Regional decomposition in age-group unemployment dynamics in Germany

In all OECD countries, the unemployment rates for younger workers are over those for older ones. For example, on an OECD average over the last decade, the youth (15 to 24 years) unemployment rate is about three times larger than the unemployment rate of workers aged 55 to 64. In addition, compared to prime-age workers (25 to 54 years), the youth unemployment rate is about twice as large, and the rate of older workers is only three-fourths of the prime-age worker’s rate. Simultaneously, based on data on unemployment duration, it will be argued that older unemployed need more time to find a new job. On the other hand, considering job tenure statistics, the risk of losing a job is lower for older workers.¹

This raises the question about age-group-related differences in the risk of losing a job and the chance to find new employment in a labor market with unequal unemployment risk across age-groups and has implications for regional labor market dynamics when group sizes shift. Due to the German labor force’s aging, it is essential to understand the differences between younger, prime-age, and older workers.

In this article, I study the relative contributions of unemployment inflows and outflows to the dynamics of different age-group unemployment rates at the regional level in Germany. I can show that younger, prime-age, and older workers differ in labor market flow rates and relative contributions to unemployment dynamics using regional administrative data from the German Federal Employment Agency. While recent studies focus on the national level and mainly on the relative contribution of separation and job finding, this study also considers the flows from/to the non-labor market to/from unemployment.

Concerning the literature on the relative importance of separation and job finding rates for the dynamics of the unemployment rate, Hall (2005) and (Shimer 2005, 2012), Petrongolo and Pissarides (2008), and Elsby et al. (2013) conclude that the job finding rate is more relevant for the US labor market, while Darby et al. (1986) and Fujita and Ramey (2009) come to the opposite conclusion and find evidence for a relatively more important contribution of job separation. Elsby et al. (2009) “find that everyone is a winner.” Smith (2011) finds evidence that increases in the unemployment rate come along with rising separations in the UK. Petrongolo and Pissarides (2008) conclude that both flow rates are equally important for the UK labor market, while job finding rates contribute relatively more to the French and Spanish unemployment fluctuations. In their analysis of different OECD countries, Elsby et al. (2013) find, on average, an inflow-outflow contribution of 15:85 for Anglo-Saxon economies and a relative ratio of 45:55 for continental European and Nordic countries.

Three studies have analyzed Germany in more detail. Nordmeier (2014) used a 2% labor force sample and analyzed the time aggregation bias for separation and job finding in monthly data between 1981 and 2007. She concludes that the job finding rate is more critical in explaining unemployment fluctuations using a two-state model. Hertweck and Sigrist (2015) used SOEP data for West Germany and the period 1984 to 2009 to analyze labor market flows disaggregated by gender, age, and educational background using a three-state model. They find that inflow fluctuation contributes more to unemployment fluctuations (60:40). Jung and Kuhn (2014) also conclude that inflow contributes more to unemployment rate volatility (60:40) using a 2% sample of the West-German labor force (1980 to 2004) and two-state and three-state models. All studies have in common that only macroeconomic conclusions can be drawn. The present article considers regional (county-level) panel data to cover within country local labor market heterogeneity.

When older and younger workers are not perfect substitutes, the literature’s findings are related to a specific demographic composition. Shimer (2001) argues that a high proportion of young workers incentivize firms to create new jobs because younger workers undertake more search activities, which reduces the firms’ recruitment costs. In this case, young workers could have a relatively higher job finding rate. In addition, Burgess (1993) and Pissarides and Wadsworth (1994) find evidence in Great Britain that job separation rates are higher for young workers because a higher proportion of such workers engage in on-the-job search activities.

Older workers’ lower job finding rates can result from age discrimination (Charness and Villeval 2009; Langot and Moreno-Galbis 2013) and assumed or actual productivity differentials (Haltiwanger et al. 1999; Hellerstein et al. 1999; Daniel and Heywood 2007). Productivity may increase with age if job experience is essential (Autor et al. 2003; Nordström Skans 2008) or decline if human capital depreciates over time, mainly due to technological change or a loss of manual abilities (Börsch-Supan 2003; Autor and Dorn 2009). Concerning cognitive abilities, the age effect is more complex. Engaging in information processing is lower among senior workers (Baltes and Lindenberger 1997), making it difficult to employ older workers in challenging jobs, such as flight control.

