt ptpt
dkdkdk
sesese
23
45
firm
size
: rel
ativ
e fir
m s
ize
(indu
stry
/ser
vice
s)
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
Chart 8: Scatter diagram: Self-employment vs Wages, EU-27 and EU-15
at
atat
atatatat at
at
bebebebe
bebebebe
be
bedededede
dededededede
eseseseseseses
eses
esfi
fififi fi fi
fi
fi
fi
fifrfrfrfrfr
frfrfrfrfr
grgrgrgrgrgrgrgr
gr
it itititit
ititititit
lu
lululu
lu
lu
lulu
lu
lu
nl
nlnlnl
nl
nlnl
nlnlnlptptptpt ptptptpt
pt
pt
dkdk
dkdk
dk
dkdkdk
dk sese
sesesese
sese ukukukuk
ukukukuk
uk uk
cy
cycy
cycycy
czczcz
czczcz
czcz
cz
ee
ee
ee
ee
huhuhu
hu
huhuhuhu
huhult
lt
ltltlt
lv lvlv
lv
lv
lvlv
lv
mt
mtmt
mt
plplplplplpl
sisisi
sisisi si
si
sisi
sksk sk
sk
sk
sksksk
bgbgbg
bgbgbg
0.2
.4.6
.8se
lfem
ploy
: rel
ativ
e pr
opor
tion
of s
elf−
empl
oyed
(in
dust
ry/s
ervi
ces)
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
at
at
at
at
atatat at
at
bebebebe
bebebe
be
be
be
dedede
dededededede
de
eseseseseseses
eses
es
fi
fi
fififi
fi
fi
fi
fi
fifrfrfr
frfr
frfrfrfrfr
grgrgr
grgrgrgrgr
gr
ititit ititititit
itit
lu
lu
lu
lu
lu
lu
lulu
lu
lu
nl
nlnl
nlnl
nlnl
nlnlnlptptptpt pt
ptptpt
pt
pt
dk
dk
dk
dk
dk
dkdk
dk
dk sese
se
sesese
sese
.1.2
.3.4
.5.6
selfe
mpl
oy: r
elat
ive
prop
ortio
n of
sel
f−em
ploy
ed (
indu
stry
/ser
vice
s)
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
Chart 9: Scatter diagram: Part-time employment vs Wages, EU-27 and EU-15
at
be
de
es
fi
fr
gr
itlu
nl
pt
dk
se
uk
cy
cz ee
hu
lt
lv
si
sk
bg
.1.2
.3.4
.5.6
teilz
eit:
rela
tive
prop
ortio
n of
par
t−tim
e w
orke
rs (
indu
stry
/ser
vice
s)
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
at
be
de
es
fi
fr
gr
itlu
nl
pt
dk
se
.1.2
.3.4
.5.6
teilz
eit:
rela
tive
prop
ortio
n of
par
t−tim
e w
orke
rs (
indu
stry
/ser
vice
s)
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
35
Chart 10: Scatter diagram: Unemployment rate vs Wages, EU-27 and EU-15
at atatatatat
at atat
bebebebe
bebebe
bebebe dededede
dededededede
es
es
es
es
eses
eseses
es
fi
fi
fi
fi fifi fifififi
frfrfrfr
frfr
frfrfrfrgrgrgr
gr
grgrgrgr
gr
it itit ititititit itit
lululululululu
lu
lulu
nl
nl
nlnlnl
nlnl
nlnlnl
ptpt
ptpt
ptptpt
ptptpt
dkdkdkdkdk dk
dkdkdk
sese
se
se
sesesese
uk
ukukuk
ukukukukuk ukcy
cy cycy
cy
cz
czczcz
czcz
czcz
eeee
ee
eeee
eeee ee
ee
huhu
hu
huhu
huhuhuhu
hu
ltlt
ltlt
lt
ltlt
lt
lv lvlvlv
lv
lvlv
lv
mt mtmtmt
plpl
pl
pl
pl
plplpl
pl
si sisisi
sisi sisi sisi
sk
sk
sksk
sk
sksk
sk bg
bg
bg
bg
bg
bg
05
1015
20al
q: n
atio
nal u
nem
ploy
men
t rat
e
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
at atatatatat
at atat
bebebebe
bebebe
bebebe dededede
dededededede
es
es
es
es
eses
eseses
es
fi
fi
fi
fi fifi fifififi
frfrfrfr
frfr
frfrfrfrgrgrgr
gr
grgrgrgr
gr
it itit ititititit itit
lululululululu
lu
lulu
nl
nl
nlnlnl
nlnl
nlnlnl
ptpt
ptpt
ptptpt
ptptpt
dkdkdkdkdk dk
dkdkdk
sese
se
se
