Labour market resilience in Europe

Labour market resilience in Europe INSPIRES Benchmark Report Version : 1 6 214 Erasmus University Rotterdam Bigos, M., Qaran, W., Fenger, M., Koster, F., & Veen, R. van der

Table of contents 1. List of tables and figures 5 2. Introduction 6 3. Analysing labour market resilience 1 2.1 Operationalisation 11 2.1.1 Dependent variables 12 2.1.2 Independent variables 13 2.2 Data 15 2.3 Analysis 15 4. Results 18 3.1 Regression results 19 3.1.1 population 2 3.1.2 Youth 22 3.1.3 Older people 24 3.1.4 Migrants 26 3.2 s 28 3.2.1 population 28 3.2.2 Youth 31 3.2.3 Older people 34 3.2.4 Migrants 37 5. Country reports 4 Conclusions 158 3

List of tables and figures Table 1 Ordinary least squares (OLS) multiply linear regression models of Labour market resilience for total population, in 29 European countries (27 and 21) 16 Table 2 Ordinary least squares (OLS) multiply linear regression models of Labour market resilience for youth, in 29 European countries (27 and 21). 17 Table 3 Ordinary least squares (OLS) multiply linear regression models of Labour market resilience for older people, in 29 European countries (27 and 21). 18 Table 4 Ordinary least squares (OLS) multiply linear regression models of Labour market resilience for migrants (27 and 21) 19 Table 5 Unemployment rates (total population): 1. in actual s (21-27), 2. s (27); 3. s (21); 4. in residual s (RE 21 RE 27) 22 Table 6 At risk of poverty and social exclusion rate (total population) : 1. in actual s (21-27); 2. s (27) ; 3. s (21) ; 4. in residual s (RE 21 RE 27) 23 Table 7 Unemployment rates (youth): 1. in actual s (21-27); 2. s (27); 3. s (21) ; 4. in residual s (RE 21 RE 27) 25 Table 8 At risk of poverty and social exclusion rate (youth) : 1. in actual s (21-27); 2. s (27) ; 3. s (21) ; 4. in residual s (RE 21 RE 27) 26 Table 9 Unemployment rates (older people): 1. in actual s (21-27); 2. s (27); 3. s (21) ; 4. in residual s (RE 21 RE 27) 28 Table 1 At risk of poverty and social exclusion rate (older people) : 1. in actual s (21-27); 2. s (27) ; 3. s (21) ; 4. in residual s (RE 21 RE 27) 29 Table 11 Unemployment rates (migrants): 1. in actual s (21-27); 2. s (27); 3. s (21); 4. in residual s (RE 21 RE 27) 31 Table 12 At risk of poverty and social exclusion rate (migrants) : 1. in actual s (21-27); 2. s (27); 3. s (21) ; 4. in residual s (RE 21 RE 27) 32 Figure 1 Gross domestic product at market prices (% on the previous period), 2-212. 11 4 5

