Labour Economics: Sonderausgabe Heft 2+3/Bd. 229 (2009) Jahrbücher für Nationalökonomie und Statistik 9783110508284, 9783828204782

Citation preview

Labour Economics

Herausgegeben von

Bernd Fitzenberger, Werner Smolny und Peter Winker

Antonczyk, Dirk, Freiburg Mit Beiträgen Bertschek, Irene, von Mannheim Bischoff, Ivo, Gießen Börsch-Supan, Axel, Mannheim Buettner, Thiess, München Ebertz, Alexander, München Entorf, Horst, Frankfurt a . M . Fitzenberger, Bernd, Freiburg Gerlach, Knut, Hannover Guertzgen, Nicole, Mannheim Hassler, Uwe, Frankfurt a . M . Heinemann, Friedrich, Mannheim

Lucius & Lucius • Stuttgart 2 0 0 9

Hennighausen, Tanja, Mannheim Hübler, Olaf, Hannover Leuschner, Ute, Freiburg Löschel, Andreas, Mannheim Ludwig, Alexander, Mannheim Meyer, Jenny, Mannheim Möller, J o a c h i m , Regensburg Oberndorfer, Ulrich, Berlin Smolny, Werner, Ulm Tubadji, Annie, Nürnberg Wolters, Jürgen, Berlin Z i m m e r m a n n , Volker, Frankfurt a . M .

Anschriften der Herausgeber des Themenheftes Prof. Bernd Fitzenberger, Ph. D. Abteilung für Empirische Wirtschaftsforschung und Ökonometrie Albert-Ludwigs-Universität Freiburg Platz der Alten Synagoge 79085 Freiburg E-Mail: [email protected] Prof. Dr. Werner Smolny Fakultät für Mathematik und Wirtschaftswissenschaften Institut für Wirtschaftspolitik Universität Ulm 89069 Ulm E-Mail: Werner.Smolny@uni-ulm Prof. Dr. Peter Winker Justus-Liebig-Universität Gießen Licher Straße 64 35394 Gießen E-Mail: [email protected]

Bibliografische Information der Deutschen Nationalbibliothek Die Deutsche Nationalbibliothek verzeichnet diese Publikation in der Deutschen Nationalbibliografie; detaillierte bibliografische Daten sind im Internet über http://dnb.d-nb.de abrufbar ISBN 978-3-8282-0478-2

© Lucius & Lucius Verlagsgesellschaft mbH • Stuttgart • 2009 Gerokstraße 51, D-79184 Stuttgart Das Werk einschließlich aller seiner Teile ist urheberrechtlich geschützt. Jede Verwertung außerhalb der engen Grenzen des Urheberrechtsgesetzes ist ohne Zustimmung des Verlags unzulässig und strafbar. Das gilt insbesondere für Vervielfältigungen, Ubersetzungen und Mikroverfilmungen sowie die Einspeicherung und Verarbeitung in elektronischen Systemen.

Satz: Mitterweger & Partner Kommunikationsgesellschaft mbH, Plankstadt Druck und Bindung: Neumann Druck, Heidelberg Printed in Germany

Jahrbücher f. Nationalökonomie u. Statistik (Lucius & Lucius, Stuttgart 2009) Bd. (Vol.) 229/2+3

Inhalt / Contents Abhandlungen / Original Papers Hassler, Uwe, Jürgen Wolters, Hysteresis in Unemployment Rates? A Comparison between Germany and the US Smolny, Werner, Wage Adjustment, Competitiveness and Unemployment East Germany after Unification Löschet, Andreas, Ulrich Oberndorfer, Oil and Unemployment in Germany Börsch-Supan, Axel, Alexander Ludwig, Living Standards in an Aging Germany: The Benefits of Reforms and the Costs of Resistance Heinemann, Friedrich, Ivo Bischoff, Tanja Hennighausen, Choosing from the Reform Menu Card - Individual Determinants of Labour Market Policy Preferences Gerlach, Knut, Olaf Hübler, Employment Adjustments on the Internal and External Labour Market - An Empirical Study with Personnel Records of a German Company Antonczyk, Dirk, Bernd Fitzenberger, Ute Leuschner, Can a Task-Based Approach Explain the Recent Changes in the German Wage Structure?. . . Guertzgen, Nicole, Firm Heterogeneity and Wages under Different Bargaining Regimes: Does a Centralised Union Care for Low-Productivity Firms? Entorf, Horst, Crime and the Labour Market: Evidence from a Survey of Inmates Möller, Joachim, Annie Tubadji, The Creative Class, Bohemians and Local Labor Market Performance Buettner, Thiess, Alexander Ebertz, Spatial Implications of Minimum Wages Zimmermann, Volker, The Impact of Innovation on Employment in Small and Medium Enterprises with Different Growth Rates Bertschek, Irene, Jenny Meyer, Do Older Workers Lower IT-Enabled Productivity? Firm-Level Evidence from Germany

119-129 130-145 146-162 163-179 180-197 198-213 214-238 239-253 254-269 270-291 292-312 313-326 327-342

Buchbesprechungen / Book Reviews Allen, Franklin, Douglas Gale, Financial Crises (International Library of Critical Writings in Economics) Allen, Franklin, Douglas Gale, Understanding Financial Crises Brandstätter, Jana, International divergierende demographische Entwicklungen und internationale Kapitalbewegungen Eekhoff, Johann, Vera Bünnagel, Susanna Kochskämper, Kai Menzel, Bürgerprivatversicherung: Ein neuer Weg für das Gesundheitswesen Kimmel, Christoph, Vermögenspreisinflation als wirtschaftspolitische Herausforderung Konrad, Kai A., Beate Jochimsen (Hrsg.), Finanzkrise im Bundesstaat Mayert, Andreas, Alterssicherung und Erziehungsentscheidungen Pickhardt, Michael, Edward Shinnick (eds.), The Shadow Economy, Corruption and Governance Sloan, Frank A., Hirschel Kasper (Hrsg.), Incentives and Choice in Health Care Walther, Steffen, Berechnungsmethoden des Produktionspotenzials

343 343 345 347 350 351 353 356 357 358

Jahrbücher f. Nationalökonomie u. Statistik (Lucius & Lucius, Stuttgart 2009) Bd. (Vol.) 229/2+3

Special Issue on Labour Economics: Guest Editorial Labour economics has emerged to be one of the leading fields of research in applied economics with important spillover effects in relation to other subdisciplines both within and outside of economics. Labour market problems are core issues in the policy debate. Modern labour economics is based on microeconomic and/or macroeconomic theory and is a distinctively empirical discipline. Research questions in labor economics have prompted the development both of new econometric techniques and of new theories. The growing availability of large scale labour market data sets at the individual level and the growth in laboratory experiments have been further stimulants. Research in labour economics is at the forefront of applied econometrics and results of labour economic research often shape policy analysis. In comparison to labour economic research in the US or the UK, labour economic research in Germany had been lacking behind for some time. But by now, this has changed a lot and this issue of the Jahrbücher für Nationalökonomie und Statistik provides evidence in accordance with this hypothesis (see also the special issue of the Jahrbücher 2008/5+6 on "Labormetrics"). The 13 contributions in this issue involve empirical research on various topics in labour economics and related areas. The Jahrbücher für Nationalökonomie und Statistik have a long lasting tradition in publishing such papers. The topics covered in this issue involve (i) macroeconomic analyses of the labour market, (ii) reform of the welfare state institutions, (iii) microeconometric analysis of wages and employment, (iv) regional labor markets, and (v) labor market effects of information technologies. (i) The paper by Uwe Hassler and Jürgen Wolters uses state-of-the-art methods in time series econometrics to analyze empirically the nature of persistence in the unemployment rates in Germany and the US. Werner Smolny analyzes the different labor market situation in East Germany and West Germany since German reunification. Andreas Löschet and Ulrich Oberndorfer readdress the pertinant question of the relationship between oil prices and unemployment in Germany, (ii) In light of the dramatic demographic change in Germany, Axel Börsch-Supan and Alexander Ludwig investigate the potential benefits of pension and labor market reform for growth and living standards, taking into account behavioral reactions to specific reforms. Friedrich Heinemann, Ivo Bischoff, and Tanja Hennighausen analyze the individual determinants of preferences for labour market policies using survey data on attitudes, (iii) Using personnel records of a German company, Knut Gerlach and Olaf Hübler scrutinize empirically employment adjustments on the internal and external labour market. Dirk Antonczyk, Bernd Fitzenberger, and Ute Leuschner describe the recent increase in wage inequality in Germany and investigate as to whether a task-based approach can explain these changes. Nicole Gürtzgen studies empirically the relationship between wages and the degree of firm heterogeneity under centralized and decentralized wage bargaining. Horst Entorf draws on a survey of prison inmates and provides microeconomic evidence on the relationship between individual anticipated labour market opportunities and the perceived probability of future recidivism. (iv) Using regional panel data for West Germany, Joachim Möller and Annie Tubadji test Florida's concept of the creative class. The study investigates whether the local concentration of the Creative Class fosters the economic development of a region and whether the creative workers flock where the Bohemians are. Thiess Buettner and Alexander Ebertz analyze empirically the spatial implications of the introduction of a minimum wage in Germany in light of the sizeable regional differences in wage levels, (v) Using firm level data, Volker Zimmermann estimates the possibly heterogeneous impact

118 • Guest Editorial

of innovation on employment in small and medium size enterprises which differ in their growth rates. Irene Bertschek and Jenny Meyer investigate empirically the question whether firms' IT-enabled labour productivity is affected by the age structure of the workforce. The authors dedicate the peer-reviewed contributions to this issue to their colleague, teacher, mentor, and friend Wolfgang Franz on the occasion of his 65th birthday on January 7, 2009. Among other research output, Wolfgang Franz has contributed to the growing body of state-of-the art research in labour economics in Germany and he is the author of the most successful textbook on labor economics in Germany ("Arbeitsmarktokonomik", Springer Publisher) which will soon appear in its seventh edition. Wolfgang Franz strongly believes that academic research in economics should be firmly based in economic theory and it should use appropriate state-of-the-art econometric - or other empirical methods. At the same time, applied economic research should be relevant, focussing on its implications for our understanding of real world problems. This is the basis for sound economic policy advice. In this vein, we hope that this issue provides sound research in labor economics which Wolfgang Franz - and all readers - find relevant. We thank the Centre for European Economic Research (ZEW) Mannheim - and in particular Thomas Kohl - for the support in producing this issue and for organizing the research workshop on Labour Economics on 6 February 2009, where some of the contributions to the issue were presented and discussed. Bernd Fitzenberger, Werner Smolny, Peter Winker

Jahrbücher f. Nationalökonomie u. Statistik (Lucius & Lucius, Stuttgart 2009) Bd. (Vol.) 229/2+3

Hysteresis in Unemployment Rates? A Comparison between Germany and the US By Uwe Hassler, Frankfurt a.M., and Jürgen Wolters, Berlin* JEL C14, C22, E24

Unemployment, hysteresis, fractional integration, change in persistence.

Summary In this paper we compare the unemployment dynamics of the US and Germany with monthly data up to 2008. With data from 1971 on the evidence is mixed when applying descriptive methods or formal unit root tests. When allowing for fractional integration, however, we find similar results to the literature in that shocks to US data seem to be transitory while having permanent effects on German unemployment. This difference in hysteresis, however, depends on the sample. Using recursive and rolling techniques we observe that shocks in US subsamples until the mid-nineties are clearly more transitory than in more recent subsamples. We conclude that hysteresis has turned into a dominating feature also on US labour market more recently.

1

Introduction

Hysteresis is a widely discussed topic in labour economics. Among others Franz (1987, 1990) has substantially contributed to this discussion. The importance especially for labour market policy stems from the fact, that the hysteresis phenomenon indicates a situation where the natural rate of unemployment is no longer constant but path dependent, i.e. lagged unemployment rates influence the current steady state. Such a situation challenges the natural rate hypothesis (Friedman 1968, Phelps 1967) which assumes stationary fluctuations of unemployment around a constant unemployment rate. Blanchard and Summers (1986) with their insider-outsider model have given one theoretical explanation for the hysteresis phenomenon. For further theories see the surveys by Bean (1994), Franz (1990) and R 0 e d (1997). Looking at Figure 1 we find that since the 1970's unemployment in Germany has more or less increased in three steps. The first two steps can be dated to the oil price shocks whereas the third one appears after the German unification. The unemployment rate hit its peak at about 12 percent in late 2005. From there we observe a d o w n w a r d movement until now, which partly can be attributed to the labour market reform (Hartz). In contrast to this picture the unemployment rate in the US also increased as a result of the oil price shocks but always returned to the former levels. Since the mid 1990s the unemployment rate stays between 4 and 6 percent. These facts have often been taken as evidence that labour markets in the US are more flexible than in Germany, since shocks

* H e l p f u l c o m m e n t s by the editor Peter W i n k e r and t w o a n o n y m o u s referees are gratefully a c k n o w ledged.

120 • U. Hassler and J. Wolters

Germany

US

Figure 1 Seasonally adjusted unemployment rates seem to have only transitory effects w h e r e a s they are considered to have p e r m a n e n t effects in G e r m a n y . T h e hysteresis p h e n o m e n o n is defined in the literature a s the feature that s h o c k s have p e r m a n e n t or a t least long-lasting effects. T h i s implies that in an a u t o r e g r e s s i v e representation for the u n e m p l o y m e n t rate the coefficients of the l a g p o l y n o m i a l s u m u p to one, such that in c a s e of hysteresis the d a t a generating p r o c e s s of the u n e m p l o y m e n t rate h a s a unit root. T h i s is a very special c a s e a n d therefore B l a n c h a r d a n d S u m m e r s ( 1 9 8 6 ) have eased this restriction to the c a s e where the sum of the coefficients in the autoregressive representation is close to one. N e v e r t h e l e s s a lot of empirical p a p e r s testing for hysteresis very o f t e n used unit r o o t tests with or w i t h o u t a l l o w i n g for structural b r e a k s ( R o e d 1 9 9 6 , A r e s t i s / B i e f a n g - F r i s a n c h o M a r i s c a l 1 9 9 9 , 2 0 0 0 , Papell et al. 2 0 0 0 , G u s t a v s s o n / O s t e r holm 2 0 0 7 ) . A failure of rejecting the unit r o o t hypothesis d o e s not m e a n that u n e m p l o y ment really f o l l o w s a unit r o o t p r o c e s s but m a y a s well be c a u s e d by a lack of p o w e r ; f o r studies o n p o w e r a g a i n s t fractionally integrated alternatives see e.g. D i e b o l d a n d R u d e busch ( 1 9 9 1 ) , H a s s l e r a n d Wolters ( 1 9 9 4 ) a n d L e e a n d Schmidt ( 1 9 9 6 ) . It is widely a c c e p t e d that the o u t c o m e of unit r o o t or stationarity testing is not a n invariant property of the underlying v a r i a b l e 1 but it strongly d e p e n d s on the s a m p l e period a n d the frequency of the o b s e r v a t i o n s , see e.g. J u s e l i u s ( 1 9 9 9 ) . M o s t studies use quarterly or annual u n e m p l o y m e n t series. Since time a g g r e g a t i o n affects the d y n a m i c b e h a v i o u r w e w o r k with monthly o b s e r v a t i o n s . U s i n g s t a n d a r d tests for stationarity or nonstationarity gives only i n f o r m a t i o n whether u n e m p l o y m e n t rates behave like a p r o c e s s integrated of order one, / ( l ) , or integrated of order zero, 1(0). T h e c o n c e p t of integrated time series h a s been e x t e n d e d , a n d the order of integration d is not restricted to be an integer number. O n e calls a stochastic p r o c e s s fractionally integrated of o r d e r d, 1(d), if the impulse r e s p o n s e f u n c t i o n a t l a g / declines with jd~\ T h i s c o n c e p t gives us the possibility to m e a s u r e in m o r e detail the degree of persistence of u n e m p l o y m e n t rates. F o r d = 1 we have pure hysteresis, and p a s t s h o c k s have infinite effect; for 0 . 5 < d < 1 w e observe nonstationarity with transitory s h o c k s to

1

It is quite clear that it is theoretically impossible for a bounded variable as the unemployment rate, to have a unit root, but what we get is a description of the behaviour of the variable for the underlying sample.

