Heterogeneity in Wage and Employment Dynamics

Transcription

1 Heterogeneity in Wage and Employment Dynamics Inauguraldissertation zur Erlangung der Würde eines Doktors der Wirtschaftswissenschaft der Fakultät für Wirtschaftswissenschaft der Ruhr-Universität Bochum Kumulative Dissertation, bestehend aus 4 Beiträgen vorgelegt von Diplom-Ökonomin Peggy Bechara, geb. David aus Witten 2011

2 Dekan: Prof. Dr. Helmut Karl Referent: Prof. Dr. Thomas K. Bauer Korreferent: Prof. Dr. Christoph M. Schmidt, Ph.D. Tag der mündlichen Prüfung:

3 Contents 1 Introduction and Overview 1 2 The Importance of Two-Sided Heterogeneity for the Cyclicality of Labour Market Dynamics Introduction Data and Concepts The data Measuring labour market transitions Labour Market Dynamics in Germany: Descriptive Evidence Steady-state results Cyclical features of aggregate worker flows Cyclical features of firm size-specific worker flows Econometric Analysis Econometric specification Estimation results Impulse response functions Summary and Discussion Conclusion Employment Polarization and Occupational Dynamics Introduction Related Literature Employment polarization Employment opportunities for different demographic groups Data and Definitions The IABS data The BIBB data Measuring occupational inflows and outflows Descriptive Evidence Polarization of Employment Opportunities Occupational dynamics Empirical Analysis Econometric specification Estimation results Conclusion

4 Contents ii 4 Labor Market Entry Conditions, Wages and Job Mobility Introduction Initial Conditions, Cohort Effects and Job Mobility Initial labour market conditions and wages Early job mobility Data Econometric Framework Job mobility Development of entry wage differentials Empirical Results Entry wage differentials The impact of the initial wage gap on job mobility Adjustment of entry year effects Sensitivity analysis Conclusion Gender Segregation and Gender Wage Differences during the Early Labour Market Career Introduction Theoretical and Empirical Background Data and Descriptive Statistics The data set Summary statistics Wages Gender segregation Work history Empirical Analysis Estimation strategy Estimation results Sensitivity analysis Conclusion Bibliography 131 Acknowledgements 143 Curriculum Vitae 145

5 Chapter 1 Introduction and Overview Over the last decades the labour markets in most industrialized economies have experienced rapid changes in economic conditions. In particular, technological progress, international trade, cyclical variations in economic activity, as well as demographic changes in the composition of the workforce have induced shifts on both the demand side and the supply side of the labour market. In order to facilitate adjustment to the altered conditions an effective reallocation across industries, occupations and firms is required. As wage and employment dynamics play a key role in this context, investigating their evolution and determinants appears to be of great importance. It is well-known that these dynamics, although often initiated by common impulses, are far from being a uniform phenomenon, but rather vary across different worker and firm groups. Analyses that are based on the "average" worker and "average" firm might thus lead to imprecise results and potentially misleading conclusions. Therefore, in this dissertation special relevance is attached to the heterogeneity that exists on both sides of the labour market. In trying to reach a better understanding of the labour market functioning, this dissertation presents empirical contributions on wage and employment dynamics, with a particular focus on worker and firm heterogeneity. Specifically, the following questions are addressed: Which types of firms employ which types of workers and how does this interaction vary over the business cycle? Which employment dynamics arise when the workforce in routine-intensive occupations is adjusted to a declining labour demand, and do these adjustment processes vary across demographic groups? Do economic conditions at the time of labour market entry entail persistent effects on workers wages? Or can job mobility act as an adjustment mechanism to mitigate these effects? And how does the

