Workers Flows in the European Union During the Great Recession

Transcription

1 Workers Flows in the European Union During the Great Recession Jose Maria Casado a, Juan Francisco Jimeno a & Cristina Fernandez a VERY PRELIMINARY VERSION This draft: October 24 Abstract Using data from the European Labour Force Survey, we measure the contribution of workers flows to the change in the unemployment rate in some EU countries during 26-22, highlighting the main socio-demographics characteristics of the individuals transiting between employment and unemployment. Besides, we analyse the role of labour market institutions on the change of the aggregate flow for each socio-demographic group across country during the crisis. For the socio-demographic characteristics, we observe that age, to a larger extent than educational attainments, are the main determinants of flows from employment into unemployment, particularly in those countries where unemployment increased by most. This suggests that labour market institutions together with the employment policies followed during the crisis (or the lack of them) have very much conditioned the employment adjustments during the Great Recession in EU labour markets. Finally, we examine if the crisis has encouraged a reallocation of the labour force to more productive sector. We still do not find evidence in favour of this issue in the European countries with the particular exception of Spain and Greece. Keywords: Labour Flows, Unemployment, Labour Market Institutions, Great Recession JEL: J6, E24, C25 The views expressed here are those of the authors and do not necessarily reflect those of the Banco de España. All remaining errors are our own. (a) Banco de España

2 Introduction One of the features of the global crisis that has received much attention is the different impact on unemployment it has had among European economies. In some countries (e.g. Belgium, Austria, Germany) unemployment barely increased, and nowadays, after almost seven years since the crisis started, it is more or less at the same level that it was in 27. In other countries (most notably, Greece, Portugal and Spain) the unemployment rate rocketed and reached similar levels or even higher than those of the high unemployment period of the early 98s. Understanding the reasons why the crisis has had such differential labour market effects is crucial for designing employment policies that support employment creation and speed up the recovery. Moreover, the degree of heterogeneity in the performance of European labour markets during the crisis needs to be taken into account when evaluating how macroeconomic policies can contribute to reduce unemployment. In principle, there could be three possible reasons for cross-country differences in the (un)employment consequences of the crisis. First, although the origins of the crisis are, to a large extent, similar in all the countries, it could be that the implications in terms of the fall of economic activity were much larger in some countries than in others. Secondly, the responses of employment policies to the crisis could have varied across countries, with some of them trying to avoid employment destruction through specific schemes favouring reduction in working hours and labour hoarding. Thirdly, faced with a similar negative shock, differences in labour market institutions across countries may also explain why in some countries the loss of employment was much larger than in others. Cross-country comparisons of the aggregate unemployment rate cannot fully provide an identification of the three factors above. In particular, the role played by institutional differences and by employment policies is better understood by looking at worker flows and at the socio-demographic composition of those flows. For instance, the impact of employment policies targeted at reducing working hours (as those implemented in Germany 2 ) would be better observed in cross-country differences regarding the flows for employment to unemployment. Similarly, the changes in unemployment due to discouraged worker effects or to added worker effects are more apparent in the flows between unemployment and inactivity. Moreover, the socio-demographic composition of these flows is also See IMF, Economic Outlook, April, 2; OECD, Employment Outlook, 2; and European Commission, Employment in Europe, 2 for detailed accounts of cross-country differentials in the initial impact of the crisis on unemployment. 2 See Burda and Hunt (2).

3 informative about both the factors behind the change in unemployment and the impact of some institutional features of the labour market during the Great Recession. Hence, in this paper we firstly look at workers flows as a step forward to understanding the cross-country differences in the rise of unemployment during the Great Recession. Using annual data from the EU-Labour Force Survey, compiled by Eurostat, we compute workers flows between employment and unemployment, unemployment and inactivity, and employment and inactivity, and their socio-demographic composition, distinguishing 8 population groups (defined by gender, age groups and educational attainment levels) in EU countries (Austria, Belgium, Germany, Denmark, Spain, France, Greece, Italy, Portugal, Sweden and UK) during the period. Secondly, we examine the role of labour market institution in accounting the different impact of the crisis on the socio-demographic flows composition across countries. Finally, we explore if the crisis has brought any reallocation of the labour force to more productive sectors in order to anticipate a possible cleansing effect of the recession. The rest of this paper is structured as follows. The next section presents an overview of the data used in this paper. Section 3 provides a decomposition of the aggregate unemployment rate into these flows. Section 4 discusses the empirical approach employed to examine the socio-demographic decomposition of the flows and the effects of the institutions. Section 5 presents the results from the first part of our empirical investigation: the socio-demographic decomposition of the flows. Additionally, we compare the demographic impact over the labour market flows of the Great Recession with respect to the crisis that took place at the beginning of the nineties. Section 6 provides de results from the second part of the empirical analysis, the impact of labour market institutions on the firsts stage results. Section 7 gives some evidence of the sectoral reallocation of the Great Recession from the point of view of worker flows despite the limitation of the EU-LFS for this analysis. Section 8 concludes with some final remarks. 2 Data and descriptives: The European Labour Force Survey The EU Labour Force Survey (EU-LFS) is a large sample survey among private households which provides detailed quarterly data on the employment status of all the individuals of the household aged 5 and over. It also collects information on many dimensions regarding the socio-demographic characteristics of the individuals, as well as on the job they are performing, in case they are employed, or the methods used to find a job, in case they are unemployed. 2