Hysteresis in Unemployment Rates? A Comparison between Germany and the US • 121

unemployment rate that are very persistent but die out eventually 2 . Therefore we do not only test for J ( l ) or 1(0) but we also fit fractionally integrated processes to the unemployment rates in US and Germany to measure the degree of persistence in a more differentiated way to make meaningful comparisons between the flexibility of labour markets in US and Germany. A similar approach has been employed by Crato and Rothman (1996) and Gil-Alana (2001), who use parametric models, while we rely on more flexible semiparametric procedures and recent powerful tests. Further, we allow for structural breaks or different regimes of persistence by presenting results from recursive and rolling regressions. The paper is organized as follows. Section 2 describes the data and their properties. Section 3 presents the concept of fractionally integrated processes and gives the results for US and German unemployment rates. A discussion and conclusions are given in the final section. 2

Data and preliminary results

We use seasonally adjusted monthly standardized unemployment rates 3 for Germany and the US taken from OECD. Monthly observations are available since January 1969 for Germany and since January 1960 for the US. To get a common sample for all calculations and to guarantee a sufficient number of initial values we choose 1971(1) to 2008(3) as our sample of size T = 447. Due to the German unification there occurs a level shift in German unemployment rates from December 1991 to January 1992 from 5.7 to 7.5 percent, see Figure 1. We accounted for this shift by multiplying the German data before 1992 with the factor 1.3158; this corrected series is called gcu. We neither present results for the logarithms of the data nor for their logistic transformations as proposed e.g. by Wallis (1987), since the results do not differ at all. For US unemployment rates (usu) as well as for the corrected German unemployment rates (gcu) Table 1 presents the results for the Augmented Dickey-Fuller (ADF Test, see Dickey and Fuller, 1979) and the Phillips-Perron test (PP, see Phillips and Perron, 1988) with the unit root as null hypothesis 4 . The lag lengths for the ADF tests are determined by the Hannan-Quinn criterion, see Hannan and Quinn (1979), and PP tests are performed with Bartlett kernel and automatic bandwidth selection according to Newey and West (1994). The table also contains the results for the KPSS test by Kwiatkowski et al. (1992) with the null hypothesis of 1(0). For US and Germany the KPSS tests reject the 1(0) hypothesis with an error probability less than one percent. The ADF tests with a constant show for both countries the same result, weak evidence for one unit root. Including a trend for the German data because of their trending behaviour in the sample period the ADF test presents weaker evidence

2

3

4

Such a feature is sometimes called "mean-reversion" or "level-reversion", although Phillips and X i a o ( 1 9 9 9 ) indicate that this is a misnomer given the nonstationarity. These rates give the numbers of unemployed persons as a percentage of the civilian labour force, which consists of civilian employees, self-employed, unpaid family workers and unemployed persons, see Eurostat. We use both unit root tests since they model autoregressive and/or heteroskedastic error terms in quite different ways. See e.g. Kirchgassner and Wolters ( 2 0 0 7 ) .

122 • U. Hassler and J. Wolters

Table 1 Unit root and KPSS tests, 1971(1) - 2008(3) Deterministic

laglength

usu

constant

4

gcu

constant

4

gcu

constant trend

4

ADF

band-

PP

band-

(p-value)

width

(p-value)

width

- 2.688 (0.077) - 2.567 (0.101) - 1.705 (0.705)

13

— 2.158 (0.222) - 2.423 (0.136) - 1.337 (0.877)

16

0.975**

16

1.760**

16

0.397**

15 15

KPSS

** indicates rejection of the null hypothesis at the 1 percent level

lags

Figure 2 Autocorrelograms

against the unit root hypothesis 5 . The PP tests confirm these results. They show a somewhat stronger evidence for the null of a unit root in all cases. All in all with these tests we do not really find different behaviour of the unemployment rates in Germany and the US 6 . Figure 2 displays the autocorrelograms of usu and gcu up to 36 lags. The first-order correlations are practically identical (0.993 and 0.992, respectively), while at lag 36 the difference is considerable (0.182 and 0.507, respectively). Hence, the memory in usu is clearly not as long and strong as in gcu. Moreover, the shapes seem to differ. With gcu we observe the linear decline that is characteristic for 1(1) series, while the more hyperbolic shape for usu is more typical for nonstationary 1(d) processes with 0.5 < d < 1, see Hassler (1997) for an analytical treatment and experimental evidence. 5

6

With the unadjusted German data displayed in Figure 1 we also applied the break-corrected test by Perron (1989), see Wolters and Hassler (2006) for details, and found even less evidence against the null hypothesis of a unit root. We applied unit root tests to differenced data, too. For Ausu and Agcu the null hypothesis of / ( I ) can be clearly rejected, which indicates that the unemployment rates should be considered as / ( l ) .

Hysteresis in Unemployment Rates? A Comparison between Germany and the US • 123

Response of GCU to GCU

Response of USU to USU

Figure 3 Response to Cholesky one s.d. innovations

A further approach to compare the dynamics of the two unemployment rates is to investigate their impulse response functions. We therefore estimate a bivariate vector autoregressive (VAR) model with the German and US unemployment rates 7 . According to the Hannan-Quinn criterion a lag length of five is chosen. Figure 3 shows the impulse response function together with the approximate 95 percent confidence intervals 8 for the US unemployment rate due to a shock of one standard deviation in this variable as well as the corresponding impulse response function for the German unemployment rate. These results show a quite different dynamic behaviour of the two unemployment rates. Whereas the effect of a shock in the US rate vanishes after about three years, a shock in the German rate has significant effects until about six years. This result casts doubts on the proposition that the equilibrium unemployment rate in Germany has been path independent in the last 37 years. Moreover, this evidence is in sharp contrast to the results of the unit root and stationarity tests which do not give significantly different results for the two unemployment rates. To get not only descriptive evidence for the different persistence in the unemployment dynamic we fit fractionally integrated models for the two unemployment rates.

3

Fractional integration

The fractionally integrated process yt is defined as (1 -L)\yt-n)=e„

f = 1 , . . . , T,

(1)

where e, is a covariance stationary and invertible autoregressive moving-average (ARMA) process, L is the lag operator and the fractional differences (1 — L)d are given by binomial expansion. The process is covariance stationary if and only if d < 0.5. For d = 1 the standard (integer) integrated process is reproduced. Upon inverting (1 - L)d

7

8

Compare for this also Reed (1997) who performed a similar exercise for European and US quarterly unemployment rates for the period 1973 to 1994. We present the analytically derived intervals which do not differ from the intervals coming from a Monte Carlo approach. The computations are done with EViews 6.

124 • U. Hassler and J. Wolters

and expanding the ARMA polynomials we obtain the moving-average representation in terms of white noise et (with impulse response function c ; ): t-1 yt = n + (1 - L)~de, =

,=o

ci£t-h

with

c;

~

c

id

l as

/

The effect of past shocks on yt dies out as long as d < 1: q —> 0. But the rate is so slow that the impulse response function is not summable as long as d > 0. Hence, d measures the degree of persistence. If d = 1 or d > 1, then the impulse response function does not die out. For further aspects of fractional integration we recommend the survey article by Baillie (1996). The following procedure allows to test for any specified value do of fractional integration. Due to our application we focus on the special null hypothesis Ho with do = 1 in (1). Demetrescu et al. (2008) propose a regression-based Lagrange Multiplier (LM) test, which they call augmented LM test (ALM) because it relies on a regression augmented by lags9. The test regression estimated by ordinary least squares (OLS) becomes t-i

+ i 2 * t - 2 + ••• + apxt-p + £ ( ,

Xt = 9x*t-\

Xf_.

— l >

,

(2)

with xt = (1 — L){yt — fi) = (1 — L)yt being the differences under Ho (do = 1). Demetrescu et al. (2008) propose to choose the lag length p in (2) depending on the sample size: p = [4(T/100) f / 4 ].The null H 0 can be tested with a i-type test statistic t,p for

CD CD To

O T o— 5 5 C CO D lO CD

o T* 5 r^ a> cn

o

i o

CD

O) Ì—

o

o

o

r-- a> T— CT) O) o o a> T O) *— CM

o

O

o

CO o o CM

m o o CM

fo o CM

Figure A1 Oil price (oil; deflated; in domestic currency)

6000000 5000000 4000000 3000000 2000000 1000000 0 CO

c-

a>

_ co f- ) 0c- ) ö ) 0CO) o

2 io N is i - CDn CO O C)O0 C)O0 CD Q )oi9)cn

2 co io o o o o o ~ o CM CM CM

Figure A2 Unemployment in Germany (number of unemployed persons)

Oil and Unemployment in Germany • 159

140

40 20 0

O

o 2 C O m N. f- r»05 o> a> '—

o

o

o 2 a> CO I-a> cn

o 2 CO CO a>

o o 2 2 m O oo C O) CT)

o o 2 2 a> CO a> o> CT)

o 2 CO o> CT) T—

o 2 IO cn a> •—

o T— o 2 2 r^. o> CT) o> CT) o>

o 2 T— o o CM

O 2 00 o o CM

o 2 in o o CSI

o 2 o o Csl

Figure A3 Production of total industry in Germany (index: base year 2000 value 100)

o 2 CO O)

o o o 2 2 2 m CT) CT5 CT> CT)

o 2 T— co cn

o 2 CO CO a>

o 2 m co 1CT) —

o 2 f00 O)

o a> oo TCT) —

o 2 CT) CT)

o T— O O O ^— 2 2 2 CO IO N. CT) CT) CT) CT) CT) CT5 CT) O) 05

O

O T— 2O o C o o o CM CM

o 2 in o o CM

o 2 r^ o o CM

Figure A4 Interest rate (real; based on rates for overnight money in Frankfurt, Germany)

160 • A. Löschel and U. Oberndorfer

o> co f^ o o o 2 O Ifi S O) -0,02 Ofc ) t0 5 t-c nN-a -0,01

co m O a> CO CD IO o o o 1- o o o o o i - n m s o i i - N ^ m i o O N ^ i o COOOCOC00005050f)C)C350000 > a ) 0 ) 0 5 c n a ) c n a > o > c » o ) o o o o T-T-T-T-T-T-T-T-T-CMCMOJCM

Figure A5 Inflation rate (annual; based on the German consumer price index)

Table A1 Unit root tests full sample period (10/1973-01/2008) ADF

Levels PP

1 Unemployment - 4.69*** -2.70 I Oil price (oil) -2.30 -2.23 - 13.05*** - 12.63*** I Oil price increase (dloilpos) ( Net oil price increase - 11.86*** - 11.26*** (nopi) - 3.25* l Interest rate - 2.35 ( Industrial production - 3.05 - 11.46*** i Inflation rate - 2.85 - 2.38

KPSS

ADF

1 s t Differences

0.35*** 4.46*** 0.32*** - 12.54*** 0.08 0.10 0.21** 0.11 0.15**

-

- 5.04*** — 4.72*** - 3.56***

PP

KPSS

- 11.22*** - 11.89***

0.06 0.17**

-

-

-

-

-26.67*** -46.51*** - 24.80***

0.08 0.05 0.05

Note: t indicates that the variable is employed in logarithmic form. ADF: Augmented Dickey-Fuller Test (null hypothesis: unit root), PP: Phillips-Perron Test (null hypothesis: unit root), KPSS: Kwiatkowski-Phillips-Schmidt-Shin Test (null hypothesis: stationarity). *, ** and *** show rejection of null hypothesis at the 10 % - , 5 % - , and 1 % level, respectively. Unit root tests include linear time trend. Lag length according to Schwarz Information Criterion. 393 obs.

Oil and Unemployment in Germany • 161

Table A2 Unit root tests post-unification sample (10/1990-01/2008) ADF

Levels PP

KPSS

ADF

1 st Differences PP

- 4.63*** _ 4 84**» 0.28*** 4.48*** - 11.17*** - 3.75** - 3.16* 0.24*** - 12.54*** - 10.08*** -11.26*** - 11.31*** 0.10

( Unemployment e Oil price (oil) i Oil price increase (dloilpos) t Net oil price increase - 13.21*** - 26.56*** 0.13* (nopi) i Interest rate - 2.02 - 0.98 0.22*** - 8.81*** 0.29*** I Industrial production - 2 . 6 3 - 2.05 - 1.71 0.34*** t Inflation rate

-

-

- 3.92** - 2.89 - 3.56**

11.72*** - 50.26*** - 15.14***

KPSS 0.13* 0.05 -

-

0.08 0.18** 0.05

Note: I indicates that the variable is employed in logarithmic form. ADF: Augmented Dickey-Fuller Test (null hypothesis: unit root), PP: Phillips-Perron Test (null hypothesis: unit root), KPSS: Kwiatkowski-Phillips-Schmidt-Shin Test (null hypothesis: stationarity). *, ** and *** show rejection of null hypothesis at the 10 % - , 5 % - , and 1 % level, respectively. Unit root tests include linear time trend. Lag length according to Schwarz Information Criterion. 208 obs.