chapter 1 Introduction This report is part of the first Work Package of the INSPIRES 1) research project, funded by the European Community s Seventh Framework Programme. It provides the follow up of the INSPIRES first and second deliverables, namely the Review Essay of Labour Market Resilience (Bigos et al., 213) and European Labour Market Resilience Dataset (ELMar). 1) The overall objectives of the INSPIRES research project can be summarized as follows: To conduct comparative assessments of the resilience and inclusiveness of labour markets in European countries; To investigate the role of innovative policies that have contributed to the resilience and inclusiveness of labour markets in Europe; To analysis strategies of policy learning that facilitate the development and transfer of these innovations within and across European nation states. One of the main goals of the INSPIRES project is to accumulate practice-oriented knowledge on the factors that positively and negatively affected resilience and inclusiveness of the labour markets in Europe in the aftermath of the 28 global economic crisis. The INSPIRES project is not only interested in the resilience of countries but also of specific vulnerable groups in the labour market. Consequently, the guiding principle of this report is to assess differences in labour market resilience between countries, and between different vulnerable groups in the labour market. In order to evaluate labour market resilience, special attention will be given to the changes that may occur in different outcome measures, such as unemployment rates or the levels of poverty and social exclusion, in the aftermath of crisis. The analysis methodologically adopts a cross-sectional approach and focuses on two points in time - before and after/during the crisis- in order to observe possible changes that may be associated with the crisis. Consequently, the following report presents a cross-country analysis of the labour market resilience of the total population, youth, older people and migrants 2) across 29 European countries, including 27 European Union Member States, Norway and Switzerland. Labour market resilience is the main concept in this report and within the entire INSPIRES project. It refers to the capacity of labour markets to absorb external shocks and mitigate their negative impact for unemployment levels as 2) As a result of data limitations for ddisabled persons, this group has been excluded from the analysis. well as the poverty and social exclusion among the labour force (see for the detailed discussion Bigos et al., 213). Although all the European countries have been affected by the crisis, there are considerable differences in the way countries managed to mitigate the negative labour market consequences. For example, while some countries almost doubled their unemployment rates, other countries proved to be more resilient and managed to maintain relatively low increases. These differences become even more apparent if we look at the position of specific vulnerable groups within European labour markets. For instance, youth unemployment in countries like Germany and the Netherlands seem to be hardly affected at all by the economic crisis, whereas countries like Spain and Greece have extremely dramatic increases. But what makes labour markets resilient? While the scale of a shock or disturbance is an important factor, it is not entirely clear what precisely constitutes the ability of the labour markets to adapt to challenges. Are the institutional arrangements or maybe structural characteristics of labour markets of key importance? This empirical puzzle - trying to understand what constitutes the strengths and weaknesses of the adaptive capacities of national labour markets in Europe - has triggered an increased interest in the concept of labour market resilience from both scholars and policy makers. Against this background, the aim of this report is to contribute to the ongoing debate on labour market resilience in several ways. First, the report extends previous studies that explicitly or implicitly investigate labour market resilience (e.g. Auer and Cazes, 2; Chapple and Lester, 21, Bartolucci et al., 211; Boeri et al., 212; OECD, 212) 6 7

Chapter 1 Introduction by examining not only the labour market resilience at the national (aggregate) level, but also by including a vulnerable group level perspective. Consequently the report provides an assessment of the position of youth, migrants and older people across 29 European countries. Furthermore, the report focuses both on the causes (factors affecting resilience) and the outcomes (i.e. unemployment, poverty, social exclusion). In addition, we provide an attempt to create a ranking of countries labour market resilience for the total population, youth, migrants and older persons. All in all, the findings presented in this report provide a basis for further milestones of the INSPIRES project. The next step of the INSPIRES project is to provide explanation for the differences in resilience between groups and countries, by adopting qualitative approach. This report is structured as follows. In chapter 2 we briefly discuss the key concept of labour market resilience, and elaborate on the operationalization of the key dependent and independent variables. In chapter 3, we discuss the methodological background of our analyses. In chapter 4 we present the results of the regression models, as well as the rankings of countries based on the actual data and changes in residuals. These results are presented for the total population, youth, older persons and migrants. The final country chapters elaborate more in detail on the differences within countries, with a special focus on the differences between vulnerable groups. Finally, we sum up the most important findings and implications of this report. The findings presented in this report suggest that the factors contributing to labour market resilience are multiple, change over time, and often vary across different vulnerable groups. Another finding (as illustrated in the country chapter) is that youth and migrants have been the most vulnerable group across Europe in the aftermath of the recent crisis, both in terms of the highest increases in unemployment rates and the risk of poverty and social exclusion. Furthermore, the report suggests that in addition to the focus on factors affecting resilience, it also is important to shed light on temporal boundaries that frame the notion of resilience. Labour market resilience depends on different labour market dynamics over time (OECD, 212). Consequently it is recommended to observe the changes in outcomes also over relatively longer period of time after the occurrence of shock. 8 9