4 The EU-LFS interviews each quarter around.5 million people in the EU, European Free Trade Association (EFTA) except Liechtenstein- and Candidate Countries. Initially, from 983, its results covered one quarter per year only (usually in spring), but form 998 to 25 it underwent a transition to a continuous survey, with interviews distributed across all weeks of the year, in order to give reliable quarterly results. One advantage of the survey for our purpose is that the definition of labour status is similar across countries. Thus, persons in employment are those who during the reference week of the interview did any work for pay or profit. This definition includes employees, entrepreneurs, family workers, unpaid family workers and self-employed persons. Unemployed persons are those that comprise the three following characteristics. First, do not have a work during the reference week, i.e., neither had a job nor were at work (for one hour or more) in paid employment or self-employment. Second, are currently available for work, i.e. were available for paid employment or self-employment before the end of the two weeks following the reference week. And third, are actively seeking work, i.e. had taken specific steps in the four week period ending with the reference week to seek paid employment or self-employment. Finally, inactive persons are those who are neither classified as employed nor as unemployed. As for socio-demographic characteristics, the age of the respondent refers to the difference between the date of the reference week and the date of birth. However, it is delivered in 5-year age bands that we end up grouping into three categories: youngest (6-29), middle-aged (3-49) and oldest (5-64). Educational attainment levels refer to the highest level of education or training successfully completed by the individual, considering both general and vocational education/training, (ISCED 997). We group them under three different categories: low, medium and high-educated individuals. Low educated refers to those respondents whose maximum level of education reaches up to compulsory education (ISCED -2); medium educated ranges general and vocational studies from compulsory education to pre-college (ISCED 3-4); and high educated refers to those respondents that pursue at least a college degree of a minimum duration of two years or a similar vocational degree (ISCED 5 and 6). This classification is sufficiently broad to end up considering and homogeneous grouping across countries, but we still may have some heterogeneity arising, for example, from the fact that compulsory years of education differ across countries and over time. Our sample contains information about workers flows in EU countries (Austria, Belgium, Germany, Denmark, Spain, France, Greece, Italy, Portugal, Sweden, and the UK). 3

5 The sample period is We restrict our analysis to males and females between 5 and 64 years of age that report their level of education as well as their employment status, both in the moment of the interview and one year before. Workers flows are constructed from the declared changes in employment status between these two years. The main summary statistics that describe the sample are reported in Table. Not surprisingly, countries are roughly similar in the socio-demographic variables, with somehow older populations in Nordic countries, and less educated ones in Southern European countries. Differences in employment market status are noticeable, with the typical European pattern of lower unemployment and inactivity rates in Nordic Countries and the UK than in Continental European countries, and, within these, higher unemployment and inactivity rates in Southern countries. Unemployment rates and annual transition probabilities between employment and unemployment along the Great Recession are reported in Figure 4. In our data, the size and cross-country differences in workers flows before the crisis are consistent with the broad patterns highlighted by the empirical literature on workers flows in Europe, as documented, among others, by Burda and Wyplosz, 994; Elsby, Hobijn and Sahin, 23; and OECD, 2. The impact of the recession on the flows between employment and unemployment are notoriously different across countries. Thus, in Germany, Austria and Belgium, they have barely changed, while the increase of the flows from employment to unemployment and the decrease of the flows between unemployment and employment have been moderate in the rest of the countries, but to Greece, Portugal, and Spain where, approximately, the former doubled and the latter halved. However, since our measures of workers flows are computed from transitions among employment status from one year to the next, they should not be taken as reflecting the incidence of job creation and job destruction. The reason is the (well-known) time aggregation bias resulting from the lack of observation of changes in workers employment status within a given year. Since atypical employment (part-time, fixed-term, temporary) have been on the rise in many EU countries and its average duration is short, job creation and job destruction rates are likely to be higher than the size of the workers flows in our sample 5. Nevertheless, the increase in the incidence of long-term unemployment during the Great Recession (Figure 2) suggests that these annual transitions may provide a good approximation to the rates at which workers are losing their jobs and the unemployed are finding new jobs. Still, to confirm if the time aggregation bias may render 3 Except for Sweden where we will not be able to exploit data for 26 because information on the status with regard to activity in the previous year is not available. 4 See the Appendix A. for information on transitions between (un)employment and inactivity. 5 Elsby et al. (22) use monthly data and quarterly observations on the duration of unemployment spells to take account of the time aggregation bias in workers flows as registered by the EU-LFS. 4