References Barsky, R . B . , L. Kilian ( 2 0 0 4 ) , Oil and the macroeconomy since the 1 9 7 0 s , Journal of E c o n o m i c Perspectives 18: 1 1 5 - 1 3 4 . Brown, S.P.A., M . K . Yiicel ( 1 9 9 9 ) , Oil prices and the economy, Southwest E c o n o m y 4 : 1 - 6 . Brown, S.P.A., M . K . Yiicel ( 2 0 0 2 ) , Energy prices and aggregate economic activity: An interpretative study, Quarterly Review of Economics and Finance 4 2 : 1 9 3 - 2 0 8 . Farzanegan, M . R . , G . M a r k w a r d t ( 2 0 0 8 ) , T h e effects of oil price shocks on the Iranian economy. Faculty of Business M a n a g e m e n t and Economy, Dresden University of Technology. Ferderer, J.P. ( 1 9 9 6 ) , Oil price volatility and the macroeconomy. Journal of M a c r o e c o n o m i c s 18: 1-26. Flaig, G. ( 2 0 0 5 ) , Time series properties of the German production index. Allgemeines Statistisches Archiv 8 9 : 4 1 9 - 4 3 4 . Franz, W. ( 1 9 8 3 ) , T h e past decade's natural rate and the dynamics of German unemployment. A case against demand policy? European Economic Review 2 1 : 5 1 - 7 6 . German Council of E c o n o m i c Experts ( 2 0 0 6 ) , Conflicting interests - missed opportunities. Report 2 0 0 6 / 2 0 0 7 , Wiesbaden. Granger, C.W.J., P. Newbold ( 1 9 7 4 ) , Spurious regressions in econometrics. J o u r n a l o f Econometrics 2 : 1 1 1 - 1 2 0 . Greening, L.A., W.B. Davis, L. Schipper, M . Khrushch ( 1 9 9 7 ) , Comparison of six decomposition methods: Application to aggregate energy intensity for manufacturing in 10 O E C D countries. Energy Economics 19: 3 7 5 - 3 9 0 . Hamilton, J . D . ( 1 9 9 6 ) , This is what happened to the oil price-macroeconomy relationship. Journal of M o n e t a r y Economics 3 8 : 2 1 5 - 2 2 0 . Hamilton, J . D . ( 1 9 8 3 ) , Oil and the macroeconomy since World War II. Journal of Political Economy 9 1 : 2 2 8 - 2 4 8 . He, Z . , K. M a e k a w a ( 2 0 0 1 ) , On spurious Granger causality. Economics Letters 7 3 : 3 0 7 - 3 1 3 . Hooker, M . A . ( 1 9 9 6 ) , W h a t happened to the oil price-macroeconomy relationship? Journal of Monetary Economics 3 8 : 1 9 5 - 2 1 3 . IEA - International Energy Agency ( 2 0 0 8 ) , World Energy O u t l o o k 2 0 0 8 . Paris. J o n e s , D.W., P.N. Leiby, I.K. Paik ( 2 0 0 4 ) , Oil price shocks and the macroeconomy: W h a t has been learned since 1 9 9 6 . T h e Energy Journal 2 5 : 1 - 3 2 . Koop, G., M . H . Pesaran, S . M . Potter ( 1 9 9 6 ) , Impulse response analysis in nonlinear multivariate models. J o u r n a l of Econometrics 7 4 : 1 1 9 - 1 4 7 .

162 • A. Löschel and U. Oberndorfer

Lin, S.X. (2008), Effect of Chinese oil consumption on world oil prices. Cass Business School, City University London. Meyer, M . , P. Winker (2005), Using H P filtered data for econometric analysis: Some evidence from M o n t e Carlo simulations. Allgemeines Statistisches Archiv 89: 3 0 1 - 3 1 8 . M o r k , K.A. (1989), Oil and the Macroeconomy W h e n Prices G o Up and D o w n : An Extension of Hamilton's Results. Journal of Political Economy 97: 7 4 0 - 7 4 4 . N a k a , A., D. Tufte (1997), Examining impulse response functions in cointegrated systems. Applied Economics 29: 1 5 9 3 - 1 6 0 3 . Papapetrou, E. (2001), Oil price shocks, stock market, economic activity and employment in Greece. Energy Economics 23: 5 1 1 - 5 3 2 . Rotemberg, J.J., M . W o o d f o r d (1996), Imperfect competition and the effects of energy price increases on economic activity. Journal of Money, Credit and Banking 28: 5 4 9 - 5 7 7 . Schmidt, T., T. Z i m m e r m a n n (2005), Effects of oil price shocks on German business cycles. RWI: Discussion Papers N o . 31, Essen. Schmidt, T., T. Z i m m e r m a n n (2007), Why are the effects of recent oil price shocks so small? Ruhr Economic Paper N o . 21, Essen. Sims, C.A. (1980), Macroeconomics and Reality. Econometrica 48: 1—48. Steiner, V. (2001), Unemployment Persistence in the West German Labour M a r k e t : Negative Duration Dependence or Sorting? O x f o r d Bulletin of Economics & Statistics 63: 9 1 - 1 1 3 . Tanaka, N . (2008), Energy security, sustainability and dialogue. Presentation at the International Energy Forum Secretariat, Riyadh. Z E W - Centre for European Economic Research (2008), Energie wird teurer. Schwerpunkt Energiemarkt 2: 3 - 4 . Dr. Andreas Löschel, Centre for European Economic Research (ZEW), P.O. Box 10 34 4 3 , 6 8 0 3 4 M a n n h e i m , Germany. E-Mail: [email protected] Ulrich Oberndorfer, Federal Ministry of Economics and Technology, Scharnhorststr. 3 4 - 3 7 , 10115 Berlin, Germany. E-Mail: [email protected]

Jahrbücher f. Nationalökonomie u. Statistik (Lucius & Lucius, Stuttgart 2009) Bd. (Vol.) 229/2+3

Living Standards in an Aging Germany: The Benefits of Reforms and the Costs of Resistance By Axel Borsch-Supan and Alexander Ludwig, Mannheim* JEL J11, J21, D13, E27, H55, F16, F21 Aging, pension reform, labor market reform, labor supply reactions.

Summary The extent of the demographic change in Europe and especially Germany is dramatic and will deeply affect future labor, financial and goods markets. The expected strain on public budgets and especially social security has received prominent attention, but aging poses many other economic challenges that threaten growth and living standards if they remain unaddressed. This paper investigates the potential benefits of pension and labor market reform for growth and living standards, taking into account behavioral reactions to specific reforms. Which behavioral reactions will strengthen, which will weaken reform policies? While Germany has a large unfunded pension system and vulnerable labor markets, Germans show remarkable resistance against pension and labor market reform. Can Germany maintain its standard of living even if behavioral reactions offset some of the current reform efforts? The paper uses a novel modeling approach to distinguish between exogenous and endogenous components of labor supply in order to shed light on these questions.

1

Introduction

This paper analyzes the aging process and its macroeconomic implications for Germany. Germany has a large pay-as-you-go financed social security system. The need for reform, in particular of the health and long-term care sectors, has already received prominent attention. In addition, Germany has a labor market characterized by low participation rates, high unemployment, and high wages. It is particularly vulnerable to the challenges of globalization due to the high tax and contribution burden in total labor compensation. In spite of these problems, Germans have been remarkably resistant to labor market and pension reform. If the government anyway manages to push such reforms through parliament, workers may thus react adversely and undo at least some of the expected effects of the reforms. The main questions posed in this paper are therefore: What can pension and labor market reforms ideally achieve? What are possible behavioral reactions to reform policies? Which direction will they take and how large are they? Hence, what can pension and labor market reforms actually achieve? And ultimately: Can Germany maintain its high living standards even if behavioral reactions offset some of the current reform efforts? * We are grateful to two anonymous referees and the editor for their helpful comments. Financial support was provided by the Deutsche Forschungsgemeinschaft, the Land Baden Württemberg, and the German Association of Insurers. The usual disclaimer applies.

164 • A. Bòrsch-Supan and A. Ludwig

Some behavioral reactions will strengthen r e f o r m . A good example is raising the statutory retirement age. It has direct effects on the labor supply by bringing older individuals t o the labor market. Indirect effects emerge f r o m endogenous labor supply reactions, e.g., t h r o u g h incentive effects generated by the t a x and contribution burden t h a t actuarially unfair social security systems impose on households. Raising the retirement age will lower social security contributions in such pension systems. In response t o rising net wages, labor supply m a y then increase at all ages. There are, however, also behavioral effects that w e a k e n policy reforms. To take up the same example, older workers, n o w forced to w o r k longer, may exploit part-time opportunities given by the pension system. In some countries (e.g., Finland) such opportunities led to a very early transition to part time w o r k with the perverse result that in some sectors h o u r s supplied actually decreased in response to pension r e f o r m . Along the same line, encouraging female labor supply, e.g. through public provision of day care facilities, m a y precipitate a decrease in male labor supply. This within-household substitution w o u l d be perfectly rational if households desire joint leisure and joint household p r o d u c t i o n . Little is k n o w n a b o u t these behavioral reactions. O n e of the key issues taken up in this paper is therefore to model and calibrate behavioral reactions to r e f o r m . Which behavioral reactions will strengthen, which will w e a k e n r e f o r m policies? W h a t are their quantitative effects? We will build a simple model of reforms a n d reform backlashes into an overlapping generations model of the Auerbach et al. (1983)/Auerbach and Kotlikoff (1987) type a n d extend it to a multi-country version m o r e appropriate for G e r m a n y (Borsch-Supan et al. 2006). 1 A first i m p o r t a n t and novel feature of our model is the distinction between exogenous labor supply c o m p o n e n t s (as key results of labor m a r k e t a n d pension reform) and endogenous labor supply c o m p o n e n t s (in order to represent possible reform backlash). To keep the language simple, w e call the exogenous labor supply c o m p o n e n t "labor force p a r t i c i p a t i o n " , a n d the endogenous labor supply c o m p o n e n t " w o r k i n g h o u r s " . This language is metaphorical as we are well a w a r e that both labor force participation a n d w o r k ing hours have endogenous as well as exogenous c o m p o n e n t s . T h e metaphorical language chosen comes f r o m our thinking of labor m a r k e t and pension reforms as lifting institutional constraints. Typical constraints are a m i n i m u m labor market entry age generated by the school system, constraining the labor force participation of the young; a n early labor m a r k e t exit age generated by the pension system, effectively constraining the labor force participation of the old; inflexible w o r k i n g hours and unavailable day care facilities, constraining female labor force participation. This view of lifting restrictions motivates our modeling strategy and the language behind it: labor m a r k e t a n d pension r e f o r m s are represented by exogenous changes of labor supply at the extensive margin (the n u m b e r of w o r k i n g persons in an economy). H o u s e h o l d s are then modeled to respond to the changes of labor supply by changing their w o r k i n g hours (the intensive margin of labor supply). Endogenous h o u r s ' supply m a y increase, e.g., if distorting social security taxes and contributions decline as an implica-

1

Similar multi-country OLG models have been developed, among others, by Feroli (2002), Henriksen (2002), Brooks (2003), Fehr et al. (2005), Domeij and Floden (2006), Aglieta et al. (2007), Attanasio et al. (2006, 2007) and Kriiger and Ludwig (2007).

Living Standards in an Aging Germany • 165

tion of pension reform. The opposite reaction is also possible: endogenous hours' supply may decrease in response to an exogenous change of the number of working persons if there is intra-household substitution between the number of persons working and the hours worked by each person. Another important feature of our model is its multi-country nature. Unlike to the U.S., no country in Continental Europe is even approximately modeled by a closed economy. Germany has a large export sector and considerable foreign direct investments. These provide a second source of opportunities during the global aging process: not all income needs to come from domestic production, and G N P may become substantially larger than GDP if foreign direct investments create large returns. We complement Germany as a country which saves more than it invests with France and Italy representing its closest trading partners, and with the U.S. representing the rest of the world currently absorbing the Continental European savings. The key results of our paper rest on a set of three-way comparisons of polar cases. The first dimension reflects labor market policies. One polar case is the complete failure to adapt those institutional restrictions that keep labor force participation so low in Germany. The result are unchanged low labor force participation rates by age and gender also in the future. The other polar case, for some an extreme, is the adaptation of all societal systems from kindergarten to retirement policies to increase age and gender specific labor force participation rates across the board. As a second dimension, we model two polar cases of the interaction between pension policy and labor supply. One extreme case is a fully-funded, voluntary private accounts system that has no labor supply distortions. The other polar case is a pay-as-you-go pension system with flat benefits financed by contributions that are perceived as 100 % taxes with the associated large labor supply distortions. Almost all pension systems, especially those in Germany, France, Italy, and the U.S., are in between these two extremes. Finally, the third dimension in these comparisons isolates behavioral effects. One extreme is a fixed hours' supply by each working individual. For the other polar case, we derive a supply function of working hours which is responsive to wages net of taxes and contributions, but also to restrictions on household labor participation. The resulting eight scenarios bracket what we think are a reasonable set of possible outcomes. We find that it is possible to maintain living standards in spite of the dramatic population aging process but that it requires further reforms. Moreover, possible reform backlashes are to be taken seriously because they can have sizeable effects. The paper is structured as follows. Section 2 briefly sets the demographic background. Section 3 describes the current labor market situation and our labor market reform scenarios. Section 4 presents the multi-country computational general equilibrium model with a combination of exogenous and endogenous labor supply components. Section 5 delivers our main results in the set up just described. We vary the institutional framework of labor markets and pensions in order to investigate the interactions between pension and employment policies and the behavioral reactions to pension and labor market reform. Since higher old-age labor force participation raises issues of age-specific productivity, they are briefly addressed in Section 6. Section 7 concludes.

166 • A. Börsch-Supan and A. Ludwig

2

Demography

While the patterns of population aging are similar in most countries, timing and extent differ substantially. We focus on Germany and compare it with its trading partners France and Italy. France is considerably younger and will age later and to a slower extent than Germany and Italy. These differences can largely be attributed to different fertility rates (France has a fertility rate close to the replacement level, see Table 1), while Germany and Italy loose about a third of their population from generation to generation due to fertility rates that are below 1.4. Life expectancy also differs among the three countries. Germany has a noticeable lower life expectancy than France and Italy, and similar for healthy life expectancy. Note that healthy life expectancy is about 10 years higher than the average retirement age, providing some room for further increases in retirement age, see section 3. We compute the future demography based on three key assumptions. First, we provide projections of mortality based on a Lee-Carter decomposition, using past mortality rate changes derived from the H u m a n Mortality Database (2008). Table 1 shows the resulting life expectancies in 2050 (column 5). They coincide with the current U N projections for Germany, but are slightly higher for France and Italy (UN has age 85 while our projections yield age 86 and 87, respectively). Second, we assume that fertility rates are exogenous and remain constant as given by Table 1. Third and similarly, we assume constant and exogenous migration flows, based on the current medium variant of the U N projections (France 100,000, Germany 150,000, Italy 135,000, and U.S. 1,100,000 net migrants per year) which is about the long-term average. It is important to note that these migration flows are small relative to the decline in the labor force projected in section 3. Figure 1 shows the total population aged 15 years and over which will be the base of our projections and simulations, and compares them with the United States. There will be population growth in France and the U.S. but significant decline in Germany and a somewhat smaller decline in Italy after 2020, mainly due to the about 30 % higher migration rate to Italy. The fifth line represents the aggregate of France, Germany, and Italy which we will call EU-3. Truly remarkable is the decline of the working age population (age 2 0 - 6 4 ) , see Figure 2. Relative to total population aged 15 and older, the U.S. will loose about 10 % of their working age individuals between 2005 and 2050. In Germany, the loss is almost twice as high with 18 % , in Italy even 22 % . Table 1 Fertility rates and life expectancy

France Germany Italy

Total fertility rate

Life expectancy at birth

Healthy life expectancy

Life expectancy in year 2050

1.89 1.34 1.29

80.3 79.0 80.4

71.3 70.2 71.0

86 84 87

Source: Eurostat (2008), OECD Health Data 2007, W H O (2006), and own computations.