chapter 2 Analysing labour market resilience 2.1 Operationalisation The following section presents the main concepts and variables used in the analysis. The operationalization is based on the conceptual framework presented in the INSPIRES Working Paper no. 1, (see Bigos et al., 213). We define labour market resilience as the capacity of labour markets to absorb external shocks and mitigate their negative impact for unemployment levels as well as the poverty and social exclusion, for the total population as well as specific vulnerable groups There are many different approaches to resilience which offer valuable tools for understanding the concept. For example, the engineering concept of resilience focuses on the stability of a system near an equilibrium or steady state and therefore, it emphasises the resistance of the system to a disruption and its speed of return to the pre-disturbance equilibrium (e.g. Holling, 1973). In contrast, the ecological resilience assumes that a system has multiple stability domains. Therefore, if a shock pushes a system away from its elasticity threshold it may move to a new domain of stability. Furthermore, the adaptive concept of resilience stresses the capacity of a system to reconfigure (adapt) its structure (firms, industries, technologies and institutions) in order to maintain an acceptable growth path in output, employment and wealth in the long run. In this report, we draw on these different approaches to resilience. First, in line with equilibrium approach we focus on specific outcome measures that may resume either their pre-shock levels or show new post-shock trajectories (Pendall et al., 21). Therefore, the equilibrium-based approaches to resilience provide us with a guideline for the empirical assessment by focusing on the performance of certain outcomes, such as the rates of unemployment and poverty and social exclusion. Second, drawing on adaptive approach to resilience, which stresses the importance of contextual (structural) conditions that determine whether a system has the capacity to reconfigure (adapt), we include a number of contextual factors in our analysis. Moreover, in line with the Varieties of Capitalism (VoC) literature, which suggests that the extent to which an economy is regulated and managed influences its capacity to react to economic challenges (Hall and Soskice, 21), we also include some of the institutional features of labour market regulation (e.g. working hours, type of contract). The first key concept for understanding labour market resilience is an exogenous shock or a challenge that triggers particular feedback mechanisms, such as specific type of policy responses. Here the concept of challenge describes the economic recession, caused by the global financial crisis, which greatly deteriorated labour markets, through the decline in productivity growth and an increase in systemic uncertainty (Bartolucci et al., 211; Jamet, 211). A second cluster of variables relates to structural features or broader context that shapes national labour markets. These relate specifically to the institutional, socio-economic demographic characteristics. Institutional structure generally refers to the systems of laws, regulations and procedures that shape employment relationship. Under the heading of socio-economic conditions we distinguish three types of 1 11