6 annual transition rates meaningless, we use information from the Spanish Labour Force Survey to construct the same transitions from quarterly data. The comparison between both measures of transitions is presented in Figure 3. Although it is true that annualized quarterly transitions are much higher than those obtained from annual data, the time evolution of both measures are very similar. 3 Workers flows and aggregate unemployment Taking employment status as one of three possible states (employment, unemployment, and inactivity), the unemployment rate can be computed as a function of the six corresponding transition rates, as follows (see Pissarides, 2): u = h eu + h eu + h iu h ie +h iu h ei h iu h ie +h iu h ei + h ue + h ie h iu +h ie h ui where h eu stands for the probability of flowing from employment to unemployment, h iu stands for the probability of flowing from inactivity to unemployment, h ie refers to the transition rate from inactivity to employment, h ei to the transition from employment to inactivity, and h ue and h ui refer to the probability of flowing out of unemployment and into employment and inactivity, respectively. One way of computing the contribution of each flow to the variation of the unemployment rate is by substituting in this formula the observed transition rate for a particular year and holding constant the rest of the rates observed in the previous period. This gives the change in unemployment that would have been observed if all the others transitions had remain invariant. Obviously, this computation does not yield an additive decomposition of the changes in the unemployment, as there is a combined effect of changes in all the flows (that we label as the residual ) that is sizeable, particularly in those countries where unemployment increased by most. Using this approach, we observe that in most of the countries it is the increase in the hazard rate from employment to unemployment, this is, the increase in the probability of losing the job and becoming unemployed, the main determinant of the increase in the unemployment rate, followed by the decrease in the finding probability among the unemployed 6. As seen in Figure 4, the increase in the flows from employment to unemployment 6 Shimer (22), using US monthly data for the period , argues that 75% of the fluctuations in the unemployment rate are driven by movements in the job-finding probability, this is, the outs of unemployment. He argues that ignoring time aggregation will bias a researcher towards finding a 5

7 and the decrease of those from unemployment to employment explain the main bulk of the increase in unemployment (around 45% and 5% respectively), both in the countries where this increase was moderate and in those where unemployment surged. This is also the case in Austria, where the increase of unemployment was negligible, but neither in Germany, where the flows from employment to unemployment decreased, nor in Belgium, where the flows from unemployment to employment increased. In Spain and Portugal, changes in these two flows contributed by pp. to the increase in unemployment, where in Greece the contribution is above 5 pp. Variations in the rest of the flows contributed very little to unemployment changes, but it is worthwhile to notice the increase in flows from unemployment to inactivity and from inactivity to employment in Austria and Germany, and the decrease of the flows from inactivity to unemployment in Germany, whose contribution to keep unemployment low was relatively important. In contrast, in the rest of the countries, the contribution of (increased) flows from inactivity to unemployment and (decreased) flows from inactivity to employment was towards pushing the unemployment rate up, while the (decreased) flows from unemployment to inactivity contributed to increase the unemployment rate (except in Portugal) 7. Finally, increasing flows from employment to inactivity only had a relatively important contribution to increase the unemployment rate in Belgium and France. A better understanding of the changing patterns and cross-country differences of workers flows among different employment states during the Great Recession requires a more detailed investigation into some individual characteristics of the individuals transiting among states. In the following Section we highlight their composition by gender, age, and educational attainment. As flows between employment and unemployment are the most important driving forces of unemployment changes, we disregard the rest of the flows 8. countercyclical employment exit probability, because when the job finding probability falls, a worker who loses her job is more likely to experience a measured spell of unemployment. 7 Increasing flows from unemployment to inactivity may be due to the so-called discouraged worker effect, by which workers quit job searching due to the lack of vacancies available. Higher flows from inactivity to unemployment may the consequence of the so-called added worker effect, that is, inactive members of household joining the labour force to compensate for employment losses of other members. In this case, a longer transition from school to jobs, as young individuals take longer to find a first job, may also explain an increase in the annual frequency of transitions from inactivity to employment. 8 Moreover, as job searching is not always a precisely observed concept in labour force surveys, there is a grey area in the definition of the boundary between inactivity and unemployment that makes the identification of flows between unemployment and inactivity a bit problematic (See Jones and Riddell, 999, 22, and Garrido and Toharia, 24). Active job search in the EU-LFS is defined as the fulfilment of one the following steps to find a job: having been in contact with a public employment office to find work, whoever took the initiative (renewing registration for administrative reasons only is not an active step), having been in contact with a private agency (temporary work agency, firm specialising in recruitment, etc.) to find work, applying to employers directly, asking among friends, relatives, unions, etc., to find 6

8 Finally, we compare the flows composition by age and gender in the current and previous crisis in order to identify the characteristic effects of the great recession on labor market. 4 Modelling Strategy To get some insights in the factors driving the changes in transitions between employment and unemployment in EU countries during the Great Recession, it is useful to look at the socio-demographic composition of workers flows to identifying in which dimensions, if any, this composition is different with respect to the situation before the crisis (27). These transitions affect mostly to workers with regular attachment to the labour market. Labour market institutions such as firing costs, unemployment benefits, wage setting, and working hours sharing schemes affect both to the size and the composition of those flows. In order to compute the changes in the socio-demographic composition of flows during the crisis and the impact of some institutions on these changes, we follow a two-step approach. Firstly, we measure the socio-demographic composition of flows between employment and unemployment (also between unemployment and employment) running separate regressions on the micro-data (EU-LFS) to obtain estimates β t, β 2t... β 8t for each country c in an equation of the form: 22 H it = X it t=26 22 β t D t + + X 8it t=26 β 8t D t Covariates (X it...x 8it ) are 8 dummy variables that identify whether the individual belongs to any of the demographic groups that arise as a result of interacting gender dummies (male; female), educational attainment dummies (low, medium and high), and age-group dummies (6-29; 3-49; 5-65). Moreover, every covariate is interacted with year dummies D t from 26 up to 22. Dependent variable H it denotes one of two different outcomes in two different sets of regressions. In the first specification, the outcome is a dummy variable that takes the value when the individual labour state changes from employment to unemployment. In the second, dummy variable takes the value when the individual labour state changes from unemployment to employment. Both models are Logit and we recover the transition probabilities from employment to unemployment P EUc unemployment to employment P UEc from the estimated β c, β 2c... β 8c coefficients. and from for each socio-demographical group across countries work, placing or answering job advertisements, studying job advertisements, taking a recruitment test or examination or being interviewed, looking for land, premises or equipment, applying for permits, licences or financial resources. 7