Considering these findings, I argue that it is not evident which implications the increasing relative appearance of older job seekers and job candidates may have relative to job-worker matching in the labor market and, ultimately, unemployment dynamics. Concerning the existing literature, only Hertweck and Sigrist (2015) analyze three age-groups using the West German sample of the socioeconomic panel. This study adds to the literature by analyzing the contributions of five age-groups to unemployment fluctuations at the regional level using administrative data. In contrast to most literature on relative flow contributions to unemployment dynamics, I consider regional panel data instead of aggregated time series. In contrast to other studies, I use administrative data that precisely cover the flows of the official unemployment rates at the county level.

Considering a three-state model (flows between employment and unemployment and between unemployment and inactivity) and monthly flow data, I find that the contributions of inflow and outflow rates to overall unemployment fluctuations are almost 50:50 using the non-steady-state approach. For all unemployed, the dynamics that arise from the inactivity and activity flow account for about 23% of unemployment dynamics. For the youth (15 to 24 years), this relative contribution accounts for 83%. The remaining unemployment fluctuations for all unemployed are explained by separation (49%) and job finding (27%), respectively. Separations are more significant for all age-groups than job finding contributions. Regions with a larger share of the labor force youth have high dynamics and very low unemployment rates. In contrast, regions with a smaller share of the youth experience twice as large unemployment rates. Agglomeration areas experience a more considerable turnover between active and passive labor markets than rural areas. The relative contributions of separation and job finding are more relevant for older workers’ unemployment, while the relative activity/non-activity flows most contribute to younger workers’ unemployment rate. In general, inflow and outflow rates decline with age. Compared to prime-age workers, I find lower inflow and exit rates for older workers. Hence, aging of the labor force lowers the dynamics of the German labor market, particularly in regions that suffer from fast aging.

The remainder of the paper is organized as follows. Section 2 provides a steady-state and a non-steady-state model to analyze relative flow contributions to unemployment dynamics. Section 3 describes the data, and Sect. 4 provides an empirical analysis of the German labor market using different models and discusses some policy implications. Finally, I summarize the main findings in Sect. 5.

Following Petrongolo and Pissarides (2008), Smith (2011), and Elsby et al. (2013), I use transition rates related to unemployment inflows and outflows to measure relative flow contributions to unemployment fluctuations. However, the three-state model should be compatible with the administrative data used in the next section and differs slightly from the standard approach.

In a two-state model, only the flows between unemployment U and employment E are considered, and workers neither enter nor exit the labor force. The three-state model considers flows out of and into the labor force from/to the stock inactivity I (see Fig. 1). I consider the following flows: from unemployment to employment UE, from employment to unemployment EU, from unemployment to inactivity UI, and from inactivity to unemployment IU. Flows between employment and inactivity are not considered. From a statistical perspective, they are not necessary to measure the unemployment rate—the focus in the empirical section. It takes only the four flows considered to explain the complete fluctuation of the official number of unemployed. However, flows between employment and inactivity affect the unemployment rate indirectly, but in this case, the focus is on the stock of employed and inactive people.²

During period t; the following transition rates according to a Poisson process are considered: job finding rate \(f_{t}=UE_{t}/U_{t-1}\), non-activity rate \(n_{t}=UI_{t}/U_{t-1}\), separation rate \( s_{t}=EU_{t}/E_{t-1}\), and activity rate \(a_{t}=IU_{t}/E_{t-1}\). The latter arrival rate should be ideally related to I. However, the stock of inactive people is difficult to measure and would include a potential measurement error that is difficult to interpret. Consequently, the considered definition differs from the activity rate in the literature. However, the advantage of this measure is that it is more comparable to flows due to separation (e.g., size and dynamics). For all transition rates, we have \(0\le f_{t},n_{t},s_{t},a_{t}\le 1\).