sesesese
05
1015
20al
q: n
atio
nal u
nem
ploy
men
t rat
e
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
Chart 11: Scatter diagram: Minimum wage-dummy vs Wages, EU-27 and EU-15
at atatatatatat atat
bebebebebebebebebebe
dededede dededededede
eseseseseseseseseses
fifififi fi fi fifififi
frfrfrfrfrfrfrfrfrfrgrgrgrgrgrgrgrgrgr
it itit ititititit itit
lulululululululu lulu nlnlnlnlnlnlnlnlnlnlptptptpt ptptptptptpt
dkdkdkdkdk dkdkdkdk sesesesesesesese ukukuk
ukukukukukuk uk
cycycycycycy cycycy
czczczczczczczczczcz eeeeeeeeeeeeeeee ee ee huhuhuhu huhuhuhuhuhu ltltltltltltltltltlt lvlv lvlv lvlvlvlvlvmt mtmtmt plplplplplplplplplplsisi sisisisisi sisi sisi sksksksk sksk sk sksksk bgbg bgbgbgbgbgbg
0.2
.4.6
.81
min
wag
e_is
: dum
my
for
exis
tenc
e of
nat
iona
l min
imum
wag
e
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
at atatatatatat atat
bebebebebebebebebebe
dededede dededededede
eseseseseseseseseses
fifififi fi fi fifififi
frfrfrfrfrfrfrfrfrfrgrgrgrgrgrgrgrgrgr
it itit ititititit itit
lulululululululu lulu nlnlnlnlnlnlnlnlnlnlptptptpt ptptptptptpt
dkdkdkdkdk dkdkdkdk sesesesesesesese0.2
.4.6
.81
min
wag
e_is
: dum
my
for
exis
tenc
e of
nat
iona
l min
imum
wag
e
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
Chart 12: Scatter diagram: Union density vs Wages, EU-27 and EU-15
at atatatatatat at
bebebebebebebebe
dededede dededede
eseseseseseseses
fifififi fi fi fifi
frfrfrfrfrfrfrfr
it itit ititititit
nlnlnlnlnlnlnlnlptptptpt ptptptpt
dkdkdkdkdk dkdk
sesesesesesesese
ukukukukukukukuk
020
4060
80ud
ens:
uni
on d
ensi
ty (
%)
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
at atatatatatat at
bebebebebebebebe
dededede dededede
eseseseseseseses
fifififi fi fi fifi
frfrfrfrfrfrfrfr
it itit ititititit
nlnlnlnlnlnlnlnlptptptpt ptptptpt
dkdkdkdkdk dkdk
sesesesesesesese
020
4060
80ud
ens:
uni
on d
ensi
ty (
%)
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
36
Chart 13: Scatter diagram: Relative export growth vs Wages, EU-27 and EU-15
at atatatatatat atat
bebebebe
bebebebebebe
dededede dededededede
eseseseseseseseseses
fififi
fifi
fi fififi
fi
frfrfrfrfrfrfrfrfrfr
grgr
grgr
grgrgr
grgrit
itit
ititititit itit
lululu
lululululu
lulu
nlnlnlnlnlnlnlnlnlnlptptptpt
ptptptptptptdkdk
dkdkdk dkdkdkdk
sesesesesesesese ukuk
ukukukukukukukuk
cycycycycycycycycy
czczczczczczcz
cz
cz
cz
ee
eeee
ee
eeeeee
ee
ee
ee
hu
hu
hu
hu
huhu
hu
hu
hu
hu
lt
ltlt
ltlt
lt
ltltlt
lt
lvlv
lvlv lvlv
lvlv
lv
mt mtmtmt
plplpl
pl
plplpl
pl
plpl
sisi sisisisisi
sisisisi
sksk
sksk
sksksk
sksk
sk
bg
bgbg
bgbg
bgbg
bg
5010
015
020
0gr
owth
_exp
k2: r
elat
ive
cum
ulat
ed e
xpor
t gro
wth
(co
untr
y/eu
ro−
zone
)
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
at atatatatat
at at
at
be
bebebe
be
bebebebe
be
dededededededededede
es
eseseseseses
eseses
fi
fifi
fi
fifi
fi
fifi
fi
frfrfrfrfr
frfrfr
frfr
grgr
grgr