9 (TO BE DECIDED THE APPROACH YET) Our second step is to run a sequence of regressions on country-level data ( observations), one for each β in the first step to compute the effect of labour market institutions (in particular the employment protection legislation, EPL) on the change in the labour status transition probability. For example, we obtain estimates ( γ, γ ) from a regression of the difference in the transition probabilities for employment to unemployment between two time periods (P EUc,t P EUc,t ) on z c, a measure of country-specific labour market institutions. P EUc,t P EUc,t =γ +γ z c +v c, Where v c is an error term that captures unobserved country-level variables, as well as possible specification errors. The weakest interpretation of our estimates γ, γ 2,..., γ 8 is that these reflect unbiased predictive (not causal) effects on the change in the labour status transition probabilities (P EUc,t P EUc,t ). In our view, assessing the predictive ability of institutional variables in explaining changes in the transition probability of flows during the crisis by comparable households across Euro area countries is in itself of considerable economic interest 9. However, an alternative and stronger claim is that γ reflects the causal impact of the institution z c on the change in the labour status transition probability. That interpretation requires ruling out endogeneity with respect to interacted country effects, arguably present in an observational cross-sectional setting such as ours. As mentioned above, the two step procedure we follow implies that each individual coefficient γ would be biased if an omitted institution were correlated with the interaction z c x ic, but not if it were correlated with other country fixed effects or country slope effects. In that sense each individual estimated effect has a stronger claim to causal validity than any effect estimated from, for example, the pooled regression. 9 Assume that there is a country-specific omitted characteristic that results both in a lower temporal employment and in a smaller response of the employment to unemployment flow of young, low educated and male workers. In this scenario, our estimate of γ would not reflect a causal impact of temporal employment on the probability of transition of the male, young and low educated workers. However, the statement that in Euro area countries with lower temporal employment, male, youngest and low educated workers had a lower growth of transition probability from employment to unemployment during the crisis would still be correct. 8

10 5 First-stage results: changes in the socio-demographic composition of flows and the role of some labour market characteristics. 5. From employment to unemployment Figures 5a and 5b provide, respectively, the change in the transition probability from employment to unemployment between and by socio-demographic groups, and the contribution of each one of them to the change in the aggregate flow during these two periods. With respect to the countries where unemployment barely changed, only in Austria and Belgium there seems to be a (statistically significant) slight increase in the flow from employment to unemployed, for males without higher education in the 3-49 age cohort. On the contrary, in the countries where unemployment raised the most, the increase in this flow is spread among all population groups. Admittedly, it is larger for those without higher education, but also larger for the youngest (5-29) regardless of their educational attainment than for older workers without education. During the second half of the sample (which can be assessed by comparing the results for with those for 27-22), the flows from employment to unemployment increased in Portugal and Greece, and, more, moderately in Italy and Spain. Another interesting pattern in the very high-unemployment countries, also somehow shared by the countries with a moderate increase of unemployment, is that the probability of losing the job increased more for males than for females. This gender difference might be explained by the higher incidence of the crisis on some specific sectors where male are overrepresented. To check this hypothesis we have interacted the gender dummy with a dummy variable that identifies workers in the construction sector one of the hardest hit in this recession and also one with the highest incidence of male employment- (see Appendix A.3). Results seem to back this hypothesis in Spain, Greece and United Kingdom, but in the rest of the countries there does not seem to be a sectorial composition effect behind the gender differences. As stressed by the literature (e.g. Nickel, 979; Ashenfelter et al, 979; and Mincer 99) education provides a powerful protection against the odds of losing the job. However, during the Great Recession in the EU countries where the increase in unemployment was higher, it seems that age, rather than education, is the main determinant of flows from employment into unemployment, with young workers, regardless of their levels of 9