The period considered in this article is 2007-2014, which is covered by a reduction of the overall unemployment rate from 9.0% in 2007 to 6.7% in 2014. I use administrative data provided by the German Federal Employment Agency from January 2007 to December 2014.⁶ Concerning the unemployed, I analyze all unemployed and the j age-groups of 15-24 years, 25-34 years, 35-44 years, 45-54 years, and 55-64 years old to point out age-related differences in unemployment dynamics. The unbalanced panel for 402 regions (counties) provides for each month a distribution over all counties in Germany (with up to 38,592 observations for each variable). The data are not seasonally adjusted. While seasonal adjustment is common for national data, the fluctuations are important for understanding individual county dynamics.

Monthly regional data provide exact information on the stocks of employment E and unemployment U and flows between these two stocks and unemployment and inactivity I. Employment cover employees subject to social insurance contribution (not the informal labor market). Unemployment means that the individual is actively searching for employment (subject to social insurance contribution) and registered as unemployed.⁷

The flows are used to calculate the arrival rates according to Sect. 2:The inflow rate is \(i_{t}=s_{t}+a_{t}\) and the exit rate is \(e_{t}=\) \( f_{t}+n_{t}\). For all transition rates we have \(0\le i_{t},e_{t},f_{t},n_{t},s_{t},a_{t}\le 1\). Following section 2.3, the calculated arrival rates are used to compute the corresponding arrival probabilities \(S_{t},A_{t},F_{t},N_{t},I_{t},\) and \({\check{E}}_{t}\). For each group, regional monthly values are calculated. With regional index r, the local equilibrium unemployment rate (2) is equal toTo illustrate the direct relationship between age-group-related unemployment rates and the unemployment rate for all unemployed, I rearrange the equilibrium unemployment rate (6) using age-specific rates weighted at the relevant population share \(p_{j}\), with \(j=1,2,3,\ldots ,J\), and \(\sum p_{j}=1\)For comparison, the same approach can be applied to the actual unemployment rate:From Eqs. (7) and (8), it follows that equal inflow and exit rates across age-groups would make weighting by population shares unnecessary. However, since age-specific unemployment rates differ, their flow rates will also differ.

In this section, I focus on transition rates and the non-steady-state approach. Additional results for transition probabilities and steady-state contributions are provided in the Appendix. In general, the time aggregation bias is very small. Hence the results do not change much when I use transition probabilities instead of transition rates. The steady-state’s flow contributions are somewhat different from the non-steady-state case results. This means that the steady-state is helpful for calculating average values but less appropriate for assessing the contributions to dynamics.

Table 1 provides average values for the arrival rates, equilibrium unemployment rates, and actual unemployment rates for each age-group and all unemployed from January 2007 to December 2014. All variables are weighted with age-group-specific labor force size. Several findings are striking. The arrival rates decline with increasing age of the groups.⁸ This applies to all measures. For exit transitions, non-activity transitions are always higher than job finding transitions. From this, it follows that an unemployed person is more likely to move to inactivity than (back) into employment for all age-groups. For inflow transitions, I find that activity transitions are always higher than transitions of separation. Since activity flows are related to the same denominator as separation flows, we can conclude for all age-groups that a person is more likely to move from inactivity to unemployment than from employment to unemployment. Hence, the stock of the unemployed is more affected by activity/non-activity flows than by separation/job finding flows. On average, the proportion of non-activity flows in all exit flows and the proportion of activity flows in all inflows are roughly 60%.

Considering age-groups seem important since transitions are more than two times larger for the youth than the group 55-64. Concerning extreme values, the unemployed aged 15-24 have about four times higher probability of finding a job within the next month than those aged 55-64 years. In contrast, the youth’s probability of losing a job within the next month is only 2.2 times larger.

The last two columns of Table 1 show average values for the equilibrium unemployment rate and the actual unemployment rate. In most cases, particularly for prime-age-groups, the equilibrium rate is near the actual one. From this, we can conclude that age-group differences in transition rates greatly approximate the unemployment rate differences.