6 1. Introduction and Overview 2 gender wage differential evolve over workers early labour market careers? All empirical investigations in this thesis are based on administrative data from Germany, though their findings may apply to other developed countries as well. Chapter 2 (co-authored by Ronald Bachmann) investigates the cyclicality of worker flows, particularly taking into account the interaction of heterogenous workers and firms in this context. This chapter contributes to the literature on the importance of heterogeneities for labour market dynamics by providing new insights, and by complementing the international picture with evidence from the German labour market. In particular, the findings of the analyses suggest that large establishments - contrary to their smaller counterparts - hire predominantly workers out of an existing employment relationship, indicating that large firms compete more successfully for employed job seekers. Regarding the cyclicality of labour market dynamics we find that small establishments hire mainly at the beginning of an economic expansion. Later on in the expansion, hirings more frequently result from direct job-to-job transitions to larger firms. Moreover, the estimation results reveal that hirings at large firms are generally more cyclical. However, this stylised fact disappears when unobserved worker heterogeneity is controlled for, indicating that composition effects play a crucial role in this context. Overall, the findings in this chapter provide an empirical foundation for the inclusion of worker heterogeneity in macroeconomic models. Chapter 3 examines the occupational mobility patterns that are associated with the processes of routinization and employment polarization for several demographic groups. It contributes to the economic literature by employing a dynamic framework providing important insights into the labour market adjustments in response to changes in occupational labour demands. A further distinction concerning the origin states of occupational inflows and the destination states of occupational outflows allows to obtain a more detailed picture of the adjustment dynamics. The findings suggest that the employment contraction in routine occupations can predominantly be attributed to young workers and women who increasingly leave routine-intensive jobs and subsequently enter other occupations or nonparticipation. For these two groups of workers the process of routinization is accompanied with rising flows into high-skill nonroutine occupations, while employment opportunities to males and older workers are increasingly found in nonroutine occupations at the bottom of the skill distribution. Thus, young workers and

7 1. Introduction and Overview 3 women appear to adapt more successfully to changes in the occupational demands. The findings of this paper suggest that the processes of routinization, i.e. the demand shift particularly towards high-skill occupations, might exacerbate the already existing shortage of specialists and thus the difficulties that companies have with recruiting highly qualified employees. Promising strategies to counter these problems is to increase the employment rate especially of female and older workers, e.g. by offering further training opportunities, improving flexible work arrangements and facilitating the entry into nonroutine jobs with high skill requirements. Chapter 4 and Chapter 5 of this thesis focus on workers early labour market career. Since a large part of labour mobility and lifetime wage growth occurs within the first years after labour market entry, analyzing the wage and employment dynamics during this period is especially interesting. In particular, Chapter 4 (co-authored by Ronald Bachmann and Thomas K. Bauer) quantifies the impact of economic conditions at the time of labour market entry on workers starting wages. Furthermore, it examines how wage differentials induced by varying entry conditions evolve over the early career and which role job mobility plays in this context. This chapter contributes to the existing literature in several respects. Firstly, distinguishing between various destination states as well as between voluntary and involuntary job transitions a detailed picture of entry wage effects and their relationship to workers mobility is provided. Secondly, the possible endogeneity of job mobility is addressed by using an instrumental-variable approach. Finally, in contrast to existing studies, the analysis covers labour market entrants of different educational levels. The empirical results suggest that there is a strong relationship between entry wage effects and workers mobility behaviour. That is, workers entering the labour market under unfavourable conditions and earning less than the average starting wage are more likely to change jobs and occupations. Furthermore, the findings reveal that entry wage differentials decrease with labour market experience. Employer changes that occur through an intervening nonemployment spell lead to a further mitigation of entry year effects. Thus, wage convergence is predominantly achieved by the loss of wage advantages which workers experience with an indirect or involuntary job change. It appears to be more difficult to alleviate initial wage disadvantages, indicating that there might be a need for more wage flexibility in Germany. Moreover, the findings in this chapter suggest that

8 1. Introduction and Overview 4 rather than focussing on a single point in time, expanding the time horizon provides important information on the persistence of wage differentials and the processes of wage convergence. The gender wage gap at the time of entering the labour market and its development during workers early careers is being investigated in Chapter 5. The analysis contributes to the existing research on gender wage differentials among young workers by providing evidence on the impact of women s segregation into certain industries, occupations, establishments and job-cells, i.e. occupations within establishments. The estimation results reveal that all types of segregation and particularly job-cell segregation are significant determinants of the gender wage gap, while skill endowments and differences in work histories are found to be of minor importance. At the time of labour market entry women s wage disadvantages can almost entirely be explained by the fact that they start their working career in lower-paying occupations and establishments. With progressing labour market experience, however, gender segregation becomes less important and cannot fully account for a slight widening of the wage differential among young men and women. Therefore, part of the early career wage gap remains unexplained. Besides data limitations (e.g. lacking information on working hours, number of children and hierarchical levels within occupations and job-cells) a possible reason for the increase in the unexplained gender wage gap might be the predominant view that family responsibilities prevent women from climbing the job ladder. Thus, the investigations in this chapter stress the importance to alleviate the problems of women s discrimination as well as their segregation into female-dominated occupations by equaling employers hiring and promotion practices and facilitating the reconciliation of family and working life. Overall, the empirical contributions in this dissertation reveal that the analyses of labour market outcomes over an extended time interval rather than focussing on crosssectional figures provide insightful information and therefore a more comprehensive picture of the labour market. In particular, static figures can only give a snapshot of wages and employment at a single point in time and thus might lead to possibly wrong implications, while expanding the time horizon and considering the underlying dynamics allows to picture the structure and functioning of the labour market in more detail. Moreover, this dissertation points out that wage and employment dynamics

9 1. Introduction and Overview 5 substantially vary across different worker and firm groups, implying that heterogeneity plays a key role for the understanding of labour market processes. As structural and technological changes generate disadvantages for one group of workers which cannot be compensated by the advantages obtained by the other, targeted measures are needed to alleviate disadvantages and to equal opportunities for the whole population.