11 education, specially hit the employment losses. While there are reasons to expect that productivity (and the incidence of productivity shocks) may have some relation with age, it seems more likely that the very uneven distribution of these flows across age-groups is mostly explain by the bias of Employment Protection Legislation (EPL) in favour of older workers, especially in those countries where duality, that is the existence of different layers of EPL for different jobs and workers, is especially acute. Dual labour markets feature to a high share of temporary employment contracts with lower employment protection. Studies show that higher employment protection leads to lower job separation rates by increasing the cost of firing but also diminish the flow from unemployment to employment by increasing the reluctance to hire workers in the first place. The results of the regression of the share of temporary workers (and the protection of temporary workers), as a proxy of EPL, on the change in the transition probability from employment to unemployment, are in line with the theory. We show a special incidence of this institution among the youngest cohorts (See Appendix A4.). Young workers were employed on temporary contracts more than adult in the previous years to the crisis and therefore the effect of EPL on the probability to move from employment to unemployment was higher among them (See Appendix A4.2). Besides, the disparity between the adult and young in this regard is particularly large in Spain, which have had some of the largest increase in young unemployment. Finally, it is important to weight the change in the probability of losing the job for each particular socio-demographic group into the aggregate change for the whole population. This is done in Figure 5b. As seen in the Figure, in most countries the most sizeable relative contribution is from medium-age males with low/medium educational attainments, except in Sweden and the UK where young workers contributed to a much noticeable extent to increase the overall flow from employment to unemployment. 5.2 From unemployment to employment The country level results for the OLS regressions are presented in Figure 2 and Table 3 where we show the estimated coefficients and their confidence intervals, as well as the results from the location scale model that provides a test of heteroscedasticity. The omitted categories for the dummy variables are detailed in Section 3.2. The first chart on the top row of Figure 2 reports the results for the reference group. Bell and Blanchflower (2) have also stressed that during the Great Recession young workers have suffered disproportionately the raise of unemployment.

12 Across all countries, the log amount of secured debt holdings is highest for households where the reference person is aged 6 to 34 years. Households with heads older than 45 years hold lower amounts of debt relative to the omitted group, 35 to 44 year olds. The full triangles below zero for the variable age 6-34 in Figure 2 are those of the location scale model and indicate that, conditional on covariates, the within-country dispersion of the logarithm of secured debt is lowest for households aged However, the dispersion in log debt amounts increases among older cohorts - notice the blue triangles above zero for the rest of the age groups. A possible interpretation is that while young adults borrow a similarly (large) amount of funds early in the life cycle, the speed of repayment of this debt varies across households, resulting in a progressively higher dispersion of outstanding debt amounts over the life cycle. Turning to the coefficients on the education variables, it is clear that higher education is associated with higher debt levels. The cross-country differences in the income profile of the amount of secured debt held are relatively similar to those corresponding to the fraction of borrowers and, as mentioned above, consistent with basic predictions of the permanent income model. Finally, we show the R-squared value for the estimated regressions in the final chart of Figure 2 and Table 3. Similar to the first specification shown in Figure, we find that the value of the R-squared varies from a low of. in Italy and Spain to a high of.32 for Luxembourg. 5.3 A comparison across crises In a complementary exercise we question whether the demographic impact over the labour market flows of the Great Recession has shown a differential behaviour in comparison with the previous crises that took place at the beginning of the nineties. With that objective, we replicate the analysis of section 4. and 4.2 and compare the magnitudes of the effects in each demographic group of both crisis periods. Given data limitations, we have done this comparison only for four countries Belgium, Spain, United Kingdom and Portugaland we restrict our analysis to the gender-age demographic dimension. Figure 7a shows the comparison across crises for the probability of flowing from employment to unemployment. The blue bars represent the ratio between the probability of losing the job during the Great Recession and the pre-crisis year, while the green dots represent the same ratio during the early nineties crisis. EU-LFS provides the education breakdown only from 992.

13 We observe that in Portugal, Belgium and United Kingdom the increase occurred during the Great Recession across demographic groups resembles their behaviour during the early nineties crisis. However, in Spain, we identify two singularities of the recent crises. Firstly, the probability of losing the job has increased proportionally more among males, while among females the proportional increase has been of the same magnitude in both episodes. This is, we clearly identify a differential gender component in the Great Recession, probably driven, as we stated above, by the asymmetric shock to the construction sector that characterized the Spanish Great Recession. Secondly, young males have experienced a higher proportional increase in the probability of losing the job than middle-aged males, while in the crisis of the early nineties both groups suffered the same proportional increase. Regarding the probability of finding a job (see Figure 7b), we observe that while in Portugal and United Kingdom the magnitude of the proportional drop has been fairly similar in both crises, in the case of Spain, the options of finding a job have been cut down relatively more during the Great Recession. Moreover, in this economy, the distribution of the incidence across demographic groups has not remained stable: the Great Recession has undermined the chances of the younger cohorts finding a job in a greater proportion. Therefore, while during the early nineties crises, the highest cut down was suffered by middle-aged males, during the recent crises the young males have been the hardest hit. 6 Second stage results: the influence of labour market institutions 7 The aftermath of the Great Recession in terms of inter-sectoral reallocation A central question in economics is how business cycles affect the allocation of resources. Recessions can be seen as being cleansing, because they are times when outdated techniques and products are squeezed out of the market (Caballero and Hammour, 994). Focusing on the labour resources, models with endogenous separations imply that recessions cleanse the labour market as low quality matches are destroyed and only exceptionally high quality matches are created (Mortensen and Pissarides, 994). However, this class of models ignores the potentially offsetting sullying effect of the crisis (Barlevy, 22) that implies that during recessions job-to-job transitions lead to worse matches, on average, because workers accept any match exceeding the quality of their currently low productive match 2

