## Jacqueline O'Reilly, Janine Leschke, Renate Ortlieb, Martin Seeleib-Kaiser, and Paola Villa

Print publication date: 2018

Print ISBN-13: 9780190864798

Published to Oxford Scholarship Online: January 2019

DOI: 10.1093/oso/9780190864798.001.0001

Show Summary Details
Page of

PRINTED FROM OXFORD SCHOLARSHIP ONLINE (www.oxfordscholarship.com). (c) Copyright Oxford University Press, 2019. All Rights Reserved. An individual user may print out a PDF of a single chapter of a monograph in OSO for personal use. date: 22 August 2019

# How do youth labor flows differ from those of older workers?

Chapter:
(p.195) 7 How do youth labor flows differ from those of older workers?
Source:
Youth Labor in Transition
Publisher:
Oxford University Press
DOI:10.1093/oso/9780190864798.003.0007

# Abstract and Keywords

This chapter analyzes youth labor market dynamics, their structure, and their policy implications, focusing on selected European Union countries during the various stages of the Great Recession and comparing flows between labor market statuses for young people (aged 1634 years) with those for prime-age individuals (aged 3554 years). The flow approach views labor market transitions as a state-dependent process, simultaneously involving all movements of individuals between employment, unemployment, and inactivity. The main result is that young workers are more likely to move between employment and unemployment in both directions. This is instructive for assessing the gap in the labor market prospects of the two age groups and particularly for understanding differences in the evolution of youth and prime-age unemployment rates. The socioeconomic determinants of transitions between employment and unemployment in both directions are estimated, with the aim of illustrating the depth of age-based labor market segmentation and marginalization.

# 7.1. Introduction

This chapter analyzes youth labor market dynamics, their structure, and their policy implications. We focus on selected European Union (EU) countries (Austria, France, Poland, and Spain) during the various stages of the Great Recession (20082009, 20102011, and 2012), comparing the results for young people (aged 1634 years) with those for prime-age individuals (aged 3554 years). The choice of countries is based on two criteria: (1) sufficient differences in youth labor market performance and/or in labor market regulations1 and (2) the availability/quality of data.2 We concentrate on the possible presence of common trends across all the countries analyzed.

Our aim is to provide new evidence regarding differences between youth and prime-age labor market dynamics, thus calling attention to the overall presence of age-based labor market segmentation and even marginalization. To this end, we apply (1) the flow approach toward labor market dynamics (Blanchard and Diamond 1990; Elsby, Smith, and Wadsworth 2011) and (2) an analysis of the socioeconomic determinants of transitions in both directions between employment and unemployment (D’Addio 1998; Kelly et al. 2013; Flek, Hála, and Mysíková 2015).

Our analysis is based on an exploration of longitudinal data from the European Union Statistics on Income and Living Conditions (EU-SILC) in an innovative way. In Sections 7.2 and 7.3, we argue in detail that existing flow analyses based on longitudinal data lack comparisons across EU countries because of data (p.196) limitations. For the same reason, they typically concern working-age populations as a whole rather than just youth. The cross-national analysis based on longitudinal data developed in this chapter represents a research novelty, but we must admit that it still far from constitutes full representativeness.

In general, youth labor market dynamics should be more pronounced compared to those of prime-age groups for many reasons. First, young people move relatively more frequently between the labor market and inactivity. In addition, two other key factors are worth noting: (1) Matching difficulties in the early years of working life lead to frequent job changes, with repeated unemployment spells in between; and (2) investment in firm-specific human capital is lower for young people; hence, when layoffs occur, the last-in, first-out (LIFO) rule is frequently applied (Bell and Blanchflower 2011). For the period of the Great Recession, there is still a lack of studies comparing the quantitative dimension of youth and prime-age labor market dynamics.

The flow approach views labor market transitions as a state-dependent process that simultaneously involves all the movements (flows) of individuals between employment, unemployment, and inactivity. It enables us to quantify the overall degree and structure of labor market dynamics over time, across countries, and for various age groups. We address the degree of difference between the gross flows and flow transition rates (transitional probabilities of moving from one labor market status to another) of young and prime-age individuals. The results should be instructive for assessing the gap between the labor market prospects of the two age groups.

The flows between labor market statuses, particularly between employment and unemployment, determine variations in unemployment rates (Petrongolo and Pissarides 2008). We focus on the link between the different unemployment performances in various countries, age groups, and periods and the concrete flow, which contributes decisively to the observed differences in the evolution of unemployment rates. Thus, our research results based on the “flow” decomposition of unemployment rate dynamics should be helpful for understanding differences in the evolution of youth and prime-age unemployment rates.

Research on youth labor market dynamics concentrates on school-to-work transitions (for an overview and/or most recent findings, see Albert, Toharia, and Davia 2008; Berloffa et al., this volume; Hadjivassiliou et al., this volume). We prefer instead to combine the flow approach outlined previously with a detailed analysis of the socioeconomic determinants of transitions between employment and unemployment. Our previous research (Flek and Mysíková 2016) shows that the flows in both directions between employment and unemployment are actually decisive for the overall youth labor market dynamics during the Great Recession.

When estimating the determinants of a likelihood of exiting employment and becoming unemployed, we intend to verify the significance of age, in particular. Furthermore, we estimate the determinants of moving from unemployment to (p.197) employment, with an emphasis on the length of previous unemployment. With increasing unemployment duration, the unemployed are likely to be stigmatized and/or discouraged from further job search. Job-finding prospects may therefore be viewed as a diminishing function of unemployment duration, net of other socioeconomic characteristics of the unemployed (Machin and Manning 1999; Shimer 2012). Based on our results, we suggest country-specific adjustments in youth unemployment policy agendas.

To summarize, the chapter addresses the following key research questions:

1. 1. How do youth labor market dynamics (expressed by the movements of young people between employment, unemployment, and inactivity) differ from the dynamics of the prime-age individuals?

2. 2. Do the most marked differences between the evolutions of the youth and the prime-age unemployment rates lie in a relatively different exposure to job loss, in the prospects for exiting unemployment, or in transitions between inactivity and the labor market?

3. 3. Does the age of a worker significantly affect the probability of job loss followed by unemployment? Or is the impact of the age variable actually offset by variables such as work experience or education?

4. 4. How do job search durations vary between young and prime-age unemployed persons? At which unemployment duration does the job-finding probability of an unemployed person drop significantly and become already comparable to the gloomy employment prospects of a long-term unemployed individual?

Section 7.2 provides a literature overview with a deeper foundation of our research questions. We outline our methodological approach in Section 7.3 and also describe how we conduct cross-national comparative flow analyses using longitudinal EU-SILC data. In Section 7.4, we focus on the youth and prime-age flows and flow transition rates. This section continues with decompositions of unemployment dynamics and the identification of the driving forces (flows) that account for the different evolution of youth and prime-age unemployment rates. In Section 7.5, we analyze the determinants of youth and prime-age flows in both directions between employment and unemployment. Section 7.6 concludes the chapter.

# 7.2. Literature overview

The literature provides us with various partial arguments pointing to the specificity of youth labor market dynamics. Only a small fraction of young labor market entrants immediately manage to find stable and satisfactory employment. The rest are first faced with unemployment or with frequent job changes (p.198) combined with repeated unemployment spells (for recent evidence, see Berloffa et al., this volume). This situation is often attributed to educational mismatch, to a lack of work experience, or to the absence of firm-specific skills on the part of young workers (International Labour Organization (ILO) 2013; McGuinness, Bergin, and Whelan, this volume).

The position of young adults in the labor market is more dynamic than that of prime-age participants even when education, skills, and other characteristics match the employer’s requirements. Young employees are still more likely to be subject to layoffs—through the practice of fixed-term labor contracts or because of the LIFO rules and seniority-weighted redundancy payments (Bell and Blanchflower 2011). Higher outflows from employment to unemployment compared to those for the prime-age segment of the workforce thus indicate that young workers actually constitute a marginalized group in the sense established by Reich, Gordon, and Edwards (1973).

As shown most recently by Elsby et al. (2011), young people are also characterized by a relatively higher frequency of outflows from unemployment into employment and by shorter unemployment spells compared to the prime-age segment. Such a seemingly positive tendency is likely to be associated with the lower reservation wages of young unemployed, with their acceptance of less stable or less significant jobs, and with lower redundancy costs linked with their future layoffs (Blanchard 1999; Berloffa et al., this volume). Thus, the relatively high outflows of young people from unemployment into employment are again closely linked with a notion of youth as a marginalized group: Young people appear to be forced to accept jobs prevailingly on secondary labor markets, with frequent and relatively brief unemployment episodes in between.

Despite the reasonably good and varied amount of findings collected so far, we believe that an accurate, cross-national view on youth labor market dynamics during the Great Recession is still largely missing. This concerns the absence of a synthetic measure of such dynamics and their structure, as well as comparisons with the labor market dynamics of prime-age individuals. The flow approach seems a promising way to fill that gap. However, the existing longitudinal flow literature lacks comparisons across countries because of data limitations. Instead, it explores national data sources such as Labor Force Surveys (Gomes 2009; Elsby et al. 2011). Also, except for the latter authors, such flow analyses concern working-age populations as a whole rather than various age groups.