10 Chapter 2 The Importance of Two-Sided Heterogeneity for the Cyclicality of Labour Market Dynamics Together with Ronald Bachmann 2.1 Introduction The analysis of the cyclicality of labour market dynamics has been a very active field of research for the last two decades. 1 Interest in this issue has been further increased by the debate about the relative importance of the ins and outs of unemployment in this context (cf. Darby, Haltiwanger, and Plant, 1986, and Shimer, 2007). While a consensus seems to emerge that both inflows into and outflows from unemployment have some role to play (cf. Elsby, Michaels, and Solon, 2009, and Fujita and Ramey, 2009), important questions remain unanswered. One crucial question, raised by Elsby, Michaels, and Solon (2009), is why job-loss-induced inflows to unemployment increase at the beginning of a recession and why outflows do not increase enough to keep unemployment duration from rising. An obvious suspect in this context is the interaction of heterogeneous agents on both sides of the labour market over the business cycle. However, as Moscarini and Postel- A preliminary version of this chapter has been circulated as Ruhr Economic Paper No. 124 and SFB 649 Discussion Paper No Analyses of the dynamics of the German labour market are contained in, e.g. Schmidt (2000), Fitzenberger and Garloff (2007), and Bachmann (2005).

11 2. Two-Sided Heterogeneity and Labour Market Dynamics 7 Vinay (2008) point out, this process is up to now little understood. They argue that, on the US labour market, specific phases of the business cycle see different types of firms hiring different types of workers, which leads to specific labour market transitions and wage dynamics. In particular, in the early phase of an economic expansion, small firms hire mainly from the ranks of the unemployed, a process which results in relatively low wages. In later phases of an economic expansion, hirings from larger firms predominate. With the pool of unemployed workers having shrunk considerably, this entails more direct job-to-job transitions from small to large firms, and higher wages. The interaction of heterogeneous firms and workers thus has important implications for both labour market transitions and the evolution of the wage structure. Our analysis aims at providing empirical evidence on the cyclicality of the German labour market, with a particular focus on both the hiring and firing behaviour of establishments belonging to different size classes, and the heterogeneity of workers. We do so by using a very rich administrative data set providing information on social security employment for West Germany and spanning three decades of workers employment history. This data set makes it possible to analyze the role of heterogeneity on both sides of the West German labour market over the business cycle. We are thus able to provide a complete set of stylized facts on this topic and to conduct a rigorous econometric analysis controlling for both observed and unobserved heterogeneities on both sides of the labour market. 2 In particular, we establish five facts: 1. Small establishments hire roughly equal proportions of previously employed and unemployed workers. By contrast, large establishments hire mainly from employment (via job-to-job transitions) and non-participation, much less so from unemployment. Similar patterns can be observed for the destination states after a job separation. 2. Employment-to-employment transitions are procyclical, and employment-to-unemployment transitions are countercyclical. 3. In recessions, unemployment supplies relatively more workers to establishments of all sizes. 2 See Abowd and Kramarz (1999) for an analysis of the determinants of worker flows between different labour market states by accounting for various individual and firm characteristics, which were shown to play an important role.