Chapter 2 Analysing labour market resilience variables, namely: firm size, regional disparities and industry structures. Demographic characteristics refer to population structure, educational level, and migration patterns. The final key concept of labour market resilience model is the outcome through which we can observe an impact of the crisis (i.e. labour force participation, poverty, social exclusion). Although, we have identified a large number of factors, we could not include al of them in the analysis. First, omitting these variables results from data limitations. Second, some of the excluded variables either appeared to have no significant effect on the key dependent variables, or they significantly undermined the goodness of fit of our regression model. Therefore, we have selected a final number of variables discussed below. 2.1.1 Dependent variables Measuring labour market resilience requires assessing the changes in outcomes experienced by specific groups in the labour market, as a result of the crisis. In other words, the outcome measures refer to the way and the extent to which the labour market have been affected by the challenges that the crisis poses. Consequently, one of the ways through which labour market resilience can be measured relates to the quantitative dimension of labour force participation, which can be measured for example by the unemployment rate. Another option is to account for the qualitative dimension of labour market participation, such as for example the involuntary part-time employment, which gives an indication of the potentially negative trend, not necessarily related to the job loss. Furthermore, the persistence of the negative tendencies can be captured by distinguishing between long term and short term effects (i.e. long-term unemployment rate). Finally, while the changes in both the quantitative and the qualitative dimension of labour market outcomes can be seen as a direct consequence of the crisis, it is also important to include more indirect consequences. Hence, we also include the concept of poverty and social exclusion, which refers primarily to exclusion from the labour market (i.e. unemployment). The process of social exclusion assumes a downward spiral or a vicious circle in which labour market marginality leads to poverty and social isolation, which in turn reinforce the risk of long-term unemployment, that imposes constraints on job search, fracturing of people s social ties and growing social isolation, psychological distress, tensions within the family and marital dissolution (Gallie et al., 23). Unemployment rate Based on the Eurostat definition, the unemployment rate is the number of people unemployed as a percentage of the labour force. Unemployed persons are all persons who were not employed during the reference week, had actively sought work during the past four weeks and were ready to begin working immediately or within two weeks (Eurostat, 213e). In addition to unemployment rate we include the long term unemployment rate, which refers to the share of unemployed persons since 12 months or more in the total number of active persons in the labour market. In-voluntary part-time employment, describes persons working on an involuntary part-time basis, who declare that they work part-time because they are unable to find full-time work. Poverty and social exclusion We use AROPE (At risk of poverty and exclusion) as an indicator that is capable of measuring both concepts (Eurostat, 213f). This indicator contains the following three sub-indicators: Poverty risk rate: people with an equalised disposable income below the poverty threshold (set at 6 % of national median disposable income (after social transfers). Severe deprivation: material deprivation is an indicator of European Union Statistics on Income and living conditions (EU-SILC) that expresses the inability to afford some items considered by most people to be desirable or even necessary to lead an adequate life. Low labour intensity: Persons are considered living in households with very low work intensity if they are aged -59 and the working age members in the household worked less than 2 % of their potential during the past year (European Commission, 213). 2.1.2 Independent variables The first key independent variable is a challenge or a shock, which refers to an economic recession triggered by external shock. The literature suggest that there are two main ways through which the crisis can affect labour markets: Figure 1 Gross domestic product at market prices (% on the previous period), 2-212. Gross domestic product at market prices (% on the previous period) 2. 15. 1. 5.. -5. -1. -15. -2. Source: Own calculation based on annual data provided by the Eurostat 2 22 24 26 Year first, through a general decline in productivity growth and second, through an increase of an atmosphere of the systemic uncertainty, which often leads to wait and see behaviour of employers (Jamet, 211; Boeri et al., 212). In our analysis we focus on productivity growth and therefore, use the Gross domestic product (GDP) in purchasing power standards, as a measure. As illustrated in figure 1 there was a sharp decline in productivity growth recorded across all European countries associated with the outbreak of the crisis. It is therefore, interesting to assess to what extent these drops had an impact on the labour market. 28 21 212 Austria Belgium Bulgaria Cyprus Czech Republic Denmark Estonia Finland France Germany Greece Hungary Ireland Italy Latvia Lithuania Luxembourg Malta Netherlands Norway Poland Portugal Romania Slovakia Slovenia Spain Sweden Switzerland United Kingdom 12 13