Elsby et al. (2011) deal explicitly with youth flows in the United Kingdom and report a higher youth labor market dynamics compared to the prime-age group. Higher youth outflow rates from employment to unemployment and vice versa appear to be in line with the theoretical assumptions of Reich et al. (1973) and Blanchard (1999). These rates confirm the presence of an age-based labor market segmentation and the marginalized status of young workers in the United Kingdom. Flek and Mysíková (2015) and Flek et al. (2015) address youth flows in the Czech Republic and provide some comparisons with neighboring (p.199) countries and/or Spain, with similar conclusions to those reported by Elsby et al. (2011). Given the relatively small number of such studies, European youth flows still need to be analyzed in a broader cross-country perspective.

The Great Recession exacerbated the difficulties for young people on the labor market, creating a situation in which youth unemployment rates increased faster than prime-age unemployment rates (ILO 2013). Because the flows of workers between labor market statuses determine variations in unemployment rates (Petrongolo and Pissarides 2008; Dixon, Freebairn, and Lim 2011; Elsby et al. 2011; Shimer 2012), a link can be derived between the different unemployment performances in various countries, age categories, or periods, on the one hand, and the concrete flow (which contributes decisively to the observed differences), on the other hand. Labor market and activation policies should then focus on that particular flow. As noted by Elsby et al. (2011), “Policy that focused on encouraging outflows from unemployment may not be as relevant in an economy in which rises in unemployment were driven by changes in the rate of outflows from employment” (p. 4).3

When estimating the socioeconomic determinants of moving from employment to unemployment, we use a standard binary probit model. Kelly et al. (2013) use an analogous approach for analyzing the outflows from youth unemployment to employment in Ireland during the Great Recession. Our main aim is to verify the presence of age-based labor market segmentation, based on the higher exposure of young people to job loss followed by unemployment.

A negative relationship between job-finding probability and unemployment duration is referred to in the literature as the “duration dependence” (Machin and Manning 1999; Shimer 2012). We plan to verify its presence in both age categories of unemployed by performing estimates based on the discrete-time proportional hazard models developed by Cox (1972) and Jenkins (1997). Among others, Albert et al. (2008) use such models when analyzing school-to-work transitions in Spain. Other examples include retirement decisions in the United Kingdom (Disney, Emmerson, and Wakefield 2006) and employment decisions after the birth of the first child in Spain (Davia and Legazpe 2014). To our knowledge, there have been only two attempts to explore this model for analyzing exits from youth unemployment into employment (D’Addio 1998; Flek et al. 2015), and both of these point to the significance of duration dependence. As with the flow analysis, however, cross-country comparisons and comparisons between age groups are still scarce.

# 7.3. Data and methodology

To our knowledge, this chapter represents one of the first attempts to use the matched longitudinal monthly data of the EU-SILC database for a comparative analysis of youth and prime-age labor market flows in Europe. Being relatively (p.200) new, this approach requires some initial description of the data, followed by methodological notes on estimation strategies.

## 7.3.1. Labor Market Dynamics and Flow Decomposition of Unemployment Rates

Some European labor markets that are potentially suitable for reference purposes, such as that of Germany, were not yet included in the versions of longitudinal EU-SILC data sets used in our analysis. Some countries, typically Scandinavian, collected some of the relevant variables only for one selected person per household. In other potentially interesting cases, such as the United Kingdom, there were other technical obstacles to the results being comparable (e.g., a high share of missing information on monthly economic activity).

We deal with young people aged 16–34 years at the beginning of all analyzed periods (2008–2009, 2010–2011, and 2012). The prime-age population, aged 35–54 years, represents a reference group. The choice of the age interval 16–34 years to represent young people is relatively straightforward in that any study aimed at youth labor market dynamics has to involve at least the early stages of work careers of young people, including university graduates. Where appropriate, we decompose the youth age band into various subgroups (16–19, 20–24, 25–29, and 30–34 years) so as to examine the possible heterogeneity of this age group.

The EU-SILC data explored in Section 7.4 consider an individual as the unit of analysis. Only the respondents with full survey participation over the chosen subperiods have been selected for analysis. Our national subsamples are therefore pure panels, where all the reported month-to-month individual labor market statuses are matched. We use the longitudinal weights provided by Eurostat specifically for these subsamples—the standard means of minimizing the attrition bias. Regarding the calendar bias, we hope to avoid it by averaging the observed status changes over the subperiods analyzed.4 Nonetheless, the retrospective nature of reports on economic activity and their self-declared character may lead to deviations from the ILO definition of unemployment.

We extract a 2-year period from longitudinal EU-SILC 2010 (version 5 of August 2014), which covers monthly economic activity for January 2008 through to December 2009, and another 2-year period from longitudinal EU-SILC 2012 (version 1 of August 2014), which includes data for January 2010 through to December 2011. Both of these subsamples provide chains of 23 monthly individual labor market statuses (i.e., employment, unemployment, and inactivity) and contain far more respondents than a single, 4-year panel of EU-SILC. We also add data for January through December 2012, from EU-SILC 2013 (version 2 of August 2015). The chains of monthly labor market statuses for a single year are obviously shorter (and thus less suitable for longitudinal analysis than the 2-year subsamples), but they enable us to incorporate the year 2012 into the analysis.

(p.201) In the past month, an individual could be employed $(Et−1)$, unemployed $(Ut−1)$, or inactive $(It−1)$. In the current month, he or she can remain in an unchanged labor market status5 or change it as follows:$(Et−1→Ut);(Et−1→It);(Ut−1→Et);(Ut−1→It);(It−1→Et);(It−1→Ut).$ Thus, the individual may move from previous to current status in six ways, and the corresponding numbers of individuals represent six gross labor market flows. Figure 7.1 in Section 7.4 compares the relative involvement of young and prime-age individuals in gross flows, where $UE=(Ut−1→Et)/(Et−1+Ut−1+It−1)$ and so forth for EU, EI, . . . . This approach represents a standard proxy for aggregate and/or group-specific labor market fluidity (Blanchard and Diamond 1990).

In contrast, transitional probabilities λ‎ presented later in Section 7.4 (see Figures 7.2 and 7.3) represent a first-order Markov process, where the probability of moving from the previous to the current status depends exclusively on the individual’s previous status (Blanchard and Diamond 1990). For instance, an unemployed individual’s average monthly job-finding probability is$λUE=(Ut−1→Et)/Ut−1.$

Next to this, we express a net change in unemployment in terms of the corresponding average monthly gross flows “in” $(Et−1→Ut;It−1→Ut)$ and “out” of unemployment $(Ut−1→Et;Ut−1→It)$, which are additionally rearranged as a product of the respective transition probability rate λ‎ and the labor market stock (E, U, I) at time (t – 1). A monthly average change in unemployment rate in percentage points is then decomposed into the contributions of the “ins” and the “outs” of unemployment. The third term shows the contribution of changes in the labor force to unemployment rate dynamics. Such a decomposition of unemployment rate dynamics was developed by Dixon et al. (2011) and applied in a slightly modified form also by Flek and Mysíková (2015). Table 7.1 in Section 7.4 reports results separately for the evolutions of the youth and prime-age unemployment rates.

## 7.3.2. Assessing the Determinants of Transitions Between Employment and Unemployment

In Section 7.5, the estimates consider an (un)employment spell as the unit of analysis, including multiple episodes. This leads to a different data organization, which is based on nonweighted subsamples. It must be admitted that a data organization of this kind makes the results potentially more prone to the calendar and/or attrition bias than in the case of the flow approach (presented in Section 7.4), which considers the individual as the unit of analysis. We concentrate initially on the determinants of transitions from employment to unemployment by using a probit model. In the 2-year subperiods, we extract from nonweighted samples all employment spells occurring at any time between the first month of the first year (January 2008 or January 2010) and the beginning of the second year (January 2009 or January 2011). For 2012, we concentrate on employment (p.202) recorded in the first month (January 2012). For all of the three subperiods, we ascertain whether or not the transitions into unemployment occur during the following 11 months.

The dependent binominal variable in a probit model equals 1 if an employment status transitioned to unemployment during the observed period, and 0 otherwise. The individual and other characteristics (e.g., age, gender, education, work experience, household size, and population density) stand for independent explanatory variables. We report results in the form of average marginal effects for pooled samples (with a dummy for prime age) and then separately for the two age groups. Among a range of potentially relevant determinants, we do not analyze the impact of previous employment duration. We are aware that the length of an employment spell can affect the probability of losing a job (e.g., because of LIFO) but, unfortunately, job tenure is not available in the data. Instead, we capture the intensity of employment by years of work experience as a regressor.