Living Standards in an Aging Germany • 167 100.0%

95.0% -*-U.S.A.

90.0%

—O- FRANCE 85.0%

-*-EU-3 -a-GERMANY

80.0%

-Û-ITALY

75.0%

70.0%

.

2005

»

2010

2015

2020

T

2025

2030

2035

2040

2045

2050

Figure 1 Population 15 years and older, indexed to 2005 = 100 % Source: Own projection based on assumptions detailed in text 100.0%

95.0% -X-U.SA

90.0%

-O-FRANCE 85.0%

-*-EU-3 -O-GERMANY

80.0%

-Û-ITALY

75.0%

70.0%

. 2005

2010

2015

2020

2025

—I 2030 2035

2040

2045

2050

Figure 2 Working age population as share of total population aged 15 +, 2005 = 100%

Source: O w n projection based on assumptions detailed in text. Working age is age 20 to 64.

3

Employment and labor market reforms

Working age population is not equal to employment. Aggregate employment is a result of labor market entry age, female labor force participation, unemployment rates, and labor market exit age, to name the four most important parameters. These parameters are strongly governed by institutional restrictions. Labor market entry age, e.g., is a function of the school system. Germany, e.g., has regulations that generate late entries into the

168 • A. Borsch-Supan and A. Ludwig

school system, a long duration in high schools and universities, and thus a late labor market entry age. Similarly, female labor force participation is a function of institutions such as kindergarten and afternoon school which tend to be provided by public entities in Europe. Unemployment is a function of the duration and generosity of unemployment compensation. Labor market exit, finally, is strongly governed by pension regulations that effectively make the early eligibility age also the effective age of labor market withdrawal. Hence, from an individual's point of view, labor supply has important exogenous components which restrict possible endogenous labor supply decisions. It is unlikely that these exogenous components remain unchanged over the course of population aging and the general change of society over the next two decades. We therefore define two polar scenarios representing the potential changes in the institutional framework restricting households' labor supply decisions: • In the status quo scenario (STATQUO), age and gender specific labor force participation rates will remain as they are at baseline in 2005; this was the scenario underlying Figure 4. • The labor market and pension reform scenario (LREFORM) includes four reform steps: • RETAGE: an increase in the retirement age by 2 years; • JOBENTRY: a decrease in the job entry age by 2 years; • FEMLFP: an adaptation of female labor force participation rates to those of men; • UNEMP: a reduction of unemployment to 40 % of its current level. The increments are motivated by actual policy proposals: in Germany, the statutory retirement age has been raised from 65 to 67 years in a series of transitions until about 2020. The change in the German high school and university system (the so-called Bologna process) is expected to decrease duration in schooling by about 2 years. Finally, 40 % of current unemployment represents the conventional estimate of the NAIRU (Ball/ Mankiw 2002). These reform steps will be phased in linearly between 2010 and 2040. The increase in retirement age (the decrease in the job entry age) is modeled as a shift of the distribution of labor force participation rates by age to the right (to the left, respectively), thereby increasing the flat part of the distribution in the middle, see Figure 3. Overall, these reform steps do not appear to be overly radical; in fact, their combination would lead in 2040 to labor force participation rates almost identical to those in Denmark today as indicated by Figure 3. Nevertheless, attempts to actually execute reforms with those goals have faced stiff opposition in France and Italy, and more recently and to a somewhat lesser extent also in Germany. This motivates our modeling approach in section 4. Figure 4 displays the resulting trajectories of the number of working individuals. Each reform step is additive to the one before. The trajectory labeled " L R E F O R M " has all four reform elements implemented. The trajectories are very different across countries. France can easily compensate the slightly declining number of individuals of working age by a combination of two or three of the above policy changes, while Germany will be able to only partially offset the loss in working age population. The three countries also differ in the efficacy of the four policy parameters. Note in particular Italy with a large jump if female labor force participation adapts to that of men.

Living Standards in an Aging Germany • 169

1.00 0.90 0.80

a

s

0.70

c

Sn

0 60

ï

0 .50

J

0.40

I

0.30

a

I

— a — M e n (DK) — » — W o m e n (DK) • 4 • • Men (OE) •

Women (DE)

0.20 0.10 0.00 1415

20

-

—

25

30

— 3S

40

-

45

50

-

— 55

60

65

70

Age

Figure 3 German and Danish labor force participation rates Source: O w n computations based on the German Mikrozensus (www.destatis.de) and Statistics Denmark (www.statbank.dk)

This is due to the very low female labor force participation currently in Italy. T h e irony is of course that, because Italy's pool of hitherto unused labor capacity (in particular w o m en) is so large, tapping it provides a very large opportunity t o counteract the effects of p o p u l a t i o n aging. Italy, while aging more t h a n Germany, is thus better off than Germany which has less r o o m to increase labor force participation.

120.0%

100.0%

^

-o

80.0%

&

o

60.0% -^LREFOR^RETAGE+JENTRY+FEMLFP+UNEMP - X - RETAGE+JENTRY+FEMLFP

40.0%

- û - RÉTAGE+JENTRY -O-RETAGE

20.0%

—0-STATQUO

0.0%

2005

2010

2015

2020

2025

2030

2035

Figure 4a Employment, indexed to 2005 = 100 %, France

2040

2045

2050

170 • A. Borsch-Supan and A. Ludwig 120.0% 100.0% =6= -O

80.0%

O

60.0% -*-LREFORM=RETAGE+JENTRY+FEMLFP+UNEMP

40.0%

—X— RETAGE+JENTRY+FEMLFP —A— RETAGE+JENTRY -O-RETAGE

20.0%

—O— STATQUO

0.0%

2005

2010

2015

2020

2025

2030

2035

2040

2045

2050

Figure 4b Employment, indexed to 2005 = 100 % , Germany 120.0%

100.0% 80.0%

60.0% - LREFORM=RETAGE+JENTRY+FEMLFP+UNEMP - X - RETAGE+JENTRY+FEMLFP

40.0%

- 6 - RETAGE+JENTRY -RETAGE

20.0%

0.0%

-STATQUO

2005

2010

2015

2020

2025

2030

2035

2040

2045

2050

Figure 4c Employment, indexed to 2005 = 100%, Italy Source: O w n calculations

4

A dynamic macroeconomic model with exogenous labor force participation and endogenous hours' supply

We use a dynamic open-economy macroeconomic model that allows us to analyze the effects of the labor market reforms described in the previous section on G D P and consumption per capita in an aging Germany. As described in the introduction, the model is a standard O L G model, but we extend it in two important dimensions. First, it represents a

Living Standards in an Aging Germany • 171

set of large open economies in which capital flows equilibrate national rates of return to productive capital. Second a n d most importantly, we model behavioral effects t o labor m a r k e t and pension reform by introducing a labor supply f u n c t i o n that has both an endogenous and an exogenous c o m p o n e n t . To keep the language simple, we call the exogenous labor supply c o m p o n e n t " l a b o r force p a r t i c i p a t i o n " , a n d the endogenous labor supply c o m p o n e n t " w o r k i n g h o u r s " . This language is metaphorical as w e are well a w a r e t h a t both labor force participation and working h o u r s have endogenous as well as exogenous components. We treat the reforms and the resulting variation in employment n u m b e r s as exogenous. H o u s e h o l d s then endogenously adjust hours w o r k e d and m a y thus strengthen or counteract parts of the labor m a r k e t reforms. M o r e formally, our main assumptions on this interplay between the exogenous variation of employment n u m b e r s and h o u r s w o r k e d are as follows. We model the decision of a household with preferences over c o n s u m p t i o n and leisure. Total labor supply of a household of age / as derived f r o m the household's optimization is the p r o d u c t of the exogenous c o m p o n e n t //, and the endogenous c o m p o n e n t hj. The crucial difference between the t w o labor supply c o m p o n e n t s is t h a t the endogenous c o m p o n e n t hj m a y not exceed an upper b o u n d (in our metaphorical language: the time e n d o w m e n t which we normalize to one) while the exogenous c o m p o n e n t /,, can take any positive value. Stated in the f o r m of equations, the household derives utility f r o m c o n s u m p t i o n c(.;,, and leisure 1 — • h t jj where the household's per period utility function is given by u{cw,

1 - k,jj

• /,,,.,) =

( y , to ^ 111 - J LU ¿ 2 u < z u O o i i < 0É 2 ^ < CD O 0- O LL

o

o

I

o

o I ff*

o I

o

I

oooooooo

O * ^ t l moo i n W O I I

in co rs ro i n (N fs» Ö (N Ö Ö ^ I I I I

QC

^

o

o

O m O I I I

00 00 O VD o o

O

ó

vo . rv * i n oo o\ , m - -a- , e o ^ T - n N i N t f f i m m i f ì m r o o

I

I

(T) *

t

in

I

o

I

-

o

o d I

o

ó

o I

5 P

o

8

m ^ r N ^ O N ^ o ^ f N ^ ( N f f i i O ^ m M O O n l D i N f ^ ^ i - i - m O O O * -

Ol a* m

3

8

m

0\ 00 rfN s f N ^ N ^ r ^ N « oooor^rNr^mvD l N ' i O O O r O f N o o o o o ' I I "

z

Q

1; s> £ u o Cö

to
a0 '

1,P'

06 )

1,7/'

The second term involves the counterfactual wage distribution for the 1999 sample of individuals using 1999 weights (superscript 99) with the 1999 distribution of personal characteristics P99. This counterfactual should give us the wage distribution for the 1999 sample of individuals with their personal characteristics had they been assigned the 2006 task distribution and did the 2006 coefficients apply. The operationalization of this intuitive definition is not straightforward. For a similar nonlinear decomposition problem, Fairlie (2005) discusses to assign the counterfactual task characteristics randomly (im-

Can a Task-Based Approach Explain the Recent Changes in the German Wage Structure? • 229

plicitly assuming independence between the different sets of characteristics) or based on the rank in the distribution of predicted outcomes. We think both strategies suffer f r o m not taking account of the relationship between personal characteristics and task assignments. Thus, we suggest and use an alternative a p p r o a c h building on our interest in the counterfactual situation of a person in the 1999 sample w h e n this person is assigned the 2 0 0 6 tasks according t o the way the labor m a r k e t in 2 0 0 6 w o r k s . To operationalize this notion, we estimate probits regressing the task d u m m y variables on personal characteristics for the 2 0 0 6 sample mimicking the task assignment in 2 0 0 6 . We use these estimates to simulate the task assignment for our counterfactual by adding a r a n d o m error term for each individual t o the fitted latent variables and then determine the sign. 1 9 T h e second c o m p o n e n t 4 defined by 4

involves the effect of the changes in task assignments and is

= 4 9 9 ( P 9 9 , T 0 *, a» 6 , a f t , a°' T ) -

9 9 q e (P",

9

T

9

, ,

a«),

where the first term is the counterfactual used to define the first c o m p o n e n t and the second term involves the counterfactual wage distribution for the actual 1999 sample of individuals using both the personal characteristics P 9 9 and the task assignment T " . The sum Ale + A^ involves the full characteristics effect in a Blinder-Oaxaca type decomposition where the differences in characteristics is evaluated at 2 0 0 6 coefficients. T h e third c o m p o n e n t Ag involves the effect of the changes in the intercepts, i.e. the changes in the conditional expectation and the conditional variance for those individuals with average 1999 characteristics. This is defined by ~ 49{P99

4

J

9 9

^

We interpret Ag as the residual change (time trend) in the wage distribution which is unexplained by changes in characteristics or slope coefficients. The f o u r t h c o m p o n e n t Ag reflects the change in the coefficients of personal characteristics and is defined by A* -

„99 (»99

T99

99

06

06 \ _

„99 ,r>99

-r-99

99

99

„06 x

T h e last c o m p o n e n t Ase reflects the change in the coefficients of task characteristics and is defined by 4

=

(P", T " , a90\aZ, aft) -

(P 9 *, T 9 9 , a 9 9 , a? 9 ,, a? 9 T ).

T h e sum Ag + A# + Ag involves the full coefficient effect in a Blinder-Oaxaca type decomposition where the differences in coefficients is evaluated at 1999 characteristics coefficients. 19

Heinze (2008) uses a conceptually similar idea to decompose the gender wage gap into the components associated with personal characteristics and the components associated with firm characteristics. To construct the counterfactual with female personal characteristics and male firm characteristics, she uses nearest neighbor matching to assign the most similar male worker to a female worker in terms of her personal characteristics and then uses the firm characteristics of this matched male worker.

230 • D. Antonczyk, B. Fitzenberger, and U. Leuschner The overall decomposition of the changes in the wage distribution between 1999 and 2006 can be summarized as A°96'"=

A^ Personal

^ Task

'^^^v

Characteristics

Residual effect '

+ Personal v

Task V

'

Coefficients

Note that our definition of counterfactuals focuses on the 1999 sample of individuals defined by their personal characteristics and their sample weights. Consequently, we also used the same simulated value of the random variable £, for all simulations in one run of the 10 simulations involved. We report bootstrapped standard errors on our decomposition estimates which take account of the sample variability of the estimated model parameters. To do so, we implement a parametric bootstrap and resample the model coefficients from their estimated asymptotic distribution estimated by maximum likelihood.

4

Results

We analyze the change in the distribution of hourly wages for full-time working males between 1999 and 2006. To present our empirical results, we proceed in three steps. First, we describe the changes in the wage distribution. Second, we present our estimated heteroscedastic interval regressions for wages. Third, we use our regression estimates to decompose the changes in the wage distribution. Our regression results are of descriptive nature as we only control for observable characteristics, i.e. we do not claim to estimate causal effects. Further detailed results are available in the longer discussion paper version of this paper Antonczyk et al. (2009). 4.1 Descriptive evidence To analyze the changes in inequality of hourly wages among full-time working males, we analyze percentile differences in log wages, namely the 80-20, the 80-50, and the 50-20 differences. 20 The 80-20 difference is a measure of overall wage inequality, while the 8050 and 50-20 differences reflect wage inequality at the top and at the bottom of the distribution, respectively. 21 For the descriptive results in this section, we first impute the midpoints of the earnings interval that is given by the respondent for the survey carried out in 1999 and we use the actual earnings reported in 2006. 2 2 These earnings data are divided by reported 20

21

22

For our analysis only full-time working men are taken into account that are between 2 5 and 55 years old, whereas working full-time is defined as working more than 2 5 hours per week. Furthermore, only German citizens employed in West Germany are considered. Finally individuals that respond that they work more than 71 hours per week are excluded from the analysis, as we consider this answer as not being reasonable. We focus here on the 2 0 %, 5 0 %, and 80 % quantiles as being respresentative for the wage distribution. The simulation results in section 4.3 provides more detailed graphical evidence, which is likely to be more reliable for reasons discussed below. N o t e that the statistical uncertainty increases at more extreme quantiles. About 15 % of the respondents in the survey carried out in 2 0 0 6 only respond whether their gross monthly earnings is below or above 1500 and 9 0 % in this group report it to be above. As we cannot impute reasonable values for these individuals, they are excluded in this descriptive part.