Chapter 2 Analysing labour market resilience The second group of independent variables relates to contextual characteristics of labour markets, specifically institutional (i.e. expenditure on labour market policies, working time, type of contract, labour tax) socio-economic (i.e. regional disparities) and demographic factors (i.e. educational level). Expenditure on (total) labour market policies Labour market interventions are the public interventions in the labour market, which aim at correcting disequilibria and which can be distinguished from other general employment policy interventions in that they act selectively to favour particular groups in the labour market. (Eurostat, 213g). These interventions can be categorized into three main groups: 1. LMP services concern all publicly funded services for jobseekers (guidance, counselling and other forms of job-search assistance) as well as any other expenditure of the public employment services (PES) not already covered in other LMP categories. 2. LMP measures (active interventions) cover interventions that aim either to provide people with new skills or experience of work in order to improve their employability or to encourage employers to create new jobs and take on people who are unemployed or otherwise disadvantaged. 3. LMP supports (passive interventions) mostly cover financial assistance designed to compensate individuals for loss of wage or salary and to support them during active job-search (i.e. mostly unemployment benefits). We use the Expenditure on total labour market policies, as a % of GDP, which consist of the following labour market interventions: labour market services, training, job rotation and job sharing, employment incentives, supported employment and rehabilitation, direct job creation, start-up incentives, out-of-work income maintenance and support and early retirement (Eurostat, 213g). Working time We use the Average number of usual weekly hours of work in main job, of total employed persons (Eurostat, 213b). The working hours variables are related to the employment protection legislation conditions, which regulate the maximum working hours for employees in order to ensure their well-being. Moreover, the concept of working time is also associated with contract type, especially to the concept of full employment. Usually, the concept of working time is assessed by indicators such as working hours annually, weekly and daily. Type of contract - temporary employment We use the Share of temporary employment, as a percentage of total dependent employment, (which includes also self-employed) (OECD, 213). Type of contract is closely related to employment protection legislation. Traditionally, labour contracts are divided along the dichotomy of permanent and temporary contracts. The expansion of temporary employment often raise concerns about increased job insecurity and growing labour market segmentation. Labour taxation We use the measure of the tax wedge on labour (Eurostat, 213c). The tax wedge provides one measure of the extent to which the tax system discourages employment, and can be defined as the difference between the salary costs of a single average worker for their employer and the net income ( take-home-pay ) that the worker receives (OECD, 27). The taxes included are personal income taxes, compulsory social security contributions paid by employees and employers, as well as payroll taxes for the few countries that have them. Regional disparities Regional disparities are reflected in an uneven ability of regions to adapt their labour markets in the face of a challenge (Chapple and Lester, 21). Regions in which there is a relative concentration of employment in cyclically sensitive industries, such as construction, tend to experience more cyclical variations in employment and unemployment than regions that are specialized in more cyclically stable industries (Robson, 26). We use the dispersion of regional GDP per inhabitant, % (Eurostat, 213d), which is measured by the sum of the absolute differences between regional and national GDP per inhabitant, weighted with the share of population and expressed in percent of the national GDP per inhabitant. The indicator is calculated from regional GDP figures based on the European System of Accounts (ESA95). The dispersion of regional GDP is zero when the GDP per inhabitant in all regions of a country is identical, and it rises where there is an increase in the distance between a region s GDP per inhabitant and the country mean. Educational level Education is an important attribute that enhances individual capacities and therefore, allows individuals to successfully and consistently perform an activity or task. We use the percentage of the adult population (25-64 years old) that has completed upper secondary education, as an indication of the share of the population that is likely to have the minimum necessary qualifications to actively participate in labour markets. It should be noted that completion of upper secondary education can be achieved in European countries after varying lengths of study, because of different national educational systems (Eurostat, 213e). 2.2 Data This report draws mostly on Eurostat data covering 29 European countries including:, 27 European Union Member States (Austria, Belgium, Bulgaria, Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherlands, Poland, Portugal, Romania, Slovakia, Slovenia, Spain, Sweden, United Kingdom), as well as Switzerland and Norway; from 2 till 212. The Eurostat data is particularly valuable as it provides the possibility to distinguish different groups on the labour market. Consequently, data representing dependent variables of labour market resilience are available for: the aggregate level and a group level (i.e. youth, older persons and migrants). We define youth as persons between the age of 15 and 24, older persons as people of age between 55 and 64. Migrants are defined as foreign citizens 3). Data on dependent variables related to labour force participation are provided by the Labour Force Survey and the EU Statistics on Income and Living Conditions (EU-SILC) (Eurostat). Data on institutional variables have been provided by both Eurostat and the Organization for Economic Cooperation and Development. Data on regional disparities and the level of education and Gross Domestic Product have been retrieved from the Eurostat. 4) 2.3 Analysis The empirical analysis follows a two-step approach. First, we estimate the strength of the causal relationship between variables, by using regression analysis. Specifically, ordinary least squares (OLS) multiple linear regression is applied in order to analyse data, consisting of national level variables (total population) and group level 3) Eurostat defines foreign population as people residing in an EU-27 Member State with citizenship of a non-member country. 4) More detailed information on specific data source is included in the Codebook of ELMAR Dataset (Bigos et al., 213a). 14 15