For the duration model estimates, we collect all unemployment spells in our three nonweighted subsamples. As with the probit estimates, we refer later for simplicity to individuals, although some of them experienced multiple unemployment spells. The data used for estimations are naturally censored. We introduce a censoring indicator “1” if an unemployment spell terminates in employment and “0” in all other cases.6

The econometric analysis is developed in two steps. As the first step, we explore the Kaplan–Meier (KM) estimator (Kaplan and Meier 1958), which represents a nonparametric estimate of the survival function. For our purposes, “survival” means the duration of unemployment; time is measured in months. At this stage, we do not account for individual or other characteristics. Instead, we simply assume that the KM survival curves will decline over time in line with the emergence of closed spells that end in a move into employment. Log-rank tests would document how (in)significantly the KM curves for young and prime-age unemployed differ—in other words, whether the duration of job search differs significantly between the two age categories.

Second, we apply a discrete-time proportional hazard model (Cox 1972).7 Our idea is to detect the “true” duration dependence of unemployment. Note that unemployed workers with “bad” characteristics (low education, etc.) tend to be less employable than those with “good” characteristics. This is likely to apply to unemployment spells of any length and leads to a selection of individuals with “bad” characteristics into long-term unemployment. But such “duration dependence” is actually spurious because it is explained by other variables and not by unemployment duration per se. In contrast, unemployment duration in and of itself may negatively affect the job-finding probability of unemployed individuals—even after controlling for their available characteristics and unobserved heterogeneity—because of the stigmatization and discouragement effects of long-term unemployment.

We assume that the baseline hazard function is piecewise constant in the chosen unemployment duration intervals (1–2 months, 3–4 months, 5–6 months, 7–10 months, 11–15 months, and 16–24 months), whereas the vector (p.203) of covariates in the model equation indicates the impact of explanatory variables on the probability of moving in a randomly chosen time from unemployment into employment. For the sake of better interpretation, the estimated coefficients are transformed into hazard ratios. For the periods 2008–2009 and 2010–2011, unemployment spells lasting between 16 and 24 months represent a reference duration interval. For 2012, a reference interval stands for unemployment spells lasting between 11 and 12 months. The set of chosen covariates is analogous to the previous probit analysis.8

# 7.4. Labor market dynamics and their age-based specificity

## 7.4.1. Comparing the Youth and Prime-Age Gross Flows

Figure 7.1 reports gross flows for young (aged 1634 years) and prime-age (aged 3554 years) individuals in the four countries considered and suggests the presence of country-specific and age-based patterns in labor market dynamics during the Great Recession. The results are presented as percentages, where, for instance, UE relates the average monthly number of individuals involved in a gross flow from unemployment to employment to total labor market stocks (and so forth for EU, EI, . . .).

In both age groups, Austria and Spain record persistently higher aggregate fluidity of their labor markets compared to France and Poland. Viewed from another perspective, all the countries involved in our analysis display an approximately two or three times lower degree of aggregate fluidity of their labor markets compared to the United States and the United Kingdom, where between 5% and 7% of the working-age population change their labor market status every month or quarter (Gomes 2009). In this respect, our results are in line with the general view, which considers the Anglo-American labor markets to be considerably more fluid than the labor markets in Continental, Southern, or Eastern Europe.9

Figure 7.1 shows that young people are relatively more involved in gross flows compared to prime-age individuals. This holds true uniformly across all the analyzed countries and periods (1: 20082009; 2: 20102011; and 3: 2012). Thus, on aggregate, the position of young people on the labor market is much less stable. This result confirms the observations of Elsby et al. (2011) for the United Kingdom, who also established that young people “churn” through the labor market relatively more frequently.

The structure of the gross flows of young people is different from that of prime-age individuals. Whereas in the latter case, gross flows between employment and unemployment (UE; EU) are almost the only source of dynamics, the youth flows involve a relatively higher frequency of transitions between inactivity and the labor market (IE; EI; UI; IU). This is fully in line with intuition, (p.204) (p.205) given that young people naturally tend to enter or exit the labor market relatively more frequently, particularly because of beginning/finishing education or because of taking/finishing parental leave.

The relatively higher frequency of transitions between inactivity and the labor market is not the sole specificity of youth labor market dynamics. The relative share of youth flows between employment and unemployment in both directions (EU; UE) is actually also higher compared to that of prime-age individuals. Figure 7.1 reveals that these two flows typically account for more than 50% of the entire youth labor market dynamics. Only Austria deviates persistently from this tendency because of its exceptionally high shares of youth transitions from employment to inactivity and vice versa (EI; IE).

Figure 7.1 Gross flows as percentages of total matched labor market stocks in four European countries (monthly averages; period 1: 2008–2009; period 2: 2010–2011; period 3: 2012; youth: 16–34; prime age 35–54).

Sources: EU-SILC longitudinal UDB 2010, version 5 of August 2014; EU-SILC longitudinal UDB 2012, version 1 of August 2014; EU-SILC UDB 2013, version 2 of August 2015; authors’ computations.

## 7.4.2. The Youth and Prime-Age Transition Rates

Figures 7.2 and 7.3 present transition rates from employment to unemployment ($λUE$) and from unemployment to employment ($λUE$) for young and prime-age individuals for the three periods and the four countries considered.

Figure 7.2 Transition rates from employment to unemployment for various age groups in four European countries (monthly averages, in %)

Sources: EU-SILC UDB 2010, version 5 of August 2014; EU-SILC UDB 2012, version 1 of August 2014; EU-SILC UDB 2013, version 2 of August 2015; authors’ computations.

Figure 7.3 Transition rates from unemployment to employment for various age groups in four European countries (monthly averages, in %).

Sources: EU-SILC UDB 2010, version 5 of August 2014; EU-SILC UDB 2012, version 1 of August 2014; EU-SILC UDB 2013, version 2 of August 2015; authors’ computations.

The values of $λEU$in Figure 7.2 confirm that a young worker (aged 1634 years) is more likely to become unemployed than a prime-age worker (aged 3554 years). This finding stems from comparisons of the last two columns (i.e., of the two age groups) for each country and applies uniformly to the four countries and the three periods analyzed, irrespective of the existing institutional differences, different unemployment performances, or other national specificities. A disproportionally high exposure of young workers to unemployment appears to be a general phenomenon, suggesting the overall presence of an age-based segmentation and marginalization on European labor markets.

Figure 7.2 also documents heterogeneity in the risk of becoming unemployed within the youth age band (1634 years). The lowest age categories (1619 and/or 2024 years) face the highest risk of becoming unemployed. But this is not to say that as the age of young workers increases, their risk of becoming unemployed becomes fully comparable with that of prime-age workers. Even the upper youth age category (3034 years) typically faces a relatively higher risk of becoming unemployed compared to prime-age workers.

The job-finding rates $(λUE)$in Figure 7.3 suggest that, with the sole exception of Austria in 20082009, a young unemployed person is relatively more “attractive” than a prime-age individual when firms hire new workers. This applies also to the job-finding rates $(λIE)$ of previously inactive young people (see Figures A7.1–A7.4 in the Appendix). As argued in more detail in the literature overview in Section 7.2, such tendencies will probably again label youth as a marginalized group, forced to accept less stable employment conditions compared to the prime-age segment of the workforce, with frequent subsequent transitions back into unemployment. (p.206) (p.207)

(p.208) One would presume that the lowest age categories of young unemployed transition back into education (inactivity) more frequently than the upper categories. This alternative transition channel should help them avoid remaining in unemployment and increase their job-finding chances in the future. However, our results presented in the Appendix suggest that even the lowest age categories of young unemployed remain mostly dependent on the labor market and that their transitions to education (inactivity) cannot be viewed as a relevant alternative.10

Our findings confirm the observations of Elsby et al. (2011) for the United Kingdom. The established (age-based) gaps in both job-loss and job-finding rates can be interpreted as typical features of marginalized groups in the sense of Reich et al. (1973) or Blanchard (1999). The results point to the need for additional policy measures aimed at higher employment stability and a better quality of jobs held by young people rather than at merely increasing their outflow rates from unemployment (inactivity) to employment.

## 7.4.3. Flow Decomposition of Unemployment Rate Dynamics

Table 7.1 decomposes changes in the unemployment rate for the four countries considered (AT, ES, FR, and PL) in terms of both movements into unemployment (from employment or inactivity) and movements out of unemployment (into employment or inactivity) over the periods 2008–2009, 2010–2011, and 2012. The results in the second column demonstrate the trend of disproportionate increases in youth unemployment rates compared to prime-age unemployment rates in the initial period of the Great Recession (20082009). As seen in the fourth column, these disproportionate increases in youth unemployment in 20082009 (in ES, FR, and PL) were generated decisively by inflows into unemployment from employment, which accounted for far higher increases in youth unemployment rates than in prime-age unemployment rates.