12 2. Two-Sided Heterogeneity and Labour Market Dynamics 8 4. Hires out of unemployment, as well as employment-to-unemployment transitions, appear to be more cyclically sensitive at large establishments than at small establishments. 5. The greater cyclicality of the transitions between employment and unemployment by large establishments seems to be entirely due to composition effects. These facts contribute to the literature on the importance of heterogeneities for the dynamics of worker flows by providing new insights and by complementing the international picture with evidence from the German labour market. In particular, Fact 1 is related to the findings by Eriksson and Lagerström (2006) who show that, on the Swedish labour market, unemployed job applicants face a lower probability to get contacted by a firm than otherwise identical employed applicants. They argue that this is due to the fact that firms view employment status as an important signal for productivity. Nagypál (2006) provides another theoretical argument for why firms might prefer hiring employed, rather than unemployed, workers. Workers arriving from unemployment are less likely to end up in a job they are happy with than employed job searchers. Therefore, the former workers are more likely to engage in job-shopping and to leave an employment relationship for a more appealing job. Given that hiring workers involves fixed costs, firms can economize on these costs by hiring employed workers. As for separations, Frederiksen and Westergaard-Nielsen (2007) analyse the effects of individual and workplace characteristics, as well as of the business cycle, on individual job separations and the associated destination states in the Danish private sector. They find that there is large heterogeneity both within and between destination states. Fact 2 confirms evidence about the procyclicality of employment-to-employment flows (cf. Fallick and Fleischman, 2004, Nagypál, 2008, and Bjelland, Fallick, Haltiwanger, and McEntarfer, 2008) and the countercyclicality of unemployment-to-employment transitions (Yashiv, 2008, and Elsby, Michaels, and Solon, 2009), which were also found for some European economies by Burda and Wyplosz (1994). Fact 4 confirms the evidence for hirings provided by Moscarini and Postel-Vinay (2010) for the US economy as well as for Denmark, France, and Brazil, and qualifies it for separations. Finally, Fact 5 provides new econometric evidence on the importance of labour market heterogeneities for the cyclicality of labour market dynamics.

13 2. Two-Sided Heterogeneity and Labour Market Dynamics 9 Our findings, and especially Facts 3-5, have important implications for the way we think about labour market dynamics, and thus for economic modeling, which are discussed extensively in Section 2.5. The plan of this chapter is as follows. The next section describes the data sets used in our analysis. The third section summarizes the descriptive empirical evidence. In particular, Section focuses on the importance of labour market heterogeneities in steady state, Section provides evidence on the cyclicality of aggregate labour market flows, and Section provides stylized facts with respect to labour market heterogeneities and the business cycle. Section 2.4 presents an econometric analysis of the cyclicality of these dynamics in order to analyse the role of composition effects in this context. In Section 2.5, we summarize the empirical evidence and discuss the implications for the theoretical modelling of labour market cyclicality. The last section concludes the analysis. 2.2 Data and Concepts The data The main data source for our analysis is the IAB Employment Sample (IABS) for the time period , which is provided by the Institute for Employment Research (IAB). The IABS is a 2% representative sample of the Employment Statistics Register, an administrative panel data set of the employment history of all dependent employees subject to social security contributions in Germany, supplemented with information on all unemployment spells of the workers covered. For 1995, the Employment Statistics Register contains the labour market history of 79.4% of all employed persons in Western Germany, and 86.2% of all employed persons in Eastern Germany. Civil servants and self-employed workers are not included in the data. 3 The basis of the employee history is the integrated notification procedure for health insurance, the statutory pension scheme, 3 The exclusion of civil servants should not influence our estimation results, as these individuals exhibit quite stable employment relationships largely unaffected by cyclical variations. By contrast, excluding self-employed workers might be relevant for our findings. Constant and Zimmermann (2004) show that transitions out of self-employment are not affected by the business cycle, while transitions into self-employment are highly procyclical. However, Caliendo and Uhlendorff (2008) investigate yearly transition rates and find that only 3% of all non-employed workers and only 1% of all wage-employed workers enter the state of self-employment, implying that transitions into and out of this state only plays a minor role for our analyses.

14 2. Two-Sided Heterogeneity and Labour Market Dynamics 10 and unemployment insurance. At the beginning and at the end of any employment spell, employers have to notify the social security agencies. This information is exact to the day. For employment spells spanning more than one calendar year, an annual report for each employee registered within the social insurance system is compulsory, and provides an update on, for example, the qualification and current occupation of the employee. 4 The IABS provides information on workers employment status, age, gender, occupation and education as well as limited information on firm characteristics (economic sector, establishment size). This data set is representative for all dependent-status workers, and contains information on all employment and unemployment spells of the workers covered. Given the relatively long time span of the data set, we are able to observe two full business cycles. From this sample, we exclude observations in East Germany, apprentices, trainees, homeworkers, part-time workers, and individuals older than This results in a sample with 1.3 million individual observations. The IABS is representative regarding employment covered by the social security system but not regarding unemployment, since only those unemployed who are entitled to transfer payments are covered. In our data, we can derive three labour market states at each point in time: employment (E) covered by social security, unemployment (U), if the worker is receiving transfer payments, and non-participation (N). 6 Non-participants are those individuals not recorded in the data sets. Therefore, this state includes those workers out of the labour market, as well as workers not covered by social security legislation, e.g. civil servants and self-employed workers. Because of the way the data are collected, both firms reports of a new employee and individuals notifications of moving into or out of unemployment are not exactly consistent with the actual change of labour market state. For example, workers might report to the unemployment office only a few days after having been laid off. The latter potential measurement error is taken into account in the following way: If the time lag between two employment or 4 A detailed description of the Employment Statistics Register and the notification procedure is given by Bender, Haas, and Klose (2000). 5 Excluding part-time workers from our sample and treating them as non-participants artificially increases our transitions into and out of non-participation. However, as the IABS data only distinguish between two categories of part-time employment and the number of working hours can be relatively low, we decided to focus on core full-time workers. How part-time workers come into play in this context is a matter of future research. 6 In the IABS data, the record on unemployment benefit recipients are unreliably measured before As we can therefore not use the worker flows to and from unemployment for the time period , we start our analysis in 1980.