14 In this context, one of the current issues under debate is whether the Great Recession is bringing any reallocation to more productive jobs in the European countries. The EU- LFS, although with some limitations 2, allows us to give some light on that debate from the point of view of worker flows, this is, looking at the sectors where unemployed are being hired or where workers were fired. When we look at the correlation of the sectoral distribution of job finding across years with that of 28 (left panel of Figure 8), we observe that, with the exception of Spain and Greece, the sector distribution of job finding during the period was very similar to that of 28: the proportion of jobs created in each sector was virtually the same before and all over the Great Recession. Another way of approaching the question is investigating whether or not hirings are taking place in the sectors that previously destroyed them. Right panel of Figure 8 shows that in 29, maybe due to the weakness of job finding, the correlation of the sector distribution of hirings and firings was very low. However, the correlation of the sector distribution of job finding in each year with that of job firing in 29 was increasing over the years. This evidence supports again the idea that there is no systematic evidence in favour of the cleansing effect at sectoral level 3. Moreover, as Figure 9 also details, the intensity of sector reallocation has been very heterogeneous across countries and seems to be higher in those countries that have experienced a sizeable increase in the unemployment rate. 8 Conclusions The economic literature has extensively analyzed the role of worker flows in determining the evolution of the unemployment rate. However, most of the studies have been conducted using aggregate data, which has the advantage of having a higher frequency and being more up-to-date than micro data, but with the drawback of not identifying the socio-demographic characteristics of the individuals that are behind those flows. In this paper, by using micro data, we highlight four findings that are informative of the employment consequences of the Great Recession. 2 The EU-LFS provides a broad classification of sector of activity distinguishing only between 5 categories: agriculture, mining, manufacturing, electricity, construction, wholesale and retail, transport, hotels and restaurants, financial intermediation, other business, public administration, education, health, entertainment services and household. Besides, given the NACE reclassification, categories are only fully comparable from 28 onwards. 3 Under this approach, given that we do not have information on the firm, we cannot conclude whether or not cleansing effects are taking place within sector, across firms. 3

15 First, using a homogeneous micro data set, we identify across eleven European countries the contribution of workers flows to the change in the unemployment rate that we have observed during the recent crisis. In line with the literature, we obtain that the increase in the probability of losing the job and becoming unemployed seems to be the main determinant of the rise in the unemployment rate, followed by the decrease in the job finding probability among unemployed. Secondly, given that we use a micro data set, we are able to observe the main sociodemographic characteristics of the individuals transiting among employment states. In particular, we get to characterize that the increase in the probability of losing the job has mainly be driven by the male workforce, independently, in most of the countries, of the sector of activity; and that the insurance role of education against unemployment has been milder, with age playing a very important role in some countries as determinant of the flow from employment into unemployment. Besides, the comparison of the demographic characteristics of the individuals transiting among employment states in this crisis with respect to one that took place at the beginning of the nineties shows that male and youngest households have been affected in a higher rate in this recession. Third, differences in the changes on the transition probabilities among employment state for each socio-demographic group could be explained by the different magnitude of the fall of economic activity in each subgroup and also by the responses of employment policies to the crisis across countries. But also, we find that the differences in labour market institutions across countries have explained why labour flows changed more in some sociodemographic groups than in others. Particularly.... Finally, despite the limitations of the EU-LFS to carry out a detailed analysis of intersectoral reallocation and although it is still too early to make a full assessment, we do not find yet important change in the distribution of job hiring and firing during the crisis. However, sectoral reallocation has been more intensive in those countries that have experienced a sizeable increase in the unemployment rate (Spain and Greece). Therefore, behind the widespread increase that we have observed in the unemployment rate during the Great Recession, workers flows have contributed in a different magnitude across economies; and that the demographic groups that drive the change in each of the flows, differ substantially across countries. The identification of the groups that are driving the rise in the unemployment rate and of the labour market institution that are conditioned this evolution should help policymakers to designing well oriented policies to reduce unemployment. 4

16 References [] Ashenfelter, O., and J. Ham (979): "Education, Unemployment, and Earnings", Journal of Political Economy, vol. 87(??), pp. S99-6, October. [2] Barlevy, Gadi. 22. The Sullying Effect of Recessions. Review of Economic Studies, 69(??): [3] Bell, D.H.F. and D.G. Blanchlower (2): Young People and the Great Recession, IZA Discussion Paper no [4] Burda, M. and C. Wyplosz (994): "Gross worker and job flows in Europe," European Economic Review, vol. 38(??), pp , June. [5] Elsby, M. B., Hobijn, and A.»ahin (23): Unemployment dynamics in the OECD, Review of Economics and Statistics, vol. 95(??), pp [6] Garrido, L. and L. Toharia (24): What does it take to be (counted as) unemployed? The case of Spain, Labour Economics (24) [7] Jones, S.R.G. and Riddell, W.C. (999): The measurement of unemployment: an empirical approach. Econometrica, 67-, [8] Jones, S.R.G. and Riddell, W.C., (22): Unemployment and Non-Employment: Heterogeneities in Labour Market States. Department of Economics Working Paper No McMaster University. [9] Mincer, J. (99): Education and Unemployment, NBER, working paper [] Mortensen, Dale T., and Christopher A. Pissarides Job Creation and Job Destruction in the theory of unemployment. Review of Economic Studies, 6(??): [] Nickel, S. (979): "Education and Lifetime Patterns of Unemployment", Journal of Political Economy, vol. 87(??), pp. S7-3, October. [2] OECD (2), Employment Outlook, OECD, Paris. [3] Pissarides, C. (2). Equilibrium Unemployment Theory, 2nd Edition, MIT Press Books, The MIT Press [4] Shimer, R. (22. "Reassessing the Ins and Outs of Unemployment," Review of Economic Dynamics, vol. 5(??), pages 27-48, April. 5