Table 7.1 Unemployment rate dynamics of young people (aged 16–34 years) and prime-age individuals (aged 35–54 years) in four European countries in 2008–2009, 2010–2011, and 2012 (monthly averages, in percentage points)

Country/period

$△( U LF )$

“Ins” (+)

$λ EU I t−1 L F t$

$λ IU E t−1 L F t$

“Outs” (–)

$− λ UE U t−1 L F t$

$− λ UI U t−1 L F t$

Contribution of changes in LFa

AT 2008–2009

Prime age

0.0311

0.7845

0.7297

0.0548

–0.7599

–0.6685

–0.0914

0.0065

Youth

–0.1660

1.3498

1.1001

0.2498

–1.5124

–1.1797

–0.3327

–0.0034

AT 2010–2011

Prime age

–0.0788

0.6160

0.5533

0.0628

–0.6992

–0.6127

–0.0866

0.0044

Youth

–0.2594

1.2005

0.9697

0.2308

–1.4543

–1.1670

–0.2873

–0.0056

AT 2012

Prime age

–0.0012

0.5719

0.5354

0.0364

–0.5743

–0.5180

–0.0563

0.0012

Youth

0.0408

1.0867

0.8436

0.2431

–1.0120

–0.9025

–0.1095

–0.0339

ES 2008–2009

Prime age

0.3080

1.1727

1.0535

0.1193

–0.8641

–0.7768

–0.0873

–0.0006

Youth

0.4432

1.9669

1.6931

0.2737

–1.4785

–1.3286

–0.1499

–0.0452

ES 2010–2011

Prime age

0.1791

1.1002

0.9189

0.1813

–0.9041

–0.8088

–0.0953

–0.0170

Youth

0.1446

1.8244

1.4199

0.4045

–1.5977

–1.3550

–0.2426

–0.0821

ES 2012

Prime age

0.0926

1.1173

1.1089

0.0084

–1.0321

–1.0091

–0.0230

0.0074

Youth

0.0036

1.5550

1.4727

0.0823

–1.5106

–1.4612

–0.0494

–0.0408

FR 2008–2009

Prime age

0.0916

0.4532

0.4292

0.0240

–0.3623

–0.3276

–0.0347

0.0007

Youth

0.1252

1.1087

0.9230

0.1858

–0.9255

–0.8313

–0.0942

–0.0581

FR 2010–2011

Prime age

–0.0407

0.3760

0.3504

0.0255

–0.4153

–0.3890

–0.0263

–0.0014

Youth

–0.0311

1.1350

0.8799

0.2551

–1.0941

–1.0059

–0.0881

–0.0721

FR 2012

Prime age

0.0242

0.4447

0.4247

0.0200

–0.4196

–0.3879

–0.0317

–0.0009

Youth

0.1678

1.3437

1.0912

0.2525

–1.1115

–1.0138

–0.0977

–0.0645

PL 2008–2009

Prime age

0.0648

0.4748

0.3910

0.0838

–0.4088

–0.3597

–0.0491

–0.0012

Youth

0.1974

0.9511

0.6529

0.2982

–0.6974

–0.6299

–0.0675

–0.0563

PL 2010–2011

Prime age

0.0238

0.5294

0.4684

0.0611

–0.5063

–0.4518

–0.0545

0.0007

Youth

–0.0214

0.9993

0.6961

0.3032

–0.9643

–0.8663

–0.0980

–0.0564

PL 2012

Prime age

0.0151

0.3902

0.3626

0.0276

–0.3796

–0.3475

–0.0321

0.0046

Youth

0.0855

0.8987

0.6498

0.2488

–0.7501

–0.7075

–0.0426

–0.0632

aComputed as $U t−1 ( 1 L F t − 1 L F t−1 )$. The results are affected by rounding.

Sources: EU-SILC UDB 2010, version 5 of August 2014; EU-SILC UDB 2012, version 1 of August 2014; EU-SILC UDB 2013, version 2 of August 2015; authors’ computations.

This is in line with our finding that the job-loss rates of young workers are persistently higher than those of prime-age workers. But in the fourth column of Table 7.1, we can see exactly how the inflows of workers into unemployment from employment contribute to the different evolutions of the unemployment rates of the two age groups. The contribution of inflows into unemployment from inactivity in the fifth column is also higher for young people, but this contribution to the different unemployment rate dynamics of the two age groups is much less relevant than the contribution of inflows of workers into unemployment from employment (likewise, the contribution of changes in the labor force in the last column is less relevant).

In contrast, had the outflows from unemployment to employment (in the seventh column in Table 7.1) been the only driver of unemployment rate dynamics, youth unemployment rates would actually have developed more favorably than prime-age unemployment rates. This confirms that the job-finding difficulties of the young unemployed cannot be viewed as the cause of disproportionate increases in youth unemployment rates in the initial stage of the Great Recession (20082009). (p.209) (p.210)

(p.211) After a short break in 20102011, when the development of youth unemployment rates started to resemble and sometimes even outperform prime-age unemployment rates, the most recent period covered by our data (2012) shows again the prevailing tendency of youth unemployment rates to increase more rapidly than prime-age rates. This could potentially be attributed to an only temporary effect of stimulus measures that were targeting the young unemployed disproportionally. Indeed, except for Spain, the 2012 balance of “ins” and “outs” reflects a new round of disproportionate increases in youth unemployment rates compared to prime-age unemployment rates. Similarly to 20082009, the main driver of these disproportions is seen in the fourth column in Table 7.1 and is embodied in a disproportionally high contribution of inflows of young workers from employment into unemployment.

Table 7.1 reveals the sources of different dynamics in youth unemployment rates. Surprisingly, the contributions of outflows from unemployment into employment in Spain and Austria were comparable in 20082009 and 20102011 (see the seventh column). The decisive source of strikingly different youth unemployment rate dynamics in these two countries was represented by a relatively much higher contribution of inflows of Spanish young workers into unemployment from employment (see the fourth column).

In 2012, France and Poland recorded the highest increases in youth unemployment rates. Both the stories behind these developments and the policy implications are somewhat different. In Poland, the only problem was embodied, at least in a given comparative perspective, in insufficient outflows from unemployment into employment (in the seventh column in Table 7.1). In contrast, France suffered simultaneously from relatively low “outs” and high “ins” of youth unemployment.

# 7.5. Determining factors of transitions between employment and unemployment

In this section, we provide an econometric analysis of the socioeconomic determinants of movements between employment and unemployment in both directions. In particular, we intend to verify within a multivariate framework the statistical significance of age for the risk of losing one’s job and becoming unemployed. Then we concentrate on unemployment durations within both age groups of interest with the aim of detecting the presence of duration dependence of unemployment, net of other individual and additional characteristics influencing the job-finding probability of an unemployed person.

## 7.5.1. Transitions from Employment to Unemployment

Tables 7.2a–7.2c evaluate the factors influencing the probability of losing one’s job and becoming unemployed. We present results for pooled samples of young and prime-age individuals in two specifications in the second and third columns. The first specification does not involve work experience and confirms (p.212) (p.213) (p.214) (p.215) (p.216) (p.217) (p.218) the significantly lower probability of prime-age workers losing their jobs and becoming unemployed. The second specification includes work experience—a variable that proves significant in all cases. With added controls for work experience, the previously established age-group effect weakens substantially; in some cases, it loses its significance or even reverses. Austria and Poland represent the most illustrative cases in that in these two countries, the controls for work experience change the sign of the age-group effect.

Table 7.2a Determinants of transitions of young people (aged 16–34 years) and prime-age individuals (aged 35–54 years) from employment to unemployment in four European countries: 2008–2009 (average marginal effects from probit model)

AT

ES

Pooled

Pooled

Youth

Prime age

Pooled

Pooled

Youth

Prime age

Prime age

–0.048***

0.044**

–0.131***

–0.022**

Male

0.033***

0.045***

0.027

0.054***

–0.032***

–0.008

0.004

–0.012

Tertiary education

–0.150***

–0.156***

–0.228***

–0.138***

–0.199***

–0.109***

–0.099***

–0.112***

Secondary education

–0.100***

–0.090***

–0.087***

–0.096***

–0.138***

–0.078***

–0.064***

–0.082***

Experience (in years)

–0.006***

–0.005**

–0.006***

–0.004***

–0.009***

–0.003***

HH size 1

0.082***

0.101***

0.137***

0.089***

a

a

a

a

HH size 2

0.072***

0.090***

0.121***

0.081***

0.010

0.010

0.002

0.014

HH size 3

0.027*

0.038***

0.073***

0.025

0.009

0.003

–0.002

0.006

Densely populated area

0.004

–0.008

0.025

–0.025*

–0.038***

–0.02**

–0.053***

–0.004

Medium-populated area

–0.017

–0.018

–0.003

–0.024

–0.011

0.003

–0.015

0.013

Pseudo R2

0.046

0.061

0.058

0.060

0.065

0.054

0.027

0.050

AUC

0.651

0.676

0.662

0.677

0.675

0.673

0.617

0.672

n

3,982

3,982

1,215

2,677

9,799

7,828

2,577

5,251

FR

PL

Pooled

Pooled

Youth

Prime age

Pooled

Pooled

Youth

Prime age

Prime age

–0.102***

–0.000

–0.065***

0.024**

Male

0.004

0.015**

0.013

0.015**

0.007

0.017***

0.017

0.018**

Tertiary education

–0.101***

–0.105***

–0.178***

–0.069***

–0.121***

–0.123***

–0.144***

–0.126***

Secondary education

–0.045***

–0.037***

–0.065***

–0.026***

–0.046***

–0.041***

–0.071***

–0.025**

Experience (in years)

–0.007***

–0.019***

–0.004***

–0.006***

–0.011***

–0.004***

HH size 1

0.067***

0.069***

0.038

0.077***

a

a

a

a

HH size 2

0.028***

0.042***

0.050**

0.033***

0.025***

0.040***

0.035*

–0.041***

HH size 3

0.017*

0.026***

0.030

0.020**

–0.011

0.000

–0.019

0.011

Densely populated area

–0.003

–0.01

–0.026

–0.004

–0.006

–0.000

–0.022

0.012

Medium-populated area

–0.013

–0.013

–0.058***

0.004

–0.006

–0.003

–0.029

0.007

Pseudo R2

0.061

0.095

0.083

0.065

0.036

0.063

0.053

0.059

AUC

0.682

0.725

0.700

0.694

0.631

0.688

0.666

0.683

n

7,449

7,449

2,387

5,018

8,782

8,694

3,097

5,597

(a) One- and two-person households are merged because of a low share of observations in the first category.