15 2. Two-Sided Heterogeneity and Labour Market Dynamics 11 unemployment notifications does not exceed 30 days, it is defined as a direct transition between the two states recorded. We count it as an intervening spell of non-participation if the time interval between the two records is larger than 30 days. The descriptive statistics of the data set as used in the econometric analysis are in Table A Measuring labour market transitions Since the data set used contains daily information on the employment and unemployment history of every individual in the sample, it is possible to calculate worker flows taking into account every change of labour market state that occurs within a given time period. We are thus able to compute the flows between employment, unemployment and non-participation, as well as direct job-to-job transitions (EE flows) using the establishment identification number, which implies that our notion of a job is establishmentbased. In addition to EE flows, our analysis focuses on the flows from employment to unemployment and to non-participation (EU and EN, respectively), and from unemployment and from non-participation to employment (UE and NE, respectively). We define as separation flows all flows emanating from employment, S t = EE sep t +EU t +EN t, and as accession flows all flows going to employment, A t = EE acc t +UE t +NE t. Following Davis and Haltiwanger (1999), we calculate the corresponding rates of each flow by using the average of current and past employment (E t 1 E t )/2 as the denominator. 7 It has to be mentioned, that our definition of establishment size classes is based on the size establishments display on June 30th of the contemporaneous year rather than the previous year. Since the IABS data are individual-based and not establishmentbased data, small establishments are less likely to be observed for two consecutive periods than large establishments. This implies that using the establishment size of the previous year would result in a selected sample, where small establishments are underrepresented. 7 The stock of employment is defined as the number of individuals employed at the reference date June 30th in year t.

16 2. Two-Sided Heterogeneity and Labour Market Dynamics Labour Market Dynamics in Germany: Descriptive Evidence In this section, we derive some stylized facts concerning the dynamics of job and worker flows in the West German labour market. In order to do so, in Section we present steady-state results, both for the aggregate labour market and for firms of different size classes. The cyclical features of aggregate workers flows on the one hand, and of firm size-specific worker flows on the other hand are presented in Sections and 2.3.3, respectively Steady-state results We first focus on the stylized facts about worker flows for different establishment size classes that are invariant over the cycle and can thus be regarded as steady-state results. Table 2.1 reports the yearly averages of the worker flow rates for the time period for establishments of different sizes. 8 Table 2.1: Worker flow rates across different establishment size categories EE NE UE EN EU All observations employees employees employees and more employees Source: Authors calculations based on IABS , for West Germany. Note: EE: Employer-to-employer flows; NE: Nonparticipation-to-employment flows; UE: unemployment-to-employment flows; EN: Employment-tononparticipation flows; EU: Employment-to-unemployment flows. All figures are weighted averages of the annual values ( ). Establishment size appears to be strongly correlated with worker flows. In particular, there is a general tendency of hiring flows and separation flows to decline with the establishment size. The finding of higher fluctuations in smaller establishments is consistent with other research (Davis and Haltiwanger, 1999, and Lane, Stevens, and Burgess, 1996). As pointed out in the introduction, firms are likely to have preferences over the 8 Differences in worker flow rates between different worker categories have been widely reported in the literature and can also be found in Bachmann and David (2009).