Table 2 provides further information for different regional subgroups on all unemployed and groups 15 to 24 and 55 to 64 years. On the left of Table 2, we compare regions with a labor force share of the youth with at least one standard deviation above the mean and at least one standard deviation below the mean. The same procedure is done for older workers. For regions with a larger youth share (first column), we find higher numbers for the age-groups and all unemployment for exit and lower numbers for inflow. As a consequence, unemployment rates are below average. The pattern is precisely the reverse when the regional youth’s share is below the average (second column). These findings align with Shimer’s (2001) assumption that the youth’s share positively affects job creation. When regions with a larger and smaller share of older workers are considered, we find values near the average rates. For the youth and all unemployed, more adverse conditions are present. In contrast, the labor market conditions worsen for the older group when the 55-64 group share is large. We can conclude that older workers compete more with other older workers than with the youth. Interesting is the finding that inflow transitions are lower in regions with a large youth share than in areas with a large percentage of older workers. In the latter case, the flow rates are similar to the group averages in Table 1.

The right side of Table 2 provides results for East and West German counties and three different types of regions: metropolitan areas (type I), urban areas (type II), and rural areas (type III). For counties in East Germany, inflow into unemployment is higher, while exiting is below the average. Hence, the adverse labor market conditions in the East can be explained by inflow and exit. The results for the West German regions go in the opposite direction, and average unemployment rates are about one percentage point below the average for all counties. Unemployment is lower in urban regions, with above-average exits and below-average inflows. However, the dynamics are higher in rural counties, with unemployment rates equal to the average for the youth and all unemployed. Overall, exit rates differ more between the three types of regions, and inflow rates more between east and west regions.

Due to the differences in exit and inflow transitions between youth and older workers, unemployment turnover must be lower, and unemployment spells must be longer for older workers. Table 3 provides the distribution of unemployment by duration. Each row sums up to 1 and shows the percentage distribution by duration for each considered age-group. Long-term unemployment is, by definition, a duration of 12 months or more. For the age-group 55-64 years, this applies to almost 50%, and within three months, only 20% of the stock will leave unemployment. In contrast, almost 57% of the age-group 15-24 years leave unemployment within the first three months, and only 8.8% of them are, on average, long-term unemployed. A closer look at the other groups indicates that the shares within the different duration categories first fall and then rise with increasing unemployment duration. These findings precisely align with the arrival rates provided in Table 1.

The monthly transition rates for all unemployed are displayed in Fig. 2.⁹ Job finding and separation rates primarily move in the opposite direction. The separation rate spikes are always in January. This is related to limited contracts that end primarily in December. The data show that the financial crisis (period of recession between the two vertical grey lines) does not affect the German labor market flows much because of the labor market policy (mainly due to short-time compensation, wage moderation, and flexible working time accounts).¹⁰ The recession covers the period 4th quarter of 2008 to the 4th quarter of 2009. During this period, employment declines by 0.5% only. The unemployment rate rose from 7.8% in 2008 to 8.1 in 2009. After the financial crisis, all series decline somewhat except the non-activity rate. For the exit flows, we observe an increasing discrepancy between job finding and non-activity, but both series are correlated moderately. Concerning the unemployment rate, separation is positively correlated, and non-activity is negatively correlated, both as expected. However, the remaining two flows do not provide a clear correlation pattern.

This article examined the relative flow contributions to different age-groups’ unemployment dynamics using regional administrative data for Germany. I consider a three-state model that allows for different flows from/to unemployment and apply steady-state and non-steady-state approaches. The overall contribution of inflow and outflow rates is 50%:50% using the non-steady-state approach.

The dynamics arising from the inactivity and activity flow account for at least 20% of unemployment dynamics. More than 80% of unemployment dynamics can be explained by activity, and inactivity flows for the youth. Across five age-groups, I find remarkable differences in flow contributions. For example, the relative contributions of separation and job finding rates are more significant for older workers than for young ones, while the activity/non-activity flows have a more significant relative contribution for younger workers. The contribution of separation rates is more important than the contribution of job find rates for all age-groups. When the distribution of age-groups is considered, we find low unemployment rates and fewer labor market participation fluctuations for regions with above-average youth shares. Also, agglomeration areas suffer more from structural labor market problems than urban or rural areas.

Based on the results, I conclude that labor force aging lowers labor market dynamics in the future. All arrival rates will decline on average if the share of older workers increases and the share of young workers declines. Compared to prime-age workers, older workers have lower inflow and exit rates. In addition, older workers’ unemployment dynamics will experience more inflow contributions and fewer exit contributions. Finally, unemployment turnover will decrease in the future, and an economic downturn will lengthen labor market shocks.