AUC, area under the curve; HH, household.

(*) p < .10.

(**) p < .05.

(***) p < .01.

Sources: EU-SILC UDB 2010, version 5 of August 2014; authors’ computations.

Table 7.2b Determinants of transitions of young people (aged 1634 years) and prime-age individuals (aged 3554 years) from employment to unemployment in four European countries: 2010–2011 (average marginal effects from probit model)

AT

ES

Pooled

Pooled

Youth

Prime age

Pooled

Pooled

Youth

Prime age

Prime age

–0.040***

0.034**

–0.115***

–0.011

Male

0.005

0.019*

0.024

0.024**

–0.021**

–0.001

–0.018

0.007

Tertiary education

–0.167***

–0.181***

–0.220***

–0.163***

–0.154***

–0.110***

–0.097***

–0.119***

Secondary education

–0.086***

–0.084***

–0.101***

–0.076***

–0.093***

–0.066***

–0.123***

–0.047***

Experience

–0.005***

–0.001

–0.006***

–0.005***

–0.006***

–0.005***

HH size 1

0.024

0.044***

0.029

0.049***

a

a

a

a

HH size 2

0.026**

0.045***

–0.001

0.064***

–0.013

0.000

–0.014

0.004

HH size 3

–0.012

0.000

–0.012

0.006

0.018*

–0.005

–0.010

–0.004

Densely populated area

0.014

0.005

0.033

–0.006

–0.050***

–0.022**

–0.036*

–0.017*

Medium-populated area

–0.004

–0.008

0.031

–0.023*

–0.030**

–0.002

–0.005

–0.001

Pseudo R2

0.044

0.063

0.045

0.072

0.052

0.060

0.028

0.062

AUC

0.652

0.686

0.641

0.695

0.661

0.682

0.622

0.690

n

4,057

3,972

1,274

2,698

7,735

6,344

1,763

4,581

FR

PL

Pooled

Pooled

Youth

Prime age

Pooled

Pooled

Youth

Prime age

Prime age

–0.104***

–0.006

–0.069***

0.034***

Male

–0.011*

0.002

–0.013

0.006

–0.030***

–0.016**

–0.033**

–0.007

Tertiary education

–0.108***

–0.113***

–0.196***

–0.076***

–0.173***

–0.170***

–0.160***

–0.185***

Secondary education

–0.043***

–0.038***

–0.078***

–0.023***

–0.074***

–0.062***

–0.056**

–0.061***

Experience

–0.006***

–0.019***

–0.004***

–0.006***

–0.014***

–0.005***

HH size 1

0.046***

0.053***

0.041

0.049***

a

a

a

a

HH size 2

0.025***

0.039***

0.034*

0.030***

0.027***

0.042***

0.052***

0.033***

HH size 3

0.015*

0.024***

0.020

0.017*

0.013

0.022***

0.025*

0.020**

Densely populated area

–0.021***

–0.014*

–0.029*

–0.007

0.003

0.004

–0.033**

0.026***

Medium-populated area

–0.003

0.002

–0.014

0.008

0.018**

0.021**

0.023

0.020**

Pseudo R2

0.068

0.105

0.099

0.061

0.040

0.069

0.066

0.065

AUC

0.692

0.737

0.726

0.685

0.644

0.688

0.682

0.689

n

7,774

7,736

2,387

5,349

8,464

8,407

3,071

5,336

(a) One- and two-person households are merged because of a low share of observations in the first category.

AUC, area under the curve; HH, household.

(*) p < .10.

(**) p < .05.

(***) p < .01.

Source: EU-SILC UDB 2012, version 1 of August 2014; authors’ computations.

Table 7.2c Determinants of transitions of young people (aged 1634 years) and prime-age individuals (aged 3554 years) from employment to unemployment in four European countries: 2012 (average marginal effects from probit model)

AT

ES

Pooled

Pooled

Youth

Prime age

Pooled

Pooled

Youth

Prime age

Prime age

–0.051***

0.014

Male

–0.024**

–0.020***

–0.004

–0.030**

0.004

Tertiary education

–0.067***

–0.067***

–0.083***

–0.107***

–0.075***

Secondary education

–0.019

–0.062***

–0.07***

–0.100***

–0.059***

Experience

–0.003**

–0.005***

–0.010***

–0.004***

HH size 1

0.010

a

a

a

a

HH size 2

–0.028*

0.019**

0.022***

–0.007

0.034***

HH size 3

–0.029*

0.007

0.010

–0.016

0.020**

Densely populated area

0.026*

–0.021***

–0.019***

–0.013

–0.021***

Medium-populated area

0.003

–0.022**

–0.018**

–0.004

–0.022**

Pseudo R2

0.044

0.031

0.051

0.045

0.047

AUC

0.668

0.634

0.667

0.648

0.666

n

b

b

1,596

b

9,003

8,927

2,424

6,503

FR

PL

Pooled

Pooled

Youth

Prime age

Pooled

Pooled

Youth

Prime age

Prime age

–0.053***

Male

–0.007

–0.025***

Tertiary education

–0.062***

–0.071***

Secondary education

–0.016**

–0.035**

Experience

–0.007***

HH size 1

0.019**

a

HH size 2

0.010

0.004

HH size 3

0.006

–0.013

Densely populated area

0.011**

0.005

Medium-populated area

0.009

0.003

Pseudo R2

0.054

0.049

AUC

0.678

0.676

n

8,629

b

b

b

b

b

3,702

b

(a) One- and two-person households are merged because of a low share of observations in the first category.

(b) For 2012, the share of employment spells transitioning into unemployment frequently amounts to less than 5%. Such results are omitted because of their presumably low representativeness.

AUC, area under the curve; HH, household.

(*) p < .10.

(**) p < .05.

(***) p < .01.

Source: EU-SILC UDB 2013, version 2 of August 2015; authors’ computations.

The pooled models show uniformly that higher education levels significantly diminish the likelihood of losing one’s job and becoming unemployed. In contrast, the effects of the remaining variables—such as gender, household size, or population density—are rather country specific or vary over time. Considering national specificities, it is worth noting that Spain is the only country in which gender has a significant effect on the probability of losing one’s job and becoming unemployed in all subperiods analyzed. Female workers in Spain thus have a higher probability of becoming unemployed compared to men.

With added controls for work experience, this gender-based difference becomes insignificant. Women’s lower work experience in Spain is thus responsible for their disadvantage in terms of sustaining employment and avoiding unemployment. This effect is not clearly apparent in any other country. In Austria, we can observe the opposite: Here, controls for work experience strengthen the men’s disadvantage.

Among young people, the gender effect (both with and without controls for work experience) usually has no significant impact on the probability of losing one’s job and becoming unemployed. This is not surprising because the gender difference in work experience cannot fully evolve at the beginning of working careers. But the gender gap in work experience may intensify during the life cycle, and women, especially in Spain, might suffer from a lack of such experience in the longer term.

The results of separate estimations for the two age groups also show that work experience significantly lowers the likelihood of losing one’s job and becoming unemployed and that this effect is in most cases more evident among young workers. Higher education likewise significantly reduces the probability of losing one’s job and becoming unemployed, and this effect is again typically stronger for young workers. Only in Spain is this specificity missing, thus indicating another difficulty faced by young workers in this country.

What really matters is not the age of a worker but, rather, his or her work experience and education. Our results confirm that young workers need to very quickly accrue relevant work experience because it diminishes their risk of becoming unemployed. The acquisition of higher education also appears to be an important factor in reducing the unemployment risk for young people.

## (p.219) 7.5.2. Duration Dependence of Unemployment

Finally, we consider the differences in the duration of unemployment between young people and prime-age workers. Figure A7.5 in the Appendix provides evidence from KM functions. The graphs confirm our empirical findings presented in Section 7.4—as well as the conceptual considerations mentioned in the literature overview in Section 7.2—that young unemployed are unlikely to suffer from longer job search compared to prime-age unemployed.11 But this is not to say that the problem of a prolonged duration of job search within the group of young unemployed should be ignored. As the Great Recession progressed, the survival functions in Figure A7.5 show dramatic declines in the job-finding prospects of young unemployed even in the shortest unemployment spells. This tendency is most apparent when comparing the periods 2008–2009 and 2012.