Can a Task-Based Approach Explain the Recent Changes in the German Wage Structure? • 231

Wages 1998/99 and 2005/06

1998/99

2005/06

Figure 1 Univariate density estimates of unconditional wages hours 2 3 to obtain hourly wages. Then, we calculate the 2 0 % quantile, the median a n d the 80 % quantile of log hourly wages and report the differences between these quantiles. N o t e that we tend to underestimate wage inequality in both years because for 1999 the earnings data are interval coded and for 2 0 0 6 we ignore those 15 % of the respondents w h o d o not report their actual earnings and w h o are likely t o have fairly high earnings (see footnote 22). Table 9 Change in Wage Inequality from 1999 to 2006 (Hourly Wages - based on Interval Midpoints) 80-20 percentile 1999 2006

80-50 percentile 1999 2006

0.57

0.29

0.64

0.55

0.71

0.57

0.57

0.56

0.60

50-20 1999

2006

0.34

0.28

0.30

0.39 Med skilled 0.31 0.31 High skilled 0.27 0.29

0.30

0.33

0.26

0.26

0.29

0.31

0.25

All Low skilled

Table 9 provides the inequality measures for both years. The overall 80-20 difference in 1999 is .57 a n d it increases t o .64 in 2 0 0 6 , i.e. we observe an increase by 7 log ppoints. T h e increase in the upper part (80-50) a m o u n t s to 5 log ppoints (.34-.29) a n d in the lower p a r t (50-20) to 2 log ppoints (.30-.28). For the reasons discussed above, these n u m b e r s may differ f r o m our subsequent simulation results discussed in section 4.3. Figure 1 provides kernel-density estimates of the wage distribution. It shows that in 2 0 0 6 the distribution is flatter and more dispersed t h a n in 1999.

23

Over 90 % of the respondents state that they work between 35 h and 55 h per week.

232 • D. Antonczyk, B. Fitzenberger, and U. Leuschner

Wage inequality differs by skill level (low, medium, high). Whereas wage inequality among medium-skilled workers has remained stable over time, wage inequality among low-skilled and high-skilled workers has increased. Specifically, low-skilled workers experienced a strong increase in wage inequality: we find evidence of an overall change of 16 log ppoints which is mainly driven by an increase of 14 log ppoints at the top of the wage distribution. High-skilled workers experienced a rise in overall wage inequality of 4 log ppoints which is associated with a similar increase in inequality both at the bottom and at the top of the wage distribution. 4.2 Parameter estimates Table 10 and Table 11 provide the different parameter estimates of the heteroscedastic interval regressions for log wages. The upper part of the tables shows the parameter estimates for the conditional expectations (mean log wages) and the lower parts show the parameter estimates for heteroscedasticity. Table 10 and Table 11 contain the parameter estimates for the two years 1999 and 2 0 0 6 without occupations and our proposed AFL-Task-Indices (AFL1-AFL4). These specifications also include the covariate NJC (sum of reported tasks) as one measure for job complexity. It is notable that the estimated parameter in the heteroscedasticity part for computer use shifts from about — .02 and being insignificant to — .09 in the second survey and now being highly significant. This can be interpreted as a wage smoothing effect of computer use. Using a computer also seems to be associated with higher average wages in 2006 than in 1999, as the parameter estimates grow over time (.15 versus .12). Considering the estimated parameters of AFL-Task-Indices 24 , we observe, that for this first specification all of them are less rewarded in 2 0 0 6 than during the first period. While the data of the 1999 survey yields positive and significant values for all estimates of our taskmeasures, they drop dramatically in 2 0 0 6 . The measures for routine cognitive (AFL3) and routine manual ( A F L 4 ) tasks become even negative, though not being significant anymore. The objective measure for job complexity (NJC) also seems to have a negative but not significant effect on wages, whereas the positive coefficient of subjective job complexity (SJC) increases over time, though not significantly so. Low-skilled and high-skilled 25 workers are likely to experience quite different life-cycle profiles in wages. Results in Antonczyk (2007), Fitzenberger (1999), and Dustmann et al. (2007) support this hypothesis. Therefore, we interact our indicator variables for lowskilled and high-skilled workers with the variables age and agel. The parameter estimates for these interaction terms show significant differences between medium-skilled and high-skilled workers whereas the age profile do not differ significantly between lowskilled and medium-skilled workers. The age profiles for high-skilled workers prove steeper than those for the other two skill groups. The results indicate that age shows less of a wage smoothing effect in 2 0 0 6 than in 1999. Moreover the estimates show that being low-skilled or high-skilled in 2 0 0 6 results in a higher wage dispersion in 2 0 0 6 than in 1999, which is in line with what was shown by the unconditional descriptive statistics. Finally the parameter estimates for the measures of job complexity indicate that wage dispersion is positively correlated with the number of tasks carried out and this effect is rather stable for the two observed years. Subjective job 24 25

The category non-routine manual is the left-out category. Medium-skilled workers are our left-out category.

Can a Task-Based Approach Explain the Recent Changes in the German Wage Structure? • 233

Table 10 Estimated Model 1999: AFL-TaskIndex without occupations Parameter for cond. expectation

Estimate (Standard Error)

= (n + l)(2n + 1)

W

i

w

w

'=2+2^>

C_ W

=

a

njc/a)] '

'

=

W

. t =

,.„. 1

'-'n'

(10)

w

2 +2[{n+l)VAR{c)/c

+ cY

c where c = j denotes the average labour-input verse labour productivity in the industry and VAR(c) sure of the industry-wide dispersion of c,. Proof. See the Appendix.

fi 1 \ '

[

coefficient, that is the average in= \ JZ'L l ici ~ c)2 represents a mea-

Proposition 1 provides a generalisation of a variety of results t h a t have already been derived for a h o m o g e n e o u s oligopoly. 5 Eqs. (9) and (10) show t h a t the firm-specific w a g e outcomes are a decreasing function of the firm-specific labour-input coefficients C;, if 5

See e.g. Corneo (1995) among others, w h o derives expressions for wf and wf under the assumption q = 1 for all t. Moreover, our analysis generalises the results of Haucap and Wey (2004), w h o consider the case n = 2, c\ = (\ — d) and ci = 1.

Firm Heterogeneity and Wages under Different Bargaining Regimes • 245

wages are determined in the decentralised and intermediate centralised wage-setting regime. The reason is that the union's marginal cost of a wage increase, dlj/diVj, unambiguously increases with c,. T h a t is the higher the labour-input coefficient the larger is the incentive to lower the firm-specific wage w, in order to improve firm's i competitive position in the product market. Conversely, if c, decreases, this induces unions in regimes (D) and (/) to settle for a higher wage as the marginal cost of a wage increase in terms of foregone employment is reduced. Moreover, in the decentralised regime (D) the firm-specific wage is the higher the lower firm's i labour-input coefficient c, relative to the industry average, c. The reason is that in the decentralised case unions generally have an incentive to cut wages in order to gain a larger share of industry employment. A low average industry productivity lowers this incentive by reducing the competitive pressure on firm i, thereby enabling its union to settle for a higher wage. N o t e that this is not the case in the intermediate centralised regime (I), where the wage is solely a function of each firm's own labour-input coefficient c;. The reason is that the competitive mechanism being at work in the decentralised regime completely disappears with an industry-wide union, which fully internalises positive externalities arising from wage increases in firm i for the employment level in the rival firms j, j ^ i. Finally, from eq. (11) it can be seen that the completely centralised regime (C) suppresses any wage response to firm-specific productivity conditions, which simply arises from the assumption that the uniform industry wage applies to all firms in the industry. Instead, the uniform industry-wage is shown to be a function of the average industry labour-input coefficient c and the variability in productivity conditions, as measured by the industrywide dispersion of the inverse labour productivity c,, VAR (c) N o t e that in a homogeneous industry with all firms exhibiting an identical labour-input-coefficient c, the uniform industry wage reduces to ^

= f

+

(.2,

Compared to the wage outcome in a homogeneous industry with all firms exhibiting an identical labour-input-coefficient c, an industry union in a heterogeneous industry therefore settles for a lower wage, since VAR(c) > 0. T h e intuition behind this result is that an industry-union setting a uniform industry-wage takes into account the marginal cost of a wage increase for all firms in the industry, that is also for those firms which have a labourinput coefficient above the average. Employment in those firms is affected more than proportionally negatively after a given wage increase. The reason is that a wage increase does not only reduce the output level to a larger extent, but also implies for a given output reduction a higher employment loss (since /, = c, jS fN f I ~ S 10 s> 3 ==£o

-= O in . 4-* ^o ai in a\ oo •K 5 ifloq^oo I O r n o i O rri ^ lo ÎOOrrM fs f Q_ o o o o o i ~ s s in m v> r-, í m o tí tom in oo\(NiO oo o ko oo 8 o oooo 1 1 1 1 —ma\ IV 00 m lOJfc t T iv m m in T — 00T m ovq o IV rv o - •3 oiOIv•o. - m (Nin IN vo a> ^ m oo r- in vo 0mrn 01a\ 5 oo in ui « vTotNIN0 rri ni rri 1 ? OinOooV-oT-rv^f 1 1 1 cLo d o d o i ¥ T —(N (N o I t •0s0f Iin O ^rn mm o'oo r- m 00 IN ui rt

XQ —1 js • ii mi 2? I 00 c s vi X .t! .t! go** 4) : u b c cD Ic O ) i/i X™ULIxLUx

2 8 4 • J. Möller and A . Tubadji

Table 3 Long-run Effects on Regional Employment and W a g e Bill (Dynamic Panel Estimates, 323 W e s t German Regions 1 9 8 5 - 2 0 0 4 ) Variable In W a g e Other Creative Class (OCC) High Skilled (HS) Firm Size (FS) Variable In W a g e Other Creative Class (OCC) High Skilled (HS) Firm Size (FS)

Long-run effect on regional employment ( x 100) Specification 2 Specification 1 Specification 3 1 -step 2-step 1 -step 2-step 1 -step 2-step 3.689 13.751 -

0.466

3.767 13.945 -

0.342

3.126 -

7.218 0.375

4.971 -

10.835 0.223

1.963 8.500

2.141 11.026

- 0.680 0.221

- 2.053 0.160

Long-run effect on regional wage bill ( x 100) Specification 1 Specification 2 Specification 3 2-step 1-step 1-step 2-step 1-step 2-step 3.556 17.038 -

0.618

4.485 15.577 -

0.451

3.229 -

9.508 0.520

5.435 -

10.822 0.306

2.585 9.888

3.206 9.870

-0.415 0.276

- 0.456 0.238

Notes: Long-run effects calculated from Table 2.

second and third specification. Note that the null is rejected especially for both variants of the Hansen test. Also under this aspect, the first specification is clearly preferable. In Table 2 b we repeat the analysis for the log of the wage bill as the dependent variable. Qualitatively all results are very similar to those of Table 2 a . Again, there is a high degree of sluggishness in the dependent variable. Also here the share of the Creative Class performs as a better measure for predicting local development. Hence the results are robust with respect to the choice of the regional economic performance indicator. Table 3 shows the implied long-run effects on regional employment and the wage bill, respectively. The results show that a 1 0 percent higher regional wage would increase employment (through migration and higher participation) by between 0 . 2 and 0 . 5 percent. An increase of the share of the Creative Class by 10 percentage points would increase regional employment by between 0.8 to 1.3 percent. This is higher than the longrun effect initiated by increasing the share of high-skilled workers. Finally, doubling average firm size would lead to 0 . 1 to 0 . 5 percent higher employment. Interestingly, the long-run effects for the wage bill are qualitatively and quantitatively rather similar. For investigating the question " w h o is attracting the Creative C l a s s ? " the dynamic panel method is used again. Here we employed the Blundell-Bond ( 1 9 9 6 ) system estimator, which - according to test statistics - seems to be more appropriate in this case. As the results of the one-step and two-step estimation methods are quite similar, Table 4 presents the results for the latter only. We first tested the hypothesis that Bohemians attract the Creative Class and then included the share of high-skilled workers as an alternative. This is done in two variants (using log employment or the log wage bill as an indictor for regional economic performance). All variants pass the test statistics with respect to the adequacy of (subsets of) instruments except for the Sargan test. There is no indication for serial correlation of order 2.

The Creative Class, Bohemians and Local Labor Market Performance • 285

Table 4: Alternative Specifications for Explaining the Concentration of the Creative Class (2-Step System G M M Dynamic Panel Estimates, 323 West German Regions, 1985-2004) Variable

coeff.

t-stat (1)

const. Creative L1 Bohemians Bohemians L1 High Skilled High Skilled L1 In Empl. In Empl.LI In Wage Bill In Wage Bill L1 In Firm Size In Firm Size L1 ShareFemales Share Fem. L1

-0.001 1.070 -0.211 0.306 -

0.016 -0.016 -

-0.002 0.002 -0.109 0.106

test-stat. F-Stat. 903.12 Arellano-Bond (AR2) -1.71 Sargan Overid.Test 115.51 Hansen Overid.Test 76.8 GMM instruments (L) 57.15 iv instruments (L) 19.65 CMM instruments (D) 55.91 iv instruments (D) 20.89 # of instruments

const. A Creative L1 A Bohemians A High Skilled A In Empl. A In Wage Bill A In Firm Size A ShareFemales

87

t-stat (2)

coeff.

t-stat

coeff.

t-stat

(3)

(4)

Specification 1 Dependent variable Other Creatives (OCC) 0.000 -0.09 -0.004 -0.52 0.08 --0.002 1.067 67.52 0.899 37.02 0.900 74.13 -0.89 -0.257 -1.12 1.44 1.26 0.343 0.485 4.40 0.463 -0.371 -3.29 --0.351 0.023 3.33 2.05 -0.023 -3.34 -2.05 2.07 0.015 0.021 -0.015 -2.08 -0.021 -0.64 -0.001 -0.48 -0.005 -1.99 --0.005 0.73 0.005 2.22 0.88 0.002 0.005 -0.087 -1.98 -2.40 -0.080 -1.64 --0.069 2.60 0.080 1.82 1.87 0.076 0.059

3.44 -3.45 -1.90 2.14 -1.53 1.44

Test statistics p.-val. test-stat. 0.000 924.9 0.089 -1.42 0.000 100.55 0.073 73.54 0.068 47.7 0.320 25.85 0.156 53.51 0.107 20.03

p.-val. 0.000 0.148 0.000 0.100 0.286 0.068 0.226 0.091

p.-val. test-stat. 0.000 812.96 0.087 -1.70 0.000 128.91 0.150 82.23 0.106 59.85 0.480 22.38 0.231 58.98 0.183 23.25 87

p.-val. 0.000 0.156 0.003 0.219 0.364 0.171 0.305 0.219

87

test-stat. 853.17 -1.45 114.91 80.00 49.88 30.13 56.09 23.91

-0.38 36.86 -

4.31 -3.22 -

87

Specification 2 Dependent variable first difference Other Creatives (A OCC) 0.009 5.33 0.006 3.32 -0.003 -1.33 -0.005 -2.30 0.161 3.32 0.149 3.20 0.051 1.25 0.022 0.52 -0.085 - 0 . 2 1 0.003 0.01 0.637 6.14 0.650 6.11 0.029 3.10 0.030 4.09 0.031 3.02 0.028 3.35 -0.001 -0.22 -0.001 -0.23 -0.004 -1.56 -0.004 - 1 . 6 2 -0.009 -0.15 -0.062 -1.02 -0.040 -0.78 -0.029 -0.54

test-stat. F-Stat. 7.68 Arellano-Bond (AR2) -0.52 Sargan Overid.Test 81.91 Hansen Overid.Test 58.38 GMM instruments (L) 43.25 iv instruments (L) 15.13 GMM instruments (D) 26.09 iv instruments (D) 32.29 # of instruments

coeff.