grgr
gr
grgr
it
it
it
itit
it
itit
itit
lulu
lu
lululululu
lu
lu
nlnlnlnlnl
nlnlnlnlnlpt
ptptpt
ptptpt
ptpt
pt
dkdk
dkdkdk
dkdk
dk
dk
sesese
sese
sese
se
6080
100
120
grow
th_e
xpk2
: rel
ativ
e cu
mul
ated
exp
ort g
row
th (
coun
try/
euro
−zo
ne)
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
Chart 14: Scatter diagram: Relative GDP growth vs Wages, EU-27 and EU-15
at atatatatatat atat bebebebebe
bebebebebe dededede dededededede
eseseseseses
eseses
es
fifififi fi fi fifi
fifi
frfrfrfrfrfrfrfrfrfrgrgrgrgrgrgrgrgr
gr
it itit ititititit itit
lulu
lu
lu
lulululu
lulu
nlnlnl
nlnlnlnlnlnlnlpt
ptptpt pt
ptptptptpt dkdkdkdk
dk dkdkdkdk sesesesesesesese ukukukukukukuk
ukuk uk
cycycy
cycycy cycycy
czcz
czczczczczcz
cz
cz ee
eeee
ee
ee
ee
ee
ee
ee
ee
huhu
huhu
huhuhuhuhu
hu
lt
lt
lt
ltltlt
lt
lt
lt
lt
lvlv lv
lv
lv
lv
lv
lv
lv
mtmtmtmt
pl
plpl
plplplplplplpl
sisi
sisisisisi
sisi
sisi
sksksk
sksksk
sksksk
sk
bgbg bgbgbg
bgbg
bg
8010
012
014
016
0gr
owth
_gdp
k2: r
elat
ive
cum
ulat
ed g
dp g
row
th (
coun
try/
euro
−zo
ne)
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
at atatatat
atat atat bebebebebe
bebebebebe
dededede de
dedededede
eseseseseses
es
eses
es
fi
fi
fifi
fifi
fifi
fifi
frfrfrfrfr
frfrfrfrfrgr
grgrgrgrgr
gr
gr
gr
it itit
ititititit
itit
lu
lu
lu
lu
lulu
lululu
lu
nlnl
nlnlnl
nlnlnlnlnl
ptptptpt pt
ptptptpt
pt dkdkdkdk
dk dkdkdkdk
sesese
sese
sese
se
9010
011
012
013
0gr
owth
_gdp
k2: r
elat
ive
cum
ulat
ed g
dp g
row
th (
coun
try/
euro
−zo
ne)
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
Chart 15: Scatter diagram: Relative internal demand growth vs Wages, EU-27 andEU-15
at atatatatatat atat
bebebebebebebebebebe dededede de
dedededede
eseseseseses
eseseses
fifififi fi fi fi
fififi
frfrfrfrfrfrfrfrfrfrgrgrgrgrgrgrgrgr
gr
it itit ititititit itit
lulu
lulululu
lulu lu
lu
nlnlnlnlnlnlnlnlnlnlpt
ptptpt ptptptptptpt dkdk
dkdkdk dkdkdk
dk sesesesesesesese
ukukukukukukukukuk uk
cycycycycycy
cycycy
czcz
czczczczczczczcz ee
eeee
ee
ee
ee
ee
ee
ee
ee
huhu
huhuhuhu
hu
huhuhu
lt
ltltltlt
lt
lt
lt
lt
lt
lv
lv lvlv
lv
lv
lv
lv
lv
mtmtmt
mt
pl
plplplpl
plplplplpl
sisi
sisisi
sisi sisi
sisi sksksk
sksk
sksk
sksk
sk
bg
bgbg
bgbg
bg
bg
bg
8010
012
014
016
018
0gr
owth
_int
ddk2
: rel
ativ
e cu
mul
ated
inte
rnal
dem
and
grow
th (
coun
try/
euro
−zo
ne)
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
atat
atatat
atat
atat
bebebebebe
bebebebebe
dededede
de
dedededede
eses
es
eseses
es
es
es
es
fi
fifi
fi fifi
fi
fifi
fi
frfrfrfr
frfrfrfrfrfr
grgrgrgrgr
grgr
gr
gr
ititit ititititit it
it
lu
lu
lu
lulu
lu
lu
lulu
lu
nl
nlnlnlnl
nlnlnl
nlnlpt
pt
pt
pt ptptpt
ptptptdkdk
dkdk
dkdkdkdk
dksesesesese
sesese
9010
011
012
013
0gr
owth
_int
ddk2
: rel
ativ
e cu
mul
ated
inte
rnal
dem
and
grow
th (
coun
try/
euro
−zo
ne)
.6 .8 1 1.2 1.4lk_neu: relative hourly wages (industry / services)
37