Austria shows its best performance within this general tendency. Figure A7.5 illustrates the gap between Austria and the remaining countries in terms of time needed by young unemployed to find a job: For instance, in 2010–2011, 44% of young Austrian unemployed managed to find a job after 4 months of unemployment duration; in France, Poland, and Spain, the shares were only 29%, 25%, and 21%, respectively. When comparing the situation after unemployment lasting for a minimum of 1 year in the same period, Austria again boasts the best job-finding prospects for young unemployed—this share amounted to 68%, as opposed to a mere 47% for Spain (58% for France and 53% for Poland).

Table 7.3 assesses the role of unemployment duration within a multivariate framework that also controls for a range of factors such as age, education, household size, and population density. The results in the first five rows of Table 7.3 show the impact of unemployment duration on individual job-finding probability in the form of hazard ratios. For each unemployment spell analyzed, the hazard ratio γ‎ in Table 7.3 indicates the probability of leaving unemployment and becoming employed relative to a reference spell. For the periods 2008–2009 and 2010–2011, unemployment spells lasting between 16 and 24 months represent a reference duration interval. For 2012, a reference interval stands for unemployment spells lasting between 11 and 12 months.

Statistically insignificant hazard ratios γ‎ would mean that there is actually no difference in job-finding prospects between the particular unemployment spell and the reference spell. The remaining rows in Table 7.3 show the hazard ratios that report the impact of explanatory variables. Suppose, for instance, that the hazard ratio reported for males takes the value “2”; then the probability that a man moves in a randomly chosen time from unemployment into employment would be, ceteris paribus, twice as high as for a woman.

In the initial stage of the Great Recession (2008–2009), the negative duration dependence of youth unemployment appeared to be absent in France and Austria.12 This means that the individual and other characteristics of the young (p.220) (p.221) (p.222) unemployed (and not the duration of unemployment per se) significantly affected their prospects of finding a job.

In contrast, those young unemployed in Poland whose unemployment spells lasted more than 10 months represented the risk group of young unemployed requiring targeting by additional policy measures, given that their chances of finding a job were not significantly better in comparison with the reference duration interval representing long-term unemployment (16–24 months). In Spain, the analogous risk group consisted of those young unemployed with unemployment spells exceeding 15 months (all shorter unemployment spells were associated with significantly higher job-finding probabilities).

Table 7.3 Youth hazard ratios of transition from unemployment to employment in four European countries (age category 1634 years)

2008–2009

2010–2011

2012

AT

ATa

ES

FR

FRa

PL

AT

ES

ESa

FR

PL

AT

ES

FR

PL

γ‎1 (1–2 months)

4.029***

0.477

3.444***

1.514**

0.582

2.589***

3.637***

2.318***

1.137

1.977***

2.280***

6.127**

4.715***

4.965***

7.061***

γ‎2 (3–4 months)

3.123***

0.629

3.517***

1.623**

0.743

3.357***

3.207***

2.578***

1.408

1.640***

3.128***

6.542***

5.186***

5.218***

9.670***

γ‎3 (5–6 months)

2.451**

0.671

3.031***

1.607**

0.862

3.006***

2.600**

2.632***

1.608

1.699***

3.236***

2.118

6.736***

4.845***

9.282***

γ‎4 (7–10 months)

1.458

0.524

2.067***

1.376*

0.894

2.924***

1.365

1.619***

1.110

1.457**

1.925***

2.934

3.208***

3.010***

5.447***

γ‎5 (11–15 months)

2.336**

1.225

2.212***

0.958

0.759

1.323

1.595

2.132***

1.719***

1.361

1.551**

Male

1.446***

1.659**

1.092

0.911

0.884

1.421***

1.051

1.122*

1.159

1.032

1.416***

1.204

0.987

0.800**

1.355***

Tertiary education

1.248

1.409

1.380***

2.158***

3.061***

1.155

1.450*

1.381***

1.542***

1.774***

1.161

1.886*

1.646***

1.735***

1.921***

Secondary education

1.319**

1.907***

1.100

1.522***

1.759***

1.156

1.410**

1.112

1.165

1.446***

0.936

1.662**

1.434***

1.549***

1.257

Age 20–24 years

1.624***

2.206***

1.283**

1.024

1.034

1.341*

2.044***

1.722***

1.968***

1.283

1.817***

0.822

2.312***

2.499***

1.846**

Age 25–29 years

1.790***

2.390***

1.327***

0.876

0.804

1.37*

2.644***

2.036***

2.417***

1.378*

2.037***

0.662

2.244***

2.235***

1.888**

Age 30–34 years

1.616**

2.072**

1.289**

0.771

0.707

1.639***

2.045***

1.753***

2.042***

0.948

1.848***

0.774

2.133***

1.907**

1.625

HH size 1

1.618**

3.452***

b

1.732***

2.315***

b

1.279

b

b

1.690***

b

1.693**

b

1.357

b

HH size 2

1.905***

2.843***

1.440***

1.438***

1.521***

1.581***

1.012

1.332***

1.475***

1.524***

1.684***

1.034

1.327***

1.395**

1.354*

HH size 3

1.143

1.314

1.118

1.141

1.202

0.883

0.840

1.087

1.121

1.208*

1.233**

0.765

0.999

0.979

1.221*

Densely populated

0.738**

0.594**

0.766***

0.643***

0.534***

0.932

0.676***

0.636***

0.574***

1.429***

0.987

0.450***

0.693***

0.817*

1.116

Medium populated

1.027

0.939

0.847**

0.684***

0.588***

0.878

0.740*

0.775***

0.731***

0.944

0.946

0.553***

0.872

0.771*

0.797*

Constant

0.012***

0.058***

0.018***

0.046***

0.118***

0.013***

0.020***

0.013***

0.023***

0.020***

0.011***

0.027***

0.004***

0.005***

0.002***

Log-likelihood

–977.8

–970.4

–4,201.8

–1,731.4

–1,728.4

-1,863.4

–872.2

–3,476.0

–3,474.1

–2,007.6

–2,614.0

–463.3

–2,674.1

–1,329.3

–1,601.2

p value

0.000

0.007

0.024

n (unemployment spells)

541

541

2,237

896

896

1,077

466

1,952

1,952

1,014

1,376

315

2,133

981

1,378

(a) Results with gamma frailty reported only when the likelihood ratio test of gamma variance (p value) significantly proved the impact of unobserved heterogeneity on model results.

(b) One- and two-person households are merged because of a low share of observations in the first category.

HH, household.

(*) p < .10.

(**) p < .05.

(***) p < .01.

Sources: EU-SILC UDB 2010, version 5 of August 2014; EU-SILC UDB 2012, version 1 of August 2014; EU-SILC UDB 2013, version 2 of August 2015; authors’ computations.

As the Great Recession progressed, the duration dependence of youth unemployment started affecting developments in all the countries analyzed. Young Austrians who remained unemployed for more than 6 months in 2010–2011 (and for more than 4 months in 2012) represented the risk group of young unemployed. It follows that those young Austrian unemployed who did not find (or did not accept) a job relatively quickly faced sharply diminishing employment prospects. This suggests that stigmatization and/or discouragement effects of prolonged youth unemployment emerged in Austria with much briefer unemployment spells than in countries with considerably higher youth unemployment. The probable reason is that in countries with high levels of youth unemployment, longer unemployment durations are considered more “natural.” The results for France and Poland confirm this assumption. In France, all young individuals who were unemployed for more than 10 months formed the risk group. This applied to both 2010–2011 and 2012. The results for Poland are very similar to those for France.

In Spain, the situation changed most dramatically between the two periods. In 2010–2011, all young unemployed with unemployment spells of between 1 and 10 months actually constituted the risk group in that their job-finding probabilities did not differ significantly from the employment prospects of long-term young unemployed (16–24 months). This further illustrates the depth of the youth unemployment problem in Spain. In contrast, the results for Spain in 2012 became comparable with those for France and Poland—the risk group of Spanish young unemployed was associated with unemployment durations exceeding 10 months.

The analysis of explanatory variables does not confirm the uniform presence of statistically significant gender- or education-based differences. But in Austria and France, the chance of finding a job gradually evolved in favor of young women. This is in line with Kelly et al. (2013), who report a lower probability of moving from unemployment to employment for young Irish men. In contrast, young Polish men have a consistently relatively higher chance of finding a job compared to young women in Poland. Spain shows no gender effects. Poland is also specific in that it lacks an education effect (except for tertiary education in 2012), whereas for the remaining countries we find convincing evidence that the (p.223) chance of a young unemployed person finding a job increases with secondary and/or tertiary education. A young person aged 25–29 years has the highest probability of moving from unemployment to employment in the majority of countries and periods analyzed. This indicates that employers tend to avoid hiring the relatively immature young unemployed. Regarding household size, it negatively affects the probability of young unemployed finding a job, although with varying significance. A significantly higher job-finding probability (relative to a household consisting of four members or more) is associated almost exclusively with small households consisting, as a maximum, of two members. This might suggest that in the absence of other members in respondents’ households (presumably their parents), who could contribute decisively to the common budget, the pressure to find a job imposed on young unemployed is significantly higher. However, this effect is not fully uniform—it is absent for Austria in 2010–2011.