17 2. Two-Sided Heterogeneity and Labour Market Dynamics 13 previous labour market state of their new hires (cf. Eriksson and Lagerström, 2006, and Nagypál, 2006). Firms are likely to prefer hiring employed workers because unemployment may be perceived as a negative signal. Furthermore, the expected duration of a new job is higher for previously employed job seekers because the match is likely to be a better fit than if the worker had been previously unemployed. In order to investigate the consequences of these mechanisms, we analyse the origin of new hires for different establishment size classes. Looking at all the establishments considered, 32.9% of new hires come from employment, 25.2% come from unemployment, and 41.9% from non-participation (cf. Table 2.2). The hiring source depends strongly on the size of the establishment. Small establishments hire roughly equal proportions of their new workers from employment and unemployment (29.8% and 28.6%, respectively). With growing establishment size, however, the proportion of hires from employment increases at the expense of hirings from unemployment. Very large establishments hire 36.8% of their new male workers from employment, but only 15.4% from unemployment. Table 2.2: Distribution of sources and destinations by establishment size Hirings from Separations to Establishment size E N U E N U All observations and more Source: Authors calculations based on IABS , for West Germany. Note: All figures are weighted averages of the annual values ( ). The larger job-to-job flows to large establishments could be explained by the fact that transitions out of an old job to a larger establishment generally lead to greater wage gains than moving to an equally-sized or smaller firm (cf. Figure B.2.1). Thus, to the extent that firms prefer hiring employed workers, large firms are able to compete more successfully for employed job seekers in the labour market. An examination of the distribution of destination states that follow a job separation leads to very similar results (cf. Table 2.2). Considering all establishments, 33.0% of the separations result in a new employment relationship, 26.1% in unemployment, and

18 2. Two-Sided Heterogeneity and Labour Market Dynamics % end in non-participation. When we split up the establishments into different size classes, we can observe strong size-specific variations in the distribution of separation destinations. In particular, for small establishments we find a roughly equal proportion of the separations to lead to a new employment (31.2%) and to unemployment (31.0%). In contrast to this, separations from very large establishments are followed by employment in 32.0% of cases, and only 17.6% are followed by an unemployment spell. The main difference is thus that many workers in large establishments exit to non-participation. This might be due to the fact that workers in large establishments tend to be older and therefore are more likely to retire than workers leaving small establishments. The preceding analysis leads us to state Fact 1. Small establishments hire roughly equal proportions of previously employed and unemployed workers. By contrast, large establishments hire mainly from employment (via job-to-job transitions) and non-participation, much less so from unemployment Cyclical features of aggregate worker flows The evolution of the accession and separation rates for the time period is displayed in Figure 2.1, with shaded areas indicating times of recession. The accession rate is clearly procyclical, as is the separation rate, but to a lesser extent than the accession rate. This implies a reduction of the aggregate employment level during recessions. These findings are in line with Bachmann (2005) who points out that during recessions, a decline in the hiring activity can be observed. Figure 2.1: Accessions and separations, , yearly rates Separations Accessions Source: Authors calculations based on IABS , for West Germany. Note: The figures are calculated as described in Section 3.2. Shaded areas are times of recession.

19 2. Two-Sided Heterogeneity and Labour Market Dynamics 15 In order to further investigate this matter, we split up the accession flows into EE flows, UE flows and NE flows. As one can see in Figure 2.2, job-to-job transitions show a clearly procyclical pattern, as do transitions from non-participation to employment. Figure 2.2: The dynamics of worker flows, , yearly rates.16 Accession flows.16 Separation flows EE UE NE EE EU EN Source: Authors calculations based on IABS , for West Germany. Note: EE: Employer-to-employer flows; NE: Nonparticipation-to-employment flows; UE: unemployment-to- employment flows; EN: Employment-to-nonparticipation flows; EU: Employment-to-unemployment flows. The figures are calculated as described in Section Shaded areas are times of recession. However, the flow from unemployment to employment, being not as volatile as the other two worker flows, rises much earlier and drops during periods of economic recovery. These observations indicate that the outflow from unemployment dominates during recessions and during the beginning of expansions, while job-to-job transitions are the most important source of accessions in the mature phase of expansions. From this, we can infer Fact 2. Employment-to-employment transitions are procyclical, and unemployment-toemployment transitions are countercyclical. The three worker flows making up separations, namely the EE flows, EU flows and EN flows, are displayed in the second panel of Figure 2.2. It becomes apparent that direct job-to-job flows and the flows from employment to non-participation are procyclical, while the flow from employment to unemployment increases during recessions and decreases in periods of economic recovery. This means that we can observe a shift from employment-to-unemployment transitions to job-to-job transitions in the mature phase of the economic expansion.