61

p.-val. test-stat. 0.000 7.61 0.604 -0.57 0.000 88.28 0.072 61.44 0.043 46.18 0.442 15.27 0.621 29.54 0.006 31.91 61

Test statistics p.-val. test-stat. 0.000 23.68 0.568 -1.55 0.000 58.74 0.042 49.22 0.023 39.01 0.432 10.2 0.437 30.5 0.007 18.72 61

p.-val. 0.000 0.122 0.068 0.272 0.101 0.807 0.389 0.227

test-stat. 20.17 -1.84 64.27 56.52 42.26 14.27 33.05 23.47

p.-val. 0.000 0.066 0.025 0.098 0.053 0.506 0.276 0.075

61

Notes: The estimated equation includes dummies for region types and years of observation; other notes see Table 2.

286 • J. Moller and A. Tubadji

We find high inertia in the regional distribution of the Creative Class, i.e. a coefficient of the lagged endogenous variable close to unity (see specification 1 in Table 4). 1 2 Note that in the equations using Bohemians as an explanatory variable this coefficient even exceeds unity. The general results are not in favor of Florida's assumptions in this respect. The coefficients for the contemporaneous and lagged Bohemian variable are not significant, whereas this is clearly the case if the share of high skilled is used instead. Moreover, our estimates suggest that employment and the wage bill do affect the regional concentration of the Creative Class. Hence creative persons seem to be concerned with regional economic conditions, an assumption which is at least to some extent questioned by Florida's theory. A closer inspection of the estimated coefficients in specification 1 reveals a clear pattern: whereas the coefficient of the lagged endogenous is close to unity, the coefficients of the contemporaneous and lagged explanatory variables are similar in absolute value but of different sign. This pattern suggests a re-specification in first differences. The outcome is shown in the lower panel of Table 4 (specification 2). The results confirm our interpretation given so far. The variants using the share of high skilled clearly outperform the variants using the share of Bohemians as an attractor for the Creative Class. According to the t-statistics only the share of high skilled - but not the share of Bohemians - seem to have an effect on the Creative Class. Moreover, the effect of the economic performance indicators is highly significant in all cases, whereas the firm size and the share of females (both as indicators for regional economic structure) do no play a prominent role. Note that also the test statistics are clearly more favorable if the share of high skilled is used instead of the share of Bohemians. All in all, the results in Table 4 are at odds with the corresponding-assertions in Florida's work.

5

Conclusions

Richard Florida's thought-provoking concept of the Creative Class can be seen a fruitful contribution for our understanding of regional economic development because it stresses the importance of professional activities and the potential role of the cultural milieu for attracting knowledge carriers and innovative people to a location. However, previous attempts to corroborate the basic pillars of Florida's theory typically suffer from serious deficiencies. Correlation does not imply a causal relationship and reverse causality might be an important issue in the context of regional development. As the issue of endogeneity of the regressors is mostly neglected in other studies, modern empirical techniques are required to obtain a deeper look at the phenomena. The present paper aims at scrutinizing two basic hypotheses of Richard Florida's concept of the Creative Class. The first is that the regional concentration of the Creative Class entails better economic performance as measured by employment growth or an increasing wage bill. Moreover, the Creative Class concept should outperform "traditional" indicators of human capital such as the share of high-skilled workers in the regional labor force. Using a large micro data set for West Germany for the observation period 1975 to 2004 containing information on professional activities, we are able to collect panel data for 323 NUTS 3 regions. Indeed, our results indicate that Florida's classification scheme

12

Here lags of order higher than 1 for the endogenous and explanatory variables were not significant.

The Creative Class, Bohemians and Local Labor Market Performance • 287

for creative people seems have remarkable explanatory power for regional economic performance. O n the basis of dynamic panel estimation we find evidence for the Creative Class playing an important role in regional economic development. In addition, the concept of measuring regional innovative capabilities by counting high-skilled persons seems to be less adequate when it comes to identify the growth potential of a region. Our econometric investigation confirms the first part of Florida's story. The empirical findings, however, are at odds with the second part. According to Florida; the Creative Class has a taste for a liberal cultural milieu which is typically indicated by a regional concentration of Bohemians, whereas favorable economic conditions do not play a major role. For German data we cannot support this view. There is no evidence for the Creative Class following the Bohemians. By contrast, we find some support for the hypothesis that creative workers prefer living in economically prosperous regions. Moreover, the concentration of other high-skilled people seems to matter more than the concentration of Bohemians. Therefore, we are skeptical vis-à-vis a simplistic adaption of Florida's concept by local policy makers true to the motto "Let's create a liberal cultural scene; this will attract creative people and the region becomes an economic hot spot". Regional economic development seems to be somewhat more complex.

288 • J. Möller and A. Tubadji

Appendix Table A1 The Creative Occupations Florida's Definition Components writers and creative or performing artists

photographers and image and sound recording equipment operators; fashion and other models artistic, entertainment, and sports associate professionals scientists, think-thank researchers

engineers

university professors editors Analysts, entrepreneurs, leading administrators opinion makers software programmers/engineers Gardening Architects

IAB Database Code Bohemians 821: Publizisten 823: 831: 832: 833: 837:

Bibliothekare, Archivare, Museumsfachleute Musiker Darstellende Künstler Bildende Künstler, Grafiker Photographen

835: Künstlerische und zugeordnete Berufe der BühnenBild- und Tontechnik 838: Artisten, Berufssportler, künstlerische Hilfsberufe Other Creative Core 881: Wirtschafts- und Sozialwissenschaftler, a.n.g., Statistiker 882: Geisteswissenschaftler, a.n.g. 883: Naturwissenschaftler, a.n.g. 601: Ingenieure des Maschinen- und Fahrzeugbaues 602: Elektroingenieure 603: Architekten, Bauingenieure 604: Vermessungsingenieure 605: Bergbau-, Hütten-, Gießereiingenieure 606: Übrige Fertigungsingenieure 607: Sonstige Ingenieure 611: Chemiker, Chemieingenieure 612: Physiker, Physikingenieure, Mathematiker 871: Hochschullehrer, Dozenten an höheren Fachschulen und Akademien Dispersed in the other categories 751: Unternehmer, Geschäftsführer, Geschäftsbereichsleiter 752: Unternehmensberater, Organisatoren 762: Leitende und administrativ entscheidende Dispersed in the other categories 774: Datenverarbeitungsfachleute 52 Gartenarchitekten, Gartenverwalter

The Creative Class, Bohemians and Local Labor Market Performance • 289

Florida's Definition Components

IAB Database Code

Creative Professionals high-tech sectors services, technicians 621 Maschinenbautechniker 622 Techniker des Elektofaches 623 Bautechniker 624 Vermessungstechniker 625 Bergbau-, Hütten-, Gießereitechniker 626 Chemietechniker, Physikotechniker 627 Übrige Fertigungstechniker 628 Sonstige Techniker 629 Industriemeister, Werkmeister 631 Biologisch-technische Sonderfachkräfte 632 Physikalisch- und mathematisch-technische 633 Chemielaboranten 634 Photolaboranten 635 Technische Zeichner 691 Bankfachleute financial services 753 Wirtschaftsprüfer, Steuerberater 813 Rechtsvertreter, -berater legal services 703 Werbefachleute business services 822 Dolmetscher, Übersetzer Alternative Classifications

IAB Database Code

Mathematics, Engineering, Natural Science, Technics engineers and technicians 601 Ingenieure des Maschinen- und Fahrzeugbaues 602 Elektroingenieure 603 Architekten, Bauingenieure 604 Vermessungsingenieure 605 Bergbau-, Hütten-, Gießereiingenieure 606 Übrige Fertigungsingenieure 607 Sonstige Ingenieure 611 Chemiker, Chemieingenieure mathematicians and natural scientists 612 Physiker, Physikingenieure, Mathematiker 883 Naturwissenschaftler, a.n.g. cultural figures

humanities

Humanities, Culture 821 Publizisten 831 Musiker 832 Darstellende Künstler 833 Bildende Künstler, Graphiker 835 Künstlerische und zugeordnete Berufe der Bühnen-, 837 Photographen 882 Geisteswissenschaftler, a.n.g.

References Arellano, M., S. Bond (1991), Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Review of Economic Studies 58: 277-97. Arellano, M., O. Bover (1995), Another look at the instrumental variables estimation of error components models. Journal of Econometrics 68: 29-51. [ Belot, M., S. Ederveen (2005), Indicators of Cultural and Institutional Barriers in OECD Countries, CPB Memorandum, The Hague.

290 • J. Möller and A. Tubadji

Bender, S., A. Haas (2002), Die Beschäftigtenstichprobe. In: G. Kleinhenz (Hrsg.), IAB-Kompendium Arbeitsmarkt- und Berufsforschung (Beiträge zur Arbeitsmarkt- und Berufsforschung 250), Nürnberg: 3-12. Blien, U., J. Südekum, K. Wolf, (2006), Local Employment Growth in West Germany: A Dynamic Panel Approach. Labour Economics 13: 445-458. Blundell, R., S. Bond (1998), Initial conditions and moment restrictions in dynamic panel data models. Journal of Econometrics 87: 11-143. Bond, S. (2002), Dynamic panel data models: A guide to micro data methods and practice. Working Paper 09/02. Institute for Fiscal Studies. London. Boschma, R., M. Fritsch (2007), Creative Class and Regional Growth - Empirical Evidence from Eight European Countries. Jena Economic Research Papers 066. Clark, T. (2002), Urban Amenities: Lakes, Opera, and Juice Bars. Do They Drive Development. The University of Chicago Working Paper and prepared as a chapter for The City as an Entertainment Machine, Research in Urban Policy 9, New York: JAI Press/Elsevier in progress. Daly, A. (2004), Richard Florida's High-Class Glasses. Grantmakers in the Arts Reader, Summer. Damelang, A., M. Steinhardt, S. Stiller (2007), Europe's Diverse Labour Force - The Case of German Cities. EURODIV Paper 49.2007 Dreher, C. (2002), Be Creative - of Die, Salon Publication, Salon.com Archive (06.06.2002) http://dir.salon.com/story/books/int/2002/06/06/florida/index.htm l?source=search&Caim=/ books/int. Florida, R. (2002a), The Rise of the Creative Class: And how it's transforming work, leisure, community, and everyday life. New York: Basic Books. Florida, R. (2002b), Bohemia and Economic Geography. Journal of Economic Geography 2: 5 5 71. Florida, R. (2004), Revenge of the Squelchers. The Next American City Journal: June. Florida, R. (2005), The Flight of the Creative Class: The new global competition for talent. London: Harper Collins. Florida, R., G. Gates (2001), Technology and Tolerance: The Importance of Diversity to HighTechnology Growth. The Brookings Institution, Survey Series, June. Fritsch, M., M. Stuetzer (2008), The Geography of Creative People in Germany. Forthcoming in: International Journal of Foresight and Innovation Policy. Gautier, P., M. Svarger, C. Teulings (2005), Marriage and the City. Centre for Economic Policy Research Discussion Paper 4939. Glaeser, E. (2005), Review of Richard Florida's The Rise of the Creative Class. Regional Science and Urban Economics 35: 593-596. Glaeser, E., J. Kolko, A. Saiz (2001), Consumer City. Journal of Economic Geography 1: 2 7 - 5 0 . Hayashi, F. (2000), Econometrics. 1 st ed. Princeton, NJ: Princeton University Press. Holtz-Eakin, D., W. Newey, H.S. Rosen (1988), Estimating vector autoregressions with panel data. Econometrica 56: 1371-95. Heilbrun, J. (1992), Art and culture as central place functions. Urban Studies 29(2): 205-215. Heilbrun, J. (1996), Growth, accessibility and the distribution of arts activity in the United States: 980 to 1990. Journal of Cultural Economics 20(4): 283-296. Hunt, G., R. Mueller (2004), North American Migration: Returns to Skill, Border Effects, And Mobility Costs. The Review of Economics and Statistics 86: 988-1007. Jacobs, J. (1961), Death and Life of Great American Cities. New York: Random House. Johnson, R., A. Onwuegbuzie, L. Turner (2007), Towards a Definition of Mixed Methods Research, Journal of Mixed Methos Research. SAGE Publications 1(2): 112-133. Kotkin, J. (2004), The Capital of What. The New York Sun, February 19, 2004. Kotkin, J. (2005), Uncool Cities. Prospect Magazine, Issue 115. Knudsen, B., R. Florida, K. Stolarick (2005), Beyond Spillovers: The Effects of Creative-Density on Innovation. Rotman School of Management, University of Toronto and The Martin Prosperity Institute. Knudsen, B., R. Florida, G. Gates, K. Stolarick (2007), Urban Density, Creativity, and Innovation, Working Paper http://creativeclass.com/rfcgdb/articles/Urban_Density_Creativity_and_Innovation.pdf.