The hazard ratios for densely populated areas are significant and lower than 1 (except for Poland in all three periods). This seems to contradict the assumption that larger cities provide more employment opportunities and thus better chances to exit from unemployment.13 Our result can be associated with longer job search in the hope of gaining a better match or more opportunities to participate in the informal economy.

Table A7.1 in the Appendix suggests that prime-age hazard functions generally display a higher sensitivity of job-finding chances to the duration of unemployment episodes. Among other findings, the impact of age on the prospect of finding a job among prime-age unemployed is worth noting, especially the significant negative impact of age categories 45+ years. This suggests that the presumed skill obsolescence and deterioration in human capital associated with these age categories function as negative signals to potential employers and diminish the chances of older unemployed finding work.

# 7.6. Conclusions

Youth are relatively more involved in gross flows than are prime-age groups. This holds true uniformly across the four countries analyzed during the period 2008–2012 and supports the existing evidence of a higher aggregate fluidity of youth labor markets compared to prime-age markets. The main result stemming from the analysis of flow transition rates is that a young worker is more likely to move between employment and unemployment in both directions compared to a prime-age worker. This finding is in line with contemporary evidence for the United Kingdom. It can be interpreted as a typical feature of marginalized groups, which have to “churn” relatively more frequently through the (secondary) labor market.

The analysis of transition rates provides the following main conclusion: The policy priority should be to reduce the gap between the unemployment risks (p.224) faced by a young and a prime-age worker. This gap is characteristic for all the labor markets analyzed and concerns countries with substantially different labor market performance, institutions, EU membership history, and other national specificities. Reducing the gap is important not only for generally improving the relative position of marginalized youth on the labor market but also for achieving more proportional evolutions in the youth and prime-age unemployment rates.

This chapter demonstrates that inflows of young workers into unemployment accounted for far higher increases in youth unemployment rates compared to prime-age unemployment rates. In contrast, had the outflows from unemployment to employment been the only driver of unemployment rate dynamics, youth unemployment rates would have developed more favorably than prime-age unemployment rates.

We analyzed in detail the determining factors of transitions from employment to unemployment. The results again stress the importance of a policy targeted at employment protection for young people, who need to gain work experience promptly after entering the labor market so as to minimize the probability of job loss. In addition, the effect of education on lowering the risk of job loss and becoming unemployed is apparent for individuals of any age; nonetheless, it seems that higher education decreases the probability of becoming unemployed more substantially for young workers.

Finally, we examined the extent to which the job-finding chances of young unemployed decline due to the duration of their unemployment, net of the impact of standard socioeconomic characteristics and unobserved heterogeneity of unemployed persons. Although the results for young unemployed appear to be generally favorable compared to those for prime-age unemployed, they simultaneously show growing negative duration dependence of youth unemployment as the Great Recession progressed. From 2010 onward, the job-finding prospects of young unemployed could be viewed as a diminishing function of unemployment duration in all countries analyzed. In 2012, the results nearly equalized across countries (except for Austria): With unemployment durations exceeding 10 months, the job-finding probability of a young unemployed person declines significantly, and those who remain unemployed for a longer time deserve additional policy attention.

Such a result may represent useful feedback for the European Youth Guarantee scheme, which promotes uniformly an offer to young people in the EU of a quality job, an apprenticeship, or training within 4 months after graduation or job loss. In contrast, our results demonstrate that the job-finding probability of a young unemployed person is already highest within the shortest unemployment spells. Although the information on unemployment durations and job-finding probabilities is never available ex ante to policymakers, it would appear that young people who are unemployed for a considerably longer time than 4 months are those who should be targeted by concentrated policy efforts and (p.225) resources. This proposition is probably even more relevant for the ongoing post-recessionary period.

(p.227)

References

Bibliography references:

Albert, Cecilia, Luis Toharia, and María A. Davia. 2008. “To Find or Not to Find a First ‘Significant’ Job.” Revista de Economía Aplicada 16 (46): 37–59.

Bell, David N. F., and David G. Blanchflower. 2011. “Young People and the Great Recession.” Oxford Review of Economic Policy 27 (2): 241–67.

Blanchard, Olivier J. 1999. European Unemployment: The Role of Shocks and Institutions. Rome: Banca d´Italia.

Blanchard, Olivier J., and Peter A. Diamond. 1990. “The Cyclical Behavior of Gross Flows of US Workers.” Brookings Papers on Economic Activity 21 (2): 85–156.

Cox, David R. 1972. “Regression Models and Life Tables.” Journal of the Royal Statistical Society, Series B (Methodological) 34 (2): 187–220.

D’Addio, Anna C. 1998. “Unemployment Durations of French Young People: The Impact of Individual, Family and Other Factors on the Hazard Rate.” CORE Discussion Paper 9851. Louvain-in-Neuve: Université Catholique de Louvain.

Davia, María A., and Nuria Legazpe. 2014. “Determinants of Employment Decisions After the First Childbirth in Spain.” Journal of Family and Economic Issues 35 (2): 214–27.

Disney, Richard, Carl Emmerson, and Matthew Wakefield. 2006. “Ill Health and Retirement in Britain: A Panel Data-based Analysis.” Journal of Health Economics 25 (4): 621–49.

Dixon, Robert, John Freebairn, and Guay C. Lim. 2011.”Net Flows in the U.S. Labor Market, 1990–2010.” Monthly Labor Review (February): 25–32.

Elsby, Michael W. L., Jennifer C. Smith, and Jonathan Wadsworth. 2011. “The Role of Worker Flows in the Dynamics and Distribution of UK Unemployment.” Discussion Paper 5784. Bonn: Institute for the Study of Labor.

Flek, Vladislav, Martin Hála, and Martina Mysíková. 2015. “Duration Dependence and Exits from Youth Unemployment in Spain and the Czech Republic.” Economic Research 28 (1): 1063–78.

Flek, Vladislav, and Martina Mysíková. 2015. “Unemployment Dynamics in Central Europe: A Labour Flow Approach.” Prague Economic Papers 24 (1): 73–87.

Flek, Vladislav, and Martina Mysíková. 2016. “Youth Transitions and Labour Market Flows.” STYLE Working Paper 5.2. Brighton, UK: CROME, University of Brighton. http://www.style-research.eu/publications/working-papers

Gomes, Pedro. 2009. “Labour Market Flows: Facts from the United Kingdom.” Working Paper 367. London: Bank of England.

(p.228) International Labour Organization. 2013. Global Employment Trends for Youth 2013: A Generation at Risk. Geneva: International Labour Organization.

Jenkins, Stephen P. 1997. “Sbe17: Discrete Time Proportional Hazards Regression.” STATA Technical Bulletin 7 (39): 22–32.

Kaplan, Edward L., and Paul Meier. 1958. “Nonparametric Estimation from Incomplete Observations.” Journal of the American Statistical Association 53 (282): 457–81.

Kelly, Elish, Seamus McGuinness, Philip J. O’Connell, David Haugh, and Alberto Gonzáles Pandiella. 2013. “Transitions in and out of Unemployment Among Young People in the Irish Recession.” Economics Department Working Paper 1084. Paris: Organization for Economic Co-operation and Development.

Machin, Stephen, and Alan Manning. 1999. “The Causes and Consequences of Long-Term Unemployment in Europe.” In Handbook of Labor Economics, edited by Orley C. Ashenfelter and David Card, 3085–139. Amsterdam: Elsevier.

Petrongolo, Barbara, and Christopher A. Pissarides. 2008. “The Ins and Outs of European Unemployment.” American Economic Review 98 (2): 256–62.

Reich, Michael, David M. Gordon, and Richard C. Edwards. 1973. “Dual Labor Markets: A Theory of Labor Market Segmentation.” American Economic Review 63 (2): 359–65.

Shimer, Robert. 2012. “Reassessing the Ins and Outs of Unemployment.” Review of Economic Dynamics 15 (2): 127–48.

Figure A7.1 Transition rates from inactivity to employment for various age groups in four European countries (monthly averages, in %)

Sources: EU-SILC UDB 2010, version 5 of August 2014; EU-SILC UDB 2012, version 1 of August 2014; EU-SILC UDB 2013, version 2 of August 2015; authors’ computations.

Figure A7.2 Transition rates from employment to inactivity for various age groups in four European countries (monthly averages, in %)

Sources: EU-SILC UDB 2010, version 5 of August 2014; EU-SILC UDB 2012, version 1 of August 2014; EU-SILC UDB 2013, version 2 of August 2015; authors’ computations.

Figure A7.3 Transition rates from inactivity to unemployment for various age groups in four European countries (monthly averages, in %)

Sources: EU-SILC UDB 2010, version 5 of August 2014; EU-SILC UDB 2012, version 1 of August 2014; EU-SILC UDB 2013, version 2 of August 2015; authors’ computations.