17 Table : SUMMARY STATISTICS COUNTRIES VARIABLES AT BE DE DK ES FR GR IT PT SE UK Age (%) Education (%) Primary and lower secondary Upper and post- secondary Tertiary Gender Male Female Unemployment rate Inactivity rate S ample period

18 Figure : UNEMPLOYMENT RATE AND LABOUR FLOWS UNEMPLOYMENT)RATE)AND)LABOUR)FLOWS UNEMPLOYMENT)RATE EMPLOYMENT)* UNEMPLOYMENT UNEMPLOYMENT)* EMPLOYMENT AT BE DE AT BE DE AT BE DE GR PT ES GR PT ES GR PT ES FR DK IT UK SE FR DK IT UK SE FR DK IT UK SE 7

19 Figure 2: LONG-TERM UNEMPLOYMENT Unemployment Duration (% of unemployed with duration > year) UNE MP L OY ME NT *DUR AT ION*(% *of*unemployed*with*durations *> *year) AUSTRIA BELGIUM GERMANY GREECE PORTUGAL SPAIN ITALY FRANCE DENMARK UNITED>KINGDOM SWEDEN SWEDEN 8

20 4 Figure 3: ANNUAL AND QUARTERLY TRANSITION RATES to Employment*to*unemployment*flow EU_LFS/Spain (Cross9Sectional/retrospective/question) LFS/Spain/ (Longitudinal/quarterly) LFS/Spain (Longitudinal/annual) LFS/Spain/ (Longitudinal/quarterly/annualized) 7 Unemployment to employment Umployment*to*employment*flow EU_LFSSpain (Cross;Sectionalretrospectivequestion) LFSSpain (Longitudinalquarterly) 9 LFSSpain (Longitudinalannual) LFSSpain (Longitudinalquarterlyannualized)

21 !! Figure 4: WORKERS FLOWS CONTRIBUTION TO THE CHANGE IN THE UNEM- PLOYMENT RATES: % 35% 3% 3% 25% 25% 2% 2% 5% 5% % % 5% 5% % %!5% AT BE DE DK FR IT SE UK ES GR PT hiu heu hei hue hui hie "residual"!5% % % 8% 8% 6% 6% 4% 4% 2% 2% % %!2%!2%!4%!4%!6%!6%!8%!8%!% AT BE DE DK FR IT SE UK ES GR PT hiu heu hei hue hui hie "residual"!%!!! 2

22 Figure 5a: FIRST STAGE RESULTS: SOCIO-DEMOGRAPHICAL FLOW DECOMPO- SITION Probability growth of flowing from employment to unemployment AT BE DE !2.!2.!2. 6!3 3!5 5!65 6!3 3!5 5!65 FE!!! 6!3 3!5 5!65 6!3 3!5 5!65 6!3 3!5 5!65 6!3 3!5 5!65 FE FE PT ES GR !2.!2.!2. 6!3 3!5 5!65 6!3 3!5 5!65 FE!!! 6!3 3!5 5!65 6!3 3!5 5!65 6!3 3!5 5!65 6!3 3!5 5!65 FE FE IT FR DK !2.!2.!2. 6!3 3!5 5!65 6!3 3!5 5!65 FE!!! 6!3 3!5 5!65 6!3 3!5 5!65 6!3 3!5 5!65 6!3 3!5 5!65 FE FE SE UK !.2 FE FE 29VS27 FE FE FE FE FE FE FE 22VS LOWEDUCATION MEDIUMEDUCATION HIGHEDUCATION LOWEDUCATION MEDIUMEDUCATION HIGHEDUCATION LOWEDUCATION MEDIUMEDUCATION HIGHEDUCATION 6!35 36!55 56!65.!2. 6!3 3!5 5!65 6!3 3!5 5!65 FE.!2.!! 6!3 3!5 5!65 6!3 3!5 5!65 FE!! Note: Each blue bar is the difference in the probability of flowing from employment to unemployment between 29 and 27 for each specific group. Dark blue indicates that the difference is significant and light blue that it is not significant. Each red dot is the difference in the probability of flowing from employment to unemployment between 22 and 27 for each specific group. When the dot is not colored, it means that the difference is not significant. 2

23 Figure 5b: FIRST STAGE RESULTS: SOCIO-DEMOGRAPHICAL FLOW DECOMPO- SITION Contribution of each demographic group to the change in the employment to unemployment aggregate flow AT BE DE !.2 6!3 3!5 5!65 6!3 3!5 5!65 FE!.2!.2!!! 6!3 3!5 5!65 6!3 3!5 5!65 6!3 3!5 5!65 6!3 3!5 5!65 FE FE PT ES GR !.2 6!3 3!5 5!65 6!3 3!5 5!65 FE!.2!.2!!! 6!3 3!5 5!65 6!3 3!5 5!65 6!3 3!5 5!65 6!3 3!5 5!65 FE FE IT FR DK !.2 6!3 3!5 5!65 6!3 3!5 5!65 FE!.2!.2!!! 6!3 3!5 5!65 6!3 3!5 5!65 6!3 3!5 5!65 6!3 3!5 5!65 FE FE SE UK !.2 FE FE 29VS27 FE FE FE FE FE FE FE LOWEDUCATION MEDIUMEDUCATION HIGHEDUCATION LOWEDUCATION MEDIUMEDUCATION HIGHEDUCATION LOWEDUCATION MEDIUMEDUCATION HIGHEDUCATION 22VS27 6!35 36!55 56!65.2.2!.2 6!3 3!5 5!65 6!3 3!5 5!65 FE!.2!! 6!3 3!5 5!65 6!3 3!5 5!65 FE!! Note: All the blue bars sum up to the aggregate variation in the employment to unemployment rate between the period 27 and 29. All the red dots sum up to the aggregate variation in the employment to unemployment rate between the period 27 and