The Creative Class, Bohemians and Local Labor Market Performance • 291

Landry, C. (2000), The Creative City: A Toolkit for Urban Innovators London. Earthscan Publications Ltd. Lee, S., R. Florida, Z . Acs (2004), Creativity and Entrepreneurship: A Regional Analysis of N e w Firm Formation. Regional Studies 38(8): 8 7 9 - 8 9 1 . Malanga, S. (2004), The Curse of the Creative Class. The City Journal: Winter. M a r k u s e n , A. (2006), Urban development and the politics of a Creative Class: evidence from the study of artists. Environment and Planning A 38(10): 1 9 2 1 - 1 9 4 0 . M a r k u s e n , A., A. Johnson (2006), Artists' Centers: Evolution and Impact on Careers, Neighborhoods and Economies. Minneapolis: H u b e r t H . H u m p h r e y Institute of Public Affairs: University of Minnesota. Mattsson, H . (2007), Mobile Talent or Privileged Sites? M a k i n g Sense of Biotech Knowledge Worker Mobility and Performance in Sweden, Social Geography 2: 1 1 5 - 1 2 3 . M c G r a n a h a n , D., T. Wojan (2007), Recasting the Creative Class to Examine Growth Processes in Rural and Urban Counties. Regional Studies 41(2): 1 9 7 - 2 1 6 . Mellander, C., R. Florida (2006), The Creative Class or H u m a n Capital? Explaining Regional Development in Sweden, CESIS Electronic Working Paper N o 79. http://www.infra.kth.se/ cesis/documents/WP79.pdf (2007-10-10). Möller, J., Aldashev, A. (2007), Wage Inequality, Reservation Wages and Labor M a r k e t Participation - Testing the Implications of a Search-Theoretical Model with Regional Data. International Regional Science Review 30(2): 1 2 0 - 1 5 1 . N a t h a n , M . (2005), The Wrong Stuff, Creative Class Theory, Diversity and City Performance. Centre for Cities, Discussion Paper no. 1, September. Nickell, S. (1981), Biases in dynamic models with fixed effects. Econometrica 49(6): 1 4 1 7 - 2 6 . Rauch, J. (1993), Productivity Gains from Geographic Concentration of H u m a n Capital: Evidence from the Cities. Journal of Urban Economics 34: 3 8 0 - 4 0 0 . Rausch, S., C. Negrey (2006), Does the Creative Engine Run? A Consideration of the Effect of Creative Class on Economic Strength and G r o w t h . Journal of Urban Affairs 28(5): 473—489. R o b a c k , J . (1982), Wages, Rent and Quality of Life. Journal of Political Economy 9 0 : 1 2 5 7 - 1 2 7 8 . R o o d m a n , D. (2006), H o w to D o xtabond2: An Introduction to "Difference" and "System" G M M in Stata. Working Paper N u m b e r 103, December. Shapiro, J. (2005), Smart Cities: Quality of Life, Productivity, and the Growth Effects of H u m a n Capital. NBER W 1 1 6 1 5 . Simon, C. (1998), H u m a n Capital and Metropolitan Employment Growth. Journal of Urban Economics: 43: 2 2 3 - 2 4 3 . Simon, C., C. Nardinelli (2002), H u m a n capital and the rise of American cities, 1 9 0 0 - 1 9 9 0 . Regional Science and Urban Economics 32(1): 5 9 - 9 6 . Simon, C. (2004), Industrial reallocation across US cities 1 9 7 7 - 1 9 9 7 . Journal of Urban Economics 56: 1 1 9 - 1 4 3 . Scott, A. (2006), Creative Cities: Conceptual Issues and Policy Questions. Journal of Urban Affairs, 28(1): 1 - 1 7 . Turok, I. (2006), The Distinctive City: 'Quality' as a Source of Competitive Advantage. Unpublished paper, University of Glasgow. Windmeijer, F. (2005), A finite sample correction for the variance of linear efficient two-step G M M estimators. Journal of Econometrics 126: 2 5 - 5 1 . Wojan, T., D. Lambert, D. M c G r a n a h a n (2007), Emoting with Their Feet: Bohemian Attraction to Creative Milieu. Journal of Economic Geography: 1 - 2 6 . Prof. Dr. Joachim Möller, Institute for Employment Research (IAB), IZA and University of Regensburg, Universitätsstraße 31, 9 3 0 5 3 Regensburg, Germany. E-Mail: [email protected] Annie Tubadji, Institute for Employment Research (IAB), Regensburger Straße 104, 90478 N ü r n b e r g , Germany. E-Mail: [email protected]

Jahrbücher f. Nationalökonomie u. Statistik (Lucius & Lucius, Stuttgart 2009) Bd. (Vol.) 229/2+3

Spatial Implications of Minimum Wages By Thiess Buettner and Alexander Ebertz, Munich* JEL J6, R12, J3 Minimum wages, urban poverty, spatial wage structure, mobility, economies of agglomeration.

Summary This paper addresses possible consequences of a minimum wage in a spatial context. An empirical analysis utilizing German data shows that a significant spatial wage structure exists and that, as a consequence, the share of workers earning wages below a minimum wage will be particularly high in rural counties even if we control for educational and occupational differences. A theoretical analysis discusses the implications for the spatial structure of the economy and shows that while the wages in the countryside will be affected positively, wages will decline in the city, where employment and population rise. Workers in the city will further suffer from an increase in housing costs. This supports concerns that urban poverty might increase as a result of the introduction of a minimum wage.

1

Introduction

F o r the policy maker minimum wages are an attractive policy t o o l . M i n i m u m wages are apparently targeted at the heart o f the poverty p r o b l e m , the motivation to fight poverty earns public respect, and the direct costs involved seem low. In fact, the evidence suggests that minimum wages do have an impact on the wage distribution raising the earnings o f those that are at the b o t t o m o f the wage distribution. Opponents argue that minimum wages also have important adverse effects on employment. T h u s , a controversial debate about the adverse consequences o f minimum wages on employment consumes a lot o f space in an empirical literature that employs sophisticated micro-level datasets and advanced e c o n o m e t r i c techniques t o show that minimum wages have or have not adverse effects on employment (e.g. see Card/Krueger 1 9 9 4 , 2 0 0 0 , Neumark/Wascher 2 0 0 0 , Brown 1999). In this paper we argue that it is generally overlooked that wage increases and adverse employment effects resulting from minimum wages are systematically different for different groups o f workers. This is already indicated by the experience with minimum wages in Germany. W h i l e the minimum wage in the construction sector shows quite limited effects in the western part o f the country it exerts rather strong adverse effects in the eastern part (e.g., Moeller/Koenig 2 0 0 8 a ) . Moreover, given the wage differential between East and West, Ragnitz and T h u m ( 2 0 0 7 ) show that a federal minimum wage would mainly be binding in the eastern part of the country.

* We thank the editor and t w o anonymous referees for helpful comments and Jens Ruhose for excellent research assistance.

Spatial Implications o f M i n i m u m W a g e s • 293

It is important to note that these asymmetries are built in, however, by the same striking simplicity of the concept that so much appeals to the policy maker: the minimum wage simply disregards all sorts of wage structures that may exist, including not only wage differences associated with skills, occupation, experience, and sex, but also differences with regard to industry, firm-size, and region. While the ignorance of these differences seems to be a necessary consequence of a social policy that is committed to combat poverty, all of these differences play a role in the economic consequences of minimum wages and, hence, are important for the effectiveness of minimum wages in reducing poverty. An important dimension of the wage structure in this regard is the spatial wage structure that shows up in higher wages in urban agglomerations as compared to rural areas. This paper argues that if there is a uniform minimum wage imposed on cities and rural towns alike, we can expect that the minimum wage is much more restrictive in the countryside but might be rather ineffective for people working in the cities. Hence, the wages of workers that live in the cities might not benefit much from minimum wages. In fact, using the German example, we present some empirical evidence below showing that the share of workers earning wages below a minimum wage would be much higher in rural as compared to urban areas. While this difference might be explained by the different composition of the work force, we provide further evidence that the regional differences in the incidence of minimum wages are mainly driven by a spatial wage structure that is associated with differences in density even if we control for differences in education and occupation. Based on these empirical findings we explore the consequences of an introduction of a uniform minimum wage in a stylized theoretical model that derives a spatial wage distribution in a migration equilibrium setting with productivity differences and housing costs. The analysis shows that imposing uniform minimum wages exerts distortive effects on the spatial structure of the economy. More specifically, we find that employment and population will rise in the more densely populated regions implying that wages of the working population in the cities might even fall. Moreover, the city population would also suffer from an increase in housing costs. This asymmetric impact is important since there is a close association between poverty and urbanization. 1 Thus, our findings support concerns that urban poverty might increase as a result of the introduction of a uniform minimum wage. The paper is organized as follows. The first part is concerned with spatial differences in the extent to which the minimum wage is binding. Section 2 provides some basic empirical evidence about these spatial differences in the incidence of minimum wages in Germany. Section 3 provides some further evidence about the spatial wage structure that gives rise to these systematic differences. The second part of the paper provides a theoretical analysis of the consequences of these spatial differences in the incidence of minimum wages. Section 4 first lays out a stylized theoretical model that shows how a spatial wage distribution emerges in the migration equilibrium setting with productivity differences and housing costs. Subsequently, minimum wages are introduced and we discuss the consequences. Section 5 provides our conclusions.

1

In the German case the poverty rate in the cities is almost twice as large as the poverty rate of rural counties: in 2004, the poverty rate in core cities has been 5.11 % compared with a figure of rural counties of 2.89 % (Source: German States' Statistical Offices).

294 • T. Buettner and A. Ebertz

2

Spatial differences in the incidence of minimum w a g e s

There is an ongoing political debate in Germany about the economy-wide introduction of minimum wages. In 1997 a minimum wage of D M 16 (€ 8.18) for West Germany (DM 15.14 ( € 7 . 7 4 ) for East Germany) has been introduced in the construction sector (see König/Möller 2008b). Current political proposals for the uniform minimum wage by some of the unions and by the Social-Democratic Party point at levels of € 6.50 or even € 7.50. In the following, we investigate the spatial patterns of the incidence of an introduction of corresponding minimum wages for the case of Germany. We make use of the regional sample of employees (Beschäftigtenstichprobe) of the Institute for Employment Research (IAB), which constitutes a two percent random sample of all German employees subject to social security contributions and provides figures on employment status, wages, and personal characteristics like age, education, and profession of the sampled individuals (for a detailed description of the data see Drews 2008). Since the data refer to the place of work at the county level, 2 this dataset is well suited to provide evidence on the spatial structure of wages in Germany. For our purpose of illustrating possible spatial consequences of minimum wages we focus only on the latest year available, 2 0 0 4 . Furthermore, we include only full-time employed individuals aged between 16 and 62. 3 Figure 1 illustrates the spatial differences in the minimum wage incidence, i.e. the average percentage of employees affected by a minimum wage at the level of counties and cities for West Germany and East Germany, respectively. Note that we include the top-coded observations when drawing percentiles from the wage distribution. As our data refer to daily wages but not to hourly wages and no information is provided about hours of work, we rely on a percentile of the wage distribution for full employed workers rather than directly applying a minimum wage. More specifically, we rely on the analysis of Ragnitz and Thum (2007) who found that a minimum wage of € 6.50 (7.50) corresponds to the 8.50 (11.30) percentile of the wage distribution in West Germany and to the 18.10 (26.00) percentile in East Germany. Ragnitz and Thum are using microdata from the survey on the salary and wage structure in the manufacturing and service sectors that have been issued by Federal Statistical Office in 2 0 0 7 . While this data refers to 2 0 0 1 our analysis focuses on 2 0 0 4 . Since the wage distribution might have changed over time, more recent data might result in different percentiles. However, our focus is not so much on the actual share of workers with wages below a minimum wage of € 6.50 or € 7.50. Rather we are interested in the spatial differences in the minimum wage incidence, regardless of the actual level. A first inspection seems to confirm that some of the cities, like Hamburg, Berlin, Cologne, or Munich, are visibly less affected by a minimum wage of € 7.50 than their less densely populated neighbor regions. Further visualization of the spatial dimension of

2 3

For reasons of privacy protection, some counties are aggregated into a region. Due to changes in individual employment status, employer, etc., for some of the sampled individuals several, possibly also simultaneous, spells are reported within one year, with the wage level possibly differing among different spells. In order not to overstate the incidence of a minimum wage in Germany, we include the highest respective wage reported for each individual worker in our analysis. To check for possible problems with simultaneous spells we conducted alternative analyses excluding all observations with a daily wage below € 4 0 to ensure that the results are not driven by such possibly defective observations. However, all results are unaffected qualitatively, and even quantitatively only minor changes were found.

Spatial Implications of Minimum Wages • 295 MW Incidence 7,50 €, in %

Figure 1 Incidence of Minimum Wages Percentage of employment spells with a wage below € 7.50 in East and West Germany. the minimum wage incidence is provided by Figure 2, which shows the average population density and the average percentage of employees affected by a minimum wage for five county types. The classification of county types is based on the typology given by the Federal Bureau of Regional Planning (Bundesamt fur Bauwesen und Raumordnung).4

4

We m o d i f y the existing classification such t h a t counties are classified a c c o r d i n g to their o w n characteristics, ignoring the dimension of the general level of agglomeration of their s u r r o u n d i n g area, t h a t is contained in the original classification. M o r e precisely, o u r c o u n t y type 1 comprises cities w i t h m o r e t h a n 1 0 0 0 0 0 inhabitants, c o u n t y type 2 c a p t u r e s all counties with density above 3 0 0 i n h a b i t a n t s per s q k m . C o u n t y type 3 refers t o all counties w i t h density a b o v e 150 but below 3 0 0 inhabitants per sqkm. C o u n t y type 4 refers to all counties with density below 150 i n h a b i t a n t s per sqkm. C o u n t y type 5 finally captures rural counties with density below 100 inhabitants per sqkm.

296 • T. Buettner and A. Ebertz

Cities (type 1)

Urban counties (type 2)

Densely pop. (type 3)

Rural (type 5)

Categories 1 0 log pop density (P/km1) O0MW incidence B,50 €(%) i0MWincidenceML7,5O€(%)

Figure 2 Incidence of Minimum Wages by County Type Percentage of employment spells affected by a minimum wage of €6.50 (€7.50) and log of density by county type. Clearly, the share of employees that earn less than the minimum wage is higher, the less densely populated the respective county is. The highest share is found for rural counties where more than 20 % of employees would be subject to a minimum wage of € 7.50. The visual impression is further underpinned by means of regression analysis, where we estimate the relationship between the local minimum wage incidence and the degree of agglomeration. More precisely, the regressions take the form MWpA+ftZ,^,, where MW, denotes the percentage of employees affected by the respective minimum wage at location /, and Zf is a vector of attributes reflecting the degree of agglomeration of region /'. Summary statistics of all variables employed in this study are reported in Table 1. Table 2 reports the results. The first set of regressions, reported in columns (1) and (2), confirms a highly significant negative relationship between the log of the population density and the percentage of workers affected by the minimum wage restriction. Doubling density would be associated with a decrease of the minimum wage incidence by about 1.68 (2.41) percentage points for the € 6.50 (€ 7.50) example. In our second set of regressions (columns (3) and (4)), we replace the density by dummy variables indicating the respective county type. The results clearly show that the minimum wage incidence is higher in less densely populated counties: the rural counties are having the highest coefficient indicating that the share of workers affected by a minimum wage is higher by about 6.19 (8.93) percentage points in rural counties as compared to cities. Columns (5) and (6) provide results that include a dummy variable for counties in East Germany. It shows a strong positive effect confirming the results by Ragnitz and Thum (2007). Of course, since there is a clear difference in terms of population size and density between regions in East and West this dummy captures some part of the spatial variation in density. This explains why the inclusion of this dummy is associated with smaller density effects. However, the qualitative results prove robust. As compared to the cities the

Spatial Implications of Minimum Wages • 297

Table 1 Descriptive Statistics Variable Daily wage Sex (is 1 for male) Age Edu.: No Edu.: Elementary school Edu.: High school Edu.: High school w. prof, training Edu.: College degree Edu.: University degree Prof, status: Simple Laborer Prof, status: Skilled Prof, status: Foreman Prof, status: Employee Prof, status: Home worker East Population density MW incidence in % , €6.50 MW incidence in % , €7.50 Cty. type 1: Cities Cty. type 2: Urban Cty. type 3: Densely Cty. type 4: Densely, rural Cty. type 5: Rural

Obs. 327130 353047 353047 353047 353047 353047 353047 353047 353047 353047 353047 353047 353047 353047 435 435 435 435 435 435 435 435 435

Mean

Std.Dev.