20 2. Two-Sided Heterogeneity and Labour Market Dynamics Cyclical features of firm size-specific worker flows We now turn to the question whether the cyclical features presented in the previous section vary between firms that differ in size. In order to do so, we first compute the share of a specific worker flow F in total hirings H for establishments of a specific size, i.e. the fraction F st /H st, with s being the establishment size class s and t the year under investigation. The results are depicted in Figure 2.3. As in Table 2.2, it again becomes obvious that for larger establishments, job-to-job transitions play the biggest Figure 2.3: The shares in hirings by establishment size, , yearly rates employees employees EE UE NE EE UE NE employees and more employees EE UE NE EE UE NE Source: Authors calculations based on IABS , for West Germany. Note: For each establishments size class the flows are computed as share of total hirings. Establishment size classes are based on size in the contemporaneous year. EE: Employer-to-employer flows; NE: Nonparticipation-to-employment flows; UE: unemployment-to-employment flows; EN: Employment-to-nonparticipation flows; EU: Employment-to-unemployment flows. role, whereas the outflow from unemployment makes up only a small part of hirings. For small establishments, however, hirings out of employment outweigh hirings out of unemployment in the mature phase of an economic expansion, while the opposite is

21 2. Two-Sided Heterogeneity and Labour Market Dynamics 17 the case during recessions and the early phase of the economic upturn. Moreover, the results show that the hiring share from employment is procyclical for all establishments, while the opposite is true for the hiring share from unemployment. This leads to Fact 3. In recessions, unemployment supplies relatively more workers to establishments of all sizes. One can see very similar patterns in Figure 2.4, which presents for different establishment size classes the three separation flows as a share of total separations, i.e. the fraction F st /S st. Here the flows between employment and unemployment also appear to become less important the larger the establishments get. Furthermore, for all establishments the separation share to employment is procyclical, while the separation share to unemployment is countercyclical. Figure 2.4: The shares in separations by establishment size, , yearly rates employees employees EE EU EN EE EU EN employees and more employees EE EU EN EE EU EN Source: Authors calculations based on IABS , for West Germany. Note: For each establishment size class the flows are computed as share of total separations. See notes to Table 2.3. In a second step, we calculate the size-specific worker flows as a share of total worker

22 2. Two-Sided Heterogeneity and Labour Market Dynamics 18 flows, i.e. the fraction F st /F t, where F t denotes the same, but economy-wide, flow in year t. In order to find out how these fractions are related to the business cycle, we compute their correlations with the growth rate of GDP for hirings and separations, as well as the underlying flows (cf. Tables 2.3 and 2.4). Table 2.3: Correlations between firm size-specific hiring flows and GDP growth rate Establishment size GDP(t) GDP(t-1) GDP(t-2) GDP(t-3) GDP(t-4) H s /H * * * * 1000 and more * EE acc s /EE acc and more * UE s /UE * * * 0.211* 1000 and more * NE s /NE * * * * 0.287* 1000 and more 0.202* Source: IABS, transformed to a quarterly data set by the authors, for West Germany 1980/I-2004/III.. Note: H refers to hirings, F s to a particular flow occurring in establishment size class s, and F denotes the same, but economy-wide, flow. A * indicates a level of significance of at least 10%. Table 2.3 shows that the hiring share of small establishments displays a strong negative correlation with GDP growth. This is due to the countercyclical UE and NE flow shares of this establishment size class, for both contemporaneous and lagged GDP growth. The hiring shares of large establishments, by contrast, are strongly procyclical for all the flows considered. Interestingly, for large establishments the correlation of the hiring shares with the different lags of GDP growth is strongest for the first lag of GDP, which can be viewed as a manifestation of the fact that larger establishments start hiring later in an economic expansion. This strong correlation is driven by the cyclical behaviour of the EE and UE flows. Thus, the cyclicality of hirings by large firms is not only due to these firms hiring employed workers, but also that they increase their hiring share from

23 2. Two-Sided Heterogeneity and Labour Market Dynamics 19 unemployment during expansions. Furthermore, the correlations are slightly larger for large establishments than for small establishments. Table 2.4: Correlations between firm size-specific separation flows and GDP growth rate Establishment size GDP(t) GDP(t-1) GDP(t-2) GDP(t-3) GDP(t-4) Sep s /Sep * * * 1000 and more 0.161* EE sep s /EE sep and more EU s /EU and more EN s /EN * * and more Source: IABS, transformed to a quarterly data set by the authors, for West Germany 1980/I-2004/III.. Note: S refers to separations, F s to a particular flow occurring in establishment size class s, and F denotes the same, but economy-wide, flow. A * indicates a level of significance of at least 10%. The evidence on separation shares is qualitatively similar (cf. Table 2.4), but generally less pronounced than the evidence on the hiring shares. The most striking result for separations is that the separation share of large establishments is very strongly correlated with current GDP. This is due to the strong increase of this share of both direct employer-to-employer flows and flows from employment to nonparticipation. The former increase can be put down to the procyclicality of job-to-job transitions, which to a large extent seem to be taking place between large firms. The latter increase could be due to the fact that employed workers are more likely to become self-employed during economic upswings. Unfortunately, we cannot investigate this hypothesis further using our data set. These results can be summarized in Fact 4a. Hires at large establishments, especially out of unemployment, appear to be more cyclically sensitive than hires at small establishments. This does not seem to be