Figure A7.4 Transition rates from unemployment to inactivity for various age groups in four European countries (monthly averages, in %)

Sources: EU-SILC UDB 2010, version 5 of August 2014; EU-SILC UDB 2012, version 1 of August 2014; EU-SILC UDB 2013, version 2 of August 2015; authors’ computations.

Figure A7.5 Survival functions for two age groups (16–34, 35–54) in four European countries (probabilities of remaining unemployed in %; 1: 2008–2009, 2: 2010–2011; 3: 2012)

Note: Analysis time: unemployment in months.

Sources: EU-SILC UDB 2010, version 5 of August 2014; EU-SILC UDB 2012, version 1 of August 2014; EU-SILC UDB 2013, version 2 of August 2015; authors’ computations.

(p.230) (p.231) (p.232) (p.233) (p.234) (p.235)

Table A7.1 Prime-age hazard ratios of transition from unemployment to employment in four European countries (age category 3554 years)

2008–2009

2010–2011

2012

AT

ES

FR

PL

AT

ES

FR

PL

AT

ATa

ES

FR

PL

γ‎1 (1–2 months)

9.106***

5.432***

2.062***

2.346***

4.956***

3.986***

1.691***

3.181***

6.297***

2.479

5.467***

4.480***

4.358***

γ‎2 (3–4 months)

10.808***

5.193***

2.030***

4.833***

3.845***

3.710***

1.903***

5.116***

6.291***

3.870**

6.379***

3.811***

7.374***

γ‎3 (5–6 months)

4.296***

4.583***

2.150***

3.972***

3.201***

3.562***

1.691***

3.668***

2.710

2.094

6.271***

3.597***

5.987***

γ‎4 (7–10 months)

2.575**

3.203***

2.042***

3.195***

1.732

2.096***

1.358*

2.648***

1.822

1.632

3.574***

2.752**

3.303***

γ‎5 (11–15 months)

2.139

2.262***

0.874

1.511

2.340**

3.197***

1.155

2.186***

Male

1.612***

1.026

0.982

1.421***

1.450***

1.271***

1.109

1.770***

1.179

1.372

1.146**

1.267*

1.951***

Tertiary education

1.530**

1.124

1.831***

1.316

1.336

1.273***

1.188

1.404

1.230

1.267

1.161

1.484*

1.516

Secondary education

1.602***

1.100

1.242

1.112

1.263*

1.011

1.257*

1.056

1.098

1.073

1.129

1.619***

1.271

Age 40–44 years

0.817

1.145*

0.766*

1.176

0.957

1.019

1.192

0.891

0.866

0.696

0.979

0.644**

1.018

Age 45–49 years

0.731**

1.038

0.824

0.737**

0.720**

0.997

0.967

0.918

0.739

0.582*

0.849*

0.777

0.818

Age 50–54 years

0.824

0.928

0.705**

0.687***

0.510***

0.771***

0.715**

0.748**

0.410***

0.260***

0.728***

0.490***

0.785

HH size 1

1.176

b

1.411**

b

0.646**

b

1.032

b

1.106

1.034

b

0.870

b

HH size 2

1.177

0.954

0.961

1.046

1.104

1.051

0.712**

0.823*

0.988

0.914

1.154

0.888

0.972

HH size 3

1.009

0.955

1.017

0.918

0.827

1.178**

0.825

0.784**

1.177

1.150

1.222**

0.863

0.926

Densely populated

0.641***

0.675***

0.796

0.951

0.428***

0.559***

1.326**

0.896

0.266***

0.166***

0.554***

0.897

0.899

Medium populated

0.685***

0.684***

1.104

1.044

0.721**

0.711***

1.146

0.802**

0.668**

0.559**

0.707***

0.827

0.845

Constant

0.014***

0.019***

0.027***

0.016***

0.046***

0.017***

0.030***

0.017***

0.037***

0.154*

0.009***

0.014***

0.005***

Log-likelihood

–996.8

–4,030.9

–1,326.9

–1,800.1

–997.3

–4,067.8

–1,677.3

–2,307.9

–532.0

–530.0

–3,629.6

–958.0

–1,305.7

p value

0.022

n (unemployment spells)

548

2,260

722

1,028

519

2,353

822

1,237

394

394

3,048

734

1,176

(a) Results with gamma frailty reported only when the likelihood ratio test of gamma variance (p value) significantly proved the impact of unobserved heterogeneity on model results.

(b) One- and two-person households are merged because of a low share of observations in the first category.

(*) p < .10.

(**) p < .05.

(***) p < .01.

HH, household.

Sources: EU-SILC UDB 2010, version 5 of August 2014; EU-SILC UDB 2012, version 1 of August 2014; EU-SILC UDB 2013, version 2 of August 2015; authors’ computations.

(p.236)

## Notes:

(1) Hadjivassiliou et al. (this volume) provide an overview of national specificities in youth labor market performance and in institutional arrangements of labor markets across the EU (including employment protection legislation, vocational education and training, active labor market policy, and collective bargaining). Our categorization of countries is analogous to that of Berloffa et al. (this volume), who analyze, among others, the clusters of Continental, Mediterranean, and Eastern European countries. Given the depth of our analysis, we concentrate on only a limited number of countries that reflect our categorization.

(2) Section 7.3 discusses the data issues in more detail.

(3) A common practice in this respect is to follow Petrongolo and Pissarides (2008) or Shimer (2012) in showing how much of the variance of the steady-state unemployment rate accounts for changes in the flow transition rates. Also see Elsby et al. (2011) for an application to youth unemployment rate dynamics in the United Kingdom. A credible compliance with this direction would require data gathered over a longer period of time than in the EU-SILC. This is why we limit ourselves to a “flow” decomposition of the observed changes in unemployment rates. Dixon et al. (2011) apply a similar framework to US data. Except for Flek and Mysíková (2015), such an approach has probably never been applied before to a cross-country analysis in Europe.

(4) EU-SILC is an annual survey in which the monthly economic status is reported retrospectively. Respondents might not always recall correctly when they changed their labor market status. Although the precise month of such changes is potentially uncertain, it should not affect our results, which are based on monthly averages for the entire subperiods analyzed.

(5) EU-SILC data do not account for direct job-to-job flows. This is why in our analysis an unchanged employment status can represent either maintaining the previous job or moving to another job.

(6) A particular unemployment spell is left censored when it is already in progress at the beginning of the observed period, and it is right censored when it does not terminate by the end of the observed period. An additional, specific type of right censoring occurs when an unemployment spell ends in inactivity rather than in employment. The KM estimators applied take into account the right-censored data, whereas the left censoring remains unaddressed by techniques available to us. The seemingly easiest solution to this problem would be to remove the censored observations from the data set. (p.226) Unfortunately, this would probably make all the estimations of unemployment durations downward biased because longer unemployment spells are more likely to be censored compared to shorter ones. Note that in the case of probit model estimations, censoring is not an issue because we do not analyze the duration of employment there.

(7) This model was implemented into a STATA routine (pmghaz) by Jenkins (1997). We utilize a refined version (pmghaz8) that has been applied, for example, by Disney et al. (2006), Albert et al. (2008), Davia and Legazpe (2014), and Flek et al. (2015).

(8) Note that these variables may not capture all the existing differences among unemployed individuals, and their unobserved heterogeneity may lead to spurious duration dependence (Jenkins 1997; Machin and Manning 1999). To account for unobserved heterogeneity, we use the mixed proportional hazard model, in which the continuous hazard rate contains a gamma-distributed random variable with unit mean and unknown variance, which is to be estimated.

(9) The results in Figure 7.1 do not involve the 55+ years age group. However, the share of elderly individuals in the working-age population and/or the specificity of their transitions are not large enough to qualitatively change the overall nature of the results (Flek and Mysíková (2015) report more details on flows among the elderly).

(10) We do not report results for unemployment-to-education transitions of young people. However, Figure A7.4 in the Appendix presents the outflow rates from unemployment to inactivity for four age bands of young unemployed. In most countries and periods, these rates do not necessarily increase with lower age. Moreover, for the low age categories of young unemployed, the outflow rates from unemployment to inactivity are too low (usually lower than 1%) to represent any real alternative to unemployment. Austria can be viewed as the only exception.

(11) Log-rank tests reveal that only in Austria (in the first subperiod analyzed) is the youth survival curve placed significantly above the prime-age survival curve; see Figure A7.5 in the Appendix.

(12) For Poland and Spain, the controls for unobserved heterogeneity did not affect the significance of the results reported in Table 7.3 for the given period. For Austria and France, the results with gamma frailty are reported in additional columns because the likelihood ratio test of gamma variance (p value) proved the impact of unobserved heterogeneity on the significance of results—with added controls for unobserved heterogeneity, all the coefficients turned out to be insignificant. To eliminate spurious duration dependence, we decided not to discuss the results where the controls for unobserved heterogeneity proved the insignificance of duration intervals for job-finding probability.

(13) D’Addio (1998) reports such an effect for young French women in the early 1990s.