Individual Data 83.1 33.7 .641 .480 40.3 10.6 .338 .131 .687 .464 .011 .103 .224 .053 .207 .045 .262 .074 .202 .402 .239 .427 .017 .128 .543 .498 .000 .019 Regional Data .257 .438 502.4 654.1 12.1 6.62 16.4 8.61 .163 .370 .101 .302 .299 .458 .474 .340 .097 .296

Min

Max

1 0 16 0 0 0 0 0 0 0 0 0 0 0

167 1 62 1 1 1 1 1 1 1 1 1 1 1

0 40.0 4.08 5.08 0 0 0 0 0

1 4010 56.1 61.3 1 1 1 1 1

Sources: IAB Beschaftigtenstichprobe 2004, federal and regional statistical offices, and own calculations. All figures for 2004.

share of employees with wages below the m i n i m u m wage is u p to 2 . 7 (3.9) percentage points higher in rural counties.

3

The spatial wage structure in Germany

T h e descriptive evidence provided so far has not touched upon the issue of w h a t is driving the wage differences that are behind the spatial differences in the incidence of m i n i m u m wages. Yet this is i m p o r t a n t for the economic consequences of m i n i m u m wages. If higher wages are the result of a spatial structure in the wages, such that wages in densely populated areas are systematically higher, the imposition of the m i n i m u m wage might distort the spatial wage structure with consequences for the spatial allocation of production. If, however, higher wages in the cities simply arise f r o m differences in the composition of the labor force in terms of skill a n d occupation, the economic consequences might be rather different. Therefore, this section further explores the sources of the observed spatial differences in the wages. As is discussed in the regional and u r b a n economics literature, differences in productivity give rise to differences in the intensity of land use which is most strikingly reflected in population density. As the largest density is generally observed in u r b a n agglomerations, the discussion a b o u t the spatial wage structure is centered a r o u n d the so-called u r b a n

298 • T. Buettner and A. Ebertz

O in . ^ :

S olf>. «> !

0

*

rsi 00 t
« * u 00 c 0) (N T3

£

^ SS m & vo

in So ÍN in CO in o IV

"g £ " « s" - 1).e w^i !n^ ifl ^x ^.rs m ¿f in P-C.Cc — >_t —i —J ^ § 5 SP .5? ra o c o-íü H i U D /t> /l> t/1 £t/1CLQ.Q.Q^ X 0J Oí 3 3 3 —+ •*— v^- M^- Q . . . . 1} 3H 2 y y - u -o -o t. i- ^ -M -M +-» -M írt LO tij). To see why, note that if population would be the same (tij = «,) also housing costs would be equal. Hence, private consumption would have to be higher x, > xr With more consumption x, and the same housing costs, however, utility would be higher in i such that the migration equilibrium is disturbed. Hence, the population size in region i would have to be larger. The additional labor supply would result in a decline in the marginal productivity of labor and in higher total cost of housing until utility is equalized across regions. Since housing costs are increasing with population size, and since wages will compensate cost differences between regions, wages will differ in spatial equilibrium w, > wr Thus, we can state the following lemma: Lemma 1 (City Size and Spatial Wage Structure) In the migration equilibrium where utility is equal across regions, locations with higher productivity display a larger population size, a larger urban area, and higher wages. Graphically, as displayed in Figure 4 an increase in productivity in region i relative to region / will result in a different labor demand curve that shows higher wages at the same level of employment. Since in the spatial equilibrium utility and, thus, consumption is equalized across regions, both regions face the same supply curve. As a consequence, the high productivity region will be larger in terms of population and display a higher wage rate than the other.

306 • T. Buettner and A. Ebertz

Figure 4 Regional Productivity Differences with Labor Mobility 4.2 Introducing a minimum wage Now let us consider an economy with several regions, which differ in productivity. Let us rank regions by size ti\ > «2 > •• • > nM- Our analysis implies, so far, that the wage rate will differ in the sense that the region with the higher productivity displays a higher wage rate such that w\ > u>z > ... > U>M- Let us introduce a minimum wage w such that the wage rate of, say, region i is higher, whereas the wage rate in region / = i + 1 is lower. To see how this affects the spatial distribution of activities, consider first the lower-productivity region where w > Wj. We can see immediately, that for the marginal product of labor to rise, employment will have to decline. Without mobility we would obtain unemployment. With mobility, however, labor will move to more productive regions where the minimum wage is not binding and, hence, employment can be increased. As a consequence, both employment and population decline in the low productivity region. The consequences of the introduction of a uniform minimum wage are illustrated in Figure 5. The initial equilibrium is characterized by levels of employment n, and »/ and in both regions utility is at the same level. With the introduction of the minimum wage, labor market equilibrium in region / requires a decline in employment to a level nj, and with higher wages and smaller housing costs there is an increase of utility in the low productivity region. The high productivity region experiences an increase in employment as well as a decline in wages as marginal productivity declines, and hence, utility decreases. However, migration cannot restore equilibrium since employment cannot be increased at the minimum wage in the low productivity region. Also for the more general case with M regions we can establish the following proposition:

Spatial Implications of Minimum Wages • 307

Figure 5 Spatial Effects of the Minimum Wage Proposition 1 (Agglomeration Effect of the Minimum Wage) Imposing a minimum wage that is binding somewhere within the spatial wage distribution distorts the locational equilibrium such that the population of densely populated regions rises whereas the population of sparsely populated regions declines. Proof: Given labor mobility the consumption level earned by workers in different regions is equalized. Taking account of the labor demand equation this implies Fm(»i) ~ h(»i) = *>(«/) - M«;)Suppose that region ;' is less productive such that nt < n,. Imposing a minimum wage that is binding in region / but not in region i implies that w =

F

>"

("/)•

With this constraint the model is overdetermined and, hence, the population size of region / is no longer determined by the above spatial equilibrium. However, in all regions where the minimum wage is binding, labor productivity is forced to rise implying that labor demand declines. With the total population size given, employment and population size in the unconstrained regions will expand. • Further results can be obtained with regard to spatial price differences. First consider wages. Intuitively, the spatial wage distribution is compressed. To see this, recall that the minimum wage is more likely to bind in the low productivity regions that display lower wages. At the same time, however, the wage level in the more productive regions declines since employment is increased.

308 • T. Buettner and A. Ebertz

Proposition 2 (Spatial Wage Distribution Effect of the Minimum Wage) Imposing a minimum wage that is binding somewhere tion tends to compress the spatial wage structure.

within the spatial wage

distribu-

Proof: We know from Proposition 1, that all regions where the minimum wage is not binding face an employment increase. All regions where the minimum wage is binding experience a decline in employment. Hence, the marginal productivity is increasing in all regions where the wage rate is below the minimum wage, whereas it declines in all regions above the minimum wage. As a consequence, wages in the latter group decline, whereas wages in the former group increase. • With regard to housing costs a different result is obtained. Due to the agglomeration effect, we have a larger demand for space in the large regions that are unconstrained and a reduction of the demand for space in the smaller, constrained regions. Let us state this as our third proposition. Proposition 3 (Location Rent Effect of the Minimum Wage) Imposing a minimum wage that is binding somewhere within the spatial wage distribution tends to raise total housing costs and location rent in the more densely populated regions and to reduce total housing costs and location rent in the less densely populated regions. Proof: We know from Proposition 1, that all regions where the minimum wage is not binding face an increase in population. All regions where the minimum wage is binding, experience a decline in population. Hence, the total housing costs are increasing in the first group but decreasing in the latter. • 4.3 Welfare implications It is tempting to consider welfare implications. We have one group of workers that experience higher wages at lower housing costs. For this group utility rises. A second group of workers in the high productivity regions experience a utility decline since wages fall and housing costs rise. A third group of workers that leave r e g i o n ; ' , . . . , M and move to regions 1 , . . . , i also experience a decline in utility. In fact, since utility is equalized across regions in the initial equilibrium, and will still be equalized across regions 1 , . . . , / ' w h e r e wages are above the minimum wage, the decline in utility experienced by the second and third groups of workers will be the same. Can we say that the gain of the one group with an increase in utility outweigh the losses of the other two groups? In order to address this question it is useful to discuss the efficiency properties of the spatial equilibrium with and without minimum wages. A standard way to approach this issue is to invoke a central planner that aims at maximizing the utility of a representative worker household in jurisdiction i under the spatial equilibrium constraint that worker utility is equalized across jurisdictions. Lcf = u(x,) +

M

M Vj[u{xi) - u(Xj)} +mY, [pi ~ (xi + %/))"/] !+< /= 1

[" + N L

M • /=!

.

The first set of constraints require that worker utility is equalized across regions - they may be referred to as mobility constraints. The second set of constraints capture the budget constraints for the households requiring that the sum of a region's households' consumption and total housing costs is equal to the total income in this region. The last

Spatial Implications of Minimum Wages • 309

constraint simply states that the total p o p u l a t i o n is fixed. The efficient spatial allocation of labor is obtained f r o m the first order condition with regard to the population size. 6 Lcp —— = fi{Fin - x, - hi - h,nn,) - 4> = 0. an, Taking account of the mobility constraints we derive the locational efficiency condition (Wildasin 1986) F.«(«i) - Hn')

~ h„(tij)nj = F,„(n,) - h(rtj) -

h„(n,)nh

implying that a reallocation of labor c a n n o t increase o u t p u t net of housing costs. N o t e that there is a discrepancy between the central planner's allocation a n d the above migration equilibrium even w i t h o u t the imposition of m i n i m u m wages. This is caused by the crowding effect that arises t h r o u g h the impact of p o p u l a t i o n changes on the total housing costs. Intuitively, w h e n moving f r o m one region t o the other the household ignores the crowding effect. Therefore, the laissez faire migration equilibrium turns out to be not efficient in our model. However, the imposition of the m i n i m u m w a g e does not improve this situation. To see this, consider the crowding effect ¿>„(w,)w, and note that it is positive and increasing in «/. Hence, this term tends t o be larger in the larger region w, > C o m p a r e d with the basic spatial equilibrium, the marginal productivity in the larger region is t o o low - in the smaller region it is t o o high. Therefore, we k n o w that the efficient allocation of labor a n d population w o u l d be such that employment and p o p u l a t i o n are smaller in region i. Thus, denoting the efficient p o p u l a t i o n size with a star we get n, > n* > n* > tij. However, with m i n i m u m wages, we k n o w t h a t w e have an increase in agglomeration relative to the spatial equilibrium distribution, > and n, < Hence n, > «, > n* > n* > n] > nr Thus, we can say that if there is an inefficient spatial equilibrium with excessive agglomeration, due to crowding effects, the imposition of m i n i m u m wages w o u l d give a f u r t h e r push t o w a r d s excessive agglomeration. Therefore, there is n o possibility for a Pareto improvement, the g r o u p of w o r k e r s that benefits f r o m m i n i m u m wages c a n n o t compensate the others.

5

Conclusions

In this paper we have discussed consequences of u n i f o r m m i n i m u m wages for the spatial structure of the economy. The starting point is the notion t h a t a m i n i m u m wage is m u c h more restrictive in the countryside but might be rather ineffective for people working in the cities. An empirical analysis exploiting G e r m a n data shows that a u n i f o r m m i n i m u m wage w o u l d affect the regional labor markets quite differently. In particular, we find that the share of workers that will be directly affected by the m i n i m u m wage is higher in rural counties as c o m p a r e d to cities a n d u r b a n counties. While this supports concerns that the

310 • T. Buettner and A. Ebertz

minimum wage is more effective in the rural as compared to urban areas, the economic consequences depend on the nature of the urban-rural wage differences. A further empirical analysis, however, shows that the wage differences are mainly associated with systematic spatial differences in the wages. Thus, the differences in the incidence of the minimum wage are driven by the spatial wage structure. According to our estimates, and based on some simplifying assumptions, for a female laborer without completed schooling and with mean age the probability to earn a wage below the minimum wage is larger in a rural county by 17 percentage points as compared to a city or urban county. For a male laborer the difference in the probability amounts to 10 percentage points. To explore the consequences of the spatial differences in the incidence of minimum wages, we present a spatial equilibrium model of the labor market, where wage differences occur due to productivity differences and housing costs. Imposing uniform minimum wages in this setting exerts some distortive effects on the spatial structure of the economy. While the wages in the countryside will tend to rise, wages would decline in the city, where employment and population increase. Workers in cities will further suffer from an increase in housing costs. Thus, a federal minimum wage will tend to spur rural-urban migration and might raise rather than reduce urban poverty. Having discussed the spatial implications of minimum wages in a rather straightforward model of the spatial equilibrium, let us briefly talk about possible limitations and extensions. A first issue is the possible existence of federal welfare programs. Such programmes would exert similar effects as minimum wages if they define a uniform reservation wage. Whether or not this is the case in Germany is not obvious, however. While the subsidies according to SGB II are, in fact, uniform, the large housing subsidy programme categorizes the cities and municipalities and assigns higher subsidies to households in urban agglomerations. A second important issue is the role of wage bargaining. In Germany wage bargaining leads to sector-specific agreements defining wage floors that are uniform across several regions. This kind of agreements may exert similar effects on the spatial wage structure. However, wage bargaining is much less restrictive as it does not apply to all firms and shows some limited regional differences (Buettner 1999). Nevertheless, our analysis suggests that these agreements might have already contributed to some excess agglomeration effect in Germany. Finally, we should note that the spatial wage structure is only one example of wage structures that are disregarded by a uniform minimum wage. With other types of systematic wage differences such as the firm-size wage distribution, similar problems will arise. Since a uniform minimum wage is more binding for smaller firms, it would distort the firm-size distribution, and in a competitive setting would benefit capital owners of larger firms, in the same way as the distortion of the spatial wage structure emphasized in this paper benefits land owners in cities.

Spatial Implications of Minimum Wages • 311

O li O I in o o '

rv in CM 0 0

tT—(N co fî — T—(N in IV O T—q 0 O0 O O