24 2. Two-Sided Heterogeneity and Labour Market Dynamics 20 the case for separations. 2.4 Econometric Analysis Econometric specification The descriptive analysis indicated that two-sided heterogeneity plays an important role for the cyclicality of labour market dynamics. We now want to analyse this issue econometrically in order to find out whether composition effects play a role in this context. For example, the increase in job-to-job transitions to large establishments during the mature phase of an economic upswing may be equally spread across all workers, which means that composition effects are not important. Alternatively, large firms may be hiring more workers of a particular type in this situation, which could be related to certain observable characteristics (e.g. skills) or certain unobservable characteristics ( high-turnover vs. low-turnover workers). As both are taken into account in our empirical analysis, we obtain a composition-adjusted effect of output growth on transition probabilities, which can be viewed as the effect on the probability of transition within worker type. In order to investigate the determinants of worker flows, we estimate a logit model Pr[Y it = 1 X it,β,γ,α i ] = Λ(α i +β X it +γ Z et +δgdp t ), (2.1) where Λ(.) is the logistic cdf with λ(z) = e z /(1 + e z ). In doing so, we estimate two specifications, a random effects model and a fixed effects model (conditional logit). 9 This allows us to explicitly take into account unobserved worker characteristics. The random effects model eliminates the individual-specific effect α i by integrating over a specified distribution of this effect, which is taken to be a random variable. In our application, this model has the advantage that it allows estimation over all individuals in the sample, i.e. also those that never make a transition. However, the random effects estimators are inconsistent if fixed effects are present which are correlated with the regressors. For this reasons, we also estimate a fixed effects model. Qualitatively, i.e. with respect to firm-size specific differences, the two models yield very similar results. 9 Regarding separations, we also estimated semi-parametric duration models as a robustness test. This yielded results very similar to the random effects model (cf. Table A.2.7 in the appendix).

25 2. Two-Sided Heterogeneity and Labour Market Dynamics 21 As dependent variables, we consider separations (i.e. the probability of an employed person to separate from his employer), as well as their components (i) transitions from employment to unemployment, and (ii) transitions from employment to another job; furthermore, we estimate the hiring transitions from unemployment to employment, and job-to-job-transitions. In particular, the logit model for separations specifies the probability whether or not an individual leaves the establishment between t 1 and t, while the logit models for the accession flows specify what happened to individuals between t 1 and t for all employees being employed at time t. These probabilities are explained by observable person characteristics X it (age, duration of previous employment, duration of previous unemployment) observable firm characteristics Z et (industry), and unobservable worker characteristics as described above. The vector GDP t, our measure of the business cycle, contains contemporaneous and lagged GDP growth (lags 1 to 4) and captures the dynamic structure of the labour market process under investigation. In order to analyse the size-specific variations in the cyclical timing of hirings and separations, we estimate both specifications for large and small firms separately Estimation results The estimation results for the cyclical behaviour of the labour market transitions under consideration are shown in Tables 2.5 to Regarding the separation probability of an existing job match the estimation results in Table 2.5 indicate that for small establishments separations are initially countercyclical (acyclical in the fixed effects specification), but from the first lag on show a procyclical pattern. The same pattern can be observed for large establishments. However, they are initially less cyclically sensitive than small firms, but from the third lag on show a stronger procyclical effect. Overall, there does not seem to be a strong difference in the cyclical sensitivity between small and large firms, but the latter are slower in their response. For the transitions from employment to employment (EE), the estimation results indicate a procyclical pattern for small establishments, with the positive impact increas- 10 Large establishments are defined as those employing more than 100 workers. Trying an alternative definition, we find very similar estimation results (cf. Tables A.2.9 to A.2.12 in the appendix). 11 In these tables the fixed effects specification yields coefficients on GDP that are much larger than in the random effects specification. As random effects regressions performed on the reduced fixed-effects sample show, this is purely due to the sample selection that goes along with the use of the conditional logit estimator.complete regression results including the coefficients for the remaining explanatory variables are listed in Tables A.2.3 to A.2.6 in the appendix.