24 ! Figure 6a: FIRST STAGE RESULTS: SOCIO-DEMOGRAPHICAL FLOW DECOMPO- SITION Probability growth of flowing from unemployment to employment AT BE DE !.!2.!3..!.!2..!.!2.!4.!3.!3.!5.!4.!4. 6!3 3!5 5!65 6!3 3!5!5. 6!3 3!5 5!65 6!3 3!5!5. 6!3 3!5 5!65 6!3 3!5 FE FE FE PT ES GR !.!.!.!2.!2.!2.!3.!3.!3.!4.!4.!4.!5.!5.!5. 6!3 3!5 5!65 6!3 3!5 5!65 FE!!! 6!3 3!5 5!65 6!3 3!5 5!65 6!3 3!5 5!65 6!3 3!5 5!65 FE FE IT FR DK !.!.!.!2.!2.!2.!3.!3.!3.!4.!4.!4.!5.!5.!5. 6!3 3!5 5!65 6!3 3!5 5!65 FE!!! 6!3 3!5 5!65 6!3 3!5 5!65 6!3 3!5 5!65 6!3 3!5 5!65 FE FE !. SE !. UK !.2 FE 29VS27 FE FE FE FE FE FE FE FE 22VS27!2.!3.!2.!3. LOWEDUCATION MEDIUMEDUCATION HIGHEDUCATION LOWEDUCATION MEDIUMEDUCATION HIGHEDUCATION LOWEDUCATION MEDIUMEDUCATION HIGHEDUCATION 6!35 36!55 56!65!4.!5. 6!3 3!5 5!65 6!3 3!5 5!65 FE!4.!5.!! 6!3 3!5 5!65 6!3 3!5 5!65 FE!! Note: Each blue bar is the difference in the probability of flowing from unemployment to employment between 29 and 27 for each specific group. Dark blue indicates that the difference is significant and light blue that it is not significant. Each red dot is the difference in the probability of flowing from unemployment to employment between 22 and 27 for each specific group. When the dot is not colored, it means that the difference is not significant. 23

25 Figure 6b: FIRST STAGE RESULTS: SOCIO-DEMOGRAPHICAL FLOW DECOMPO- SITION Contribution of each demographic group to the change in the unemployment to employment aggregate flow AT BE DE !!!!2!2!2!3!3!3!4!4!4!5 6!3 3!5 5!65 6!3 3!5 5!65 FE!5!5!!! 6!3 3!5 5!65 6!3 3!5 5!65 6!3 3!5 5!65 6!3 3!5 5!65 FE FE PT ES GR !!!!2!2!2!3!3!3!4!4!4!5 6!3 3!5 5!65 6!3 3!5 5!65 FE!5!5!!! 6!3 3!5 5!65 6!3 3!5 5!65 6!3 3!5 5!65 6!3 3!5 5!65 FE FE IT FR DK !!!!2!2!2!3!3!3!4!4!4!5 6!3 3!5 5!65 6!3 3!5 5!65 FE!5!5!!! 6!3 3!5 5!65 6!3 3!5 5!65 6!3 3!5 5!65 6!3 3!5 5!65 FE FE!!2 SE!!2 UK !.2 FE FE 29VS27 FE FE FE FE FE FE FE LOWEDUCATION MEDIUMEDUCATION HIGHEDUCATION LOWEDUCATION MEDIUMEDUCATION HIGHEDUCATION LOWEDUCATION MEDIUMEDUCATION HIGHEDUCATION 22VS27!3!3 6!35 36!55 56!65!4!5 FE FE FE FE FE FE FE FE FE LOW/ MEDIUM/ HIGH/ LOW/ MEDIUM/ HIGH/ LOW/ MEDIUM/ HIGH/ EDUCATION EDUCATION EDUCATION EDUCATION EDUCATION EDUCATION EDUCATION EDUCATION EDUCATION 6!3 3!5 5!65!4!5 FE FE FE FE FE FE FE FE FE LOW/ MEDIUM/ HIGH/ LOW/ MEDIUM/ HIGH/ LOW/ MEDIUM/ HIGH/ EDUCATION EDUCATION EDUCATION EDUCATION EDUCATION EDUCATION EDUCATION EDUCATION EDUCATION 6!3 3!5 5!65!!!! Note: All the blue bars sum up to the aggregate variation in the unemployment to employment rate between the period 27 and 29. All the red dots sum up to the aggregate variation in the unemployment to employment rate between the period 27 and