Chapter 2 Analysing labour market resilience variables (youth, older persons and migrants). The regression analysis allows estimating the nature and strength of causal relationship between variables. Furthermore, this analysis enables to combine many variables in order to produce an optimal prediction for the dependent variable and it separates the effects of independent variables on the dependent variable, so we can examine the unique contribution of each variable (Allison, 1999). In order to account for the differences that may have occurred as a result of the crisis we run each model twice: once for 27 and once for 21. Consequently, we present the results of 26 models (see Tables 1, 2, 3, 4). With small exceptions, each of the regression models has the same sets of independent variables, as discussed in the previous section. The general format of regression equation is as follows: Labour market resilience (UR, AROPE) = α + β1 GDP_PPS+ β2 Institutional (β2a Exp on lmp, β2b Tax, β2ctemp, β2d) + β3 Socio Economic (Regional Disparities) + β3 Demographic (Education) + εt, Besides the regression results we calculate the predicted s and (unstandardized) residuals for all observations, based on the estimated regression equations. The residual reflect the difference between observed (actual) and predicted s. These s are reported in detail for two dependent variables: unemployment rate and AROPE for the total population ad and specific vulnerable groups in the country chapter. The residuals are particularly useful in identifying outliers, or observations that deviate significantly from the predictions. Therefore, based on the calculated residuals (for 27 and 21) and change in residuals (21-27) we create a ranking of countries and compare them with ranking based on the actual changes in unemployment rates and AROPE. The ranking is composed by ordering countries according to the scale from the lowest to highest s for both dependent variables. Next we develop a classification, where the countries with the lowest s (below ), can be seen as more resilient, and those that have s well above are less resilient, based on our regression models. Put differently, countries with residual scores below the predicted have performed better than expected by our models, as they managed to maintain lower unemployment rates and AROPE rates than predicted. Labour market resilience is represented by two key dependent variables: unemployment rate and at risk of poverty and social exclusion, α is a vector of coefficients, β are the partial regression coefficients of independent variables, and εi is the error term that captures all other factors which influence the dependent variable. 16 17

chapter 3 Results The discussion of the findings begins with elaboration of the regression results, followed by presentation of the ranking of countries based on scores in residuals, which have been calculated based on regression models. In both cases our discussion is structured by vulnerable groups. 3.1 Regression results In tables 1, 2, 3 and 4 we report the results from estimating several different regression equations. For total population we have three dependent variables: unemployment rate, AROPE and long-term unemployment rate. For youth and older people we have a set of four dependent variables: unemployment rate, AROPE rate, long-term unemployment rate and involuntary part-time employment rate. For migrants we use both unemployment rate and AROPE rate. For each regression equation we use two sets of panel data. The first, consisting of data for 27 (before the crisis), the second for 21 (after the crisis/during recovery). For each regression equation two columns of numbers are shown. The column labelled B, contains the partial regression coefficients and the other labelled Std E contains their estimated standard errors. The row labelled Constant gives the estimates of the intercepts in the equations. The Adjusted R squared (the coefficient of multiple determination), is the measure of how well the model predicts the dependent variable, knowing only the set of independent variables in the model (Allison, 1999). In other words, the larger the of the R-squared, the better the set of explanatory variables predicts the dependent variable (Agresti & Finlay, 29). Consequently, R- squared is often considered a measure of how good the particular models are. Finally, F-statistics column provides the s of F distribution. The larger s of the F test statistic provide stronger evidence against H (Agresti & Finlay, 29). It is important to note that even if the P- is small for the F statistic, this does not imply that every independent variable has an effect on dependent variable, but merely that at least one of them has an effect (Agresti & Finlay, 29). 18 19

Chapter 3 Results 3.1.1 population Table 1 Ordinary least squares (OLS) multiply linear regression models of Labour market resilience for total population, in 29 European countries (27 and 21). 1. Unemployment rate 2. AROPE 3. Long term unemployment Model variable Model 1(27) Model 2(21) Model 3 (27) Model 4(21) Model 5(27) Model 6 (21) B Std E B Std E B Std E B Std E B Std E B Std E Gross Domestic Product (PPS) -.2.4.22 **.9 -.3.17 -.35 **.14 -.3.4 -.11**.5 Expenditure on LMP.97.64 2.71** 1.7 -.81 2.92.95 1.69.41.61 1.32**.59 Working hours.39 *.2.73.45.153.91.25.71.21.19.39.25 Share of temporary employment.8.6 -.12.14.117.26 -.6.22.1.5 -.1.8 Tax wedge.3.5 -.16.11.81.21.3.18.2.4 -.8.6 Dispersion of regional GDP.5.6.2.12.6 **.25.54 ***.18.4.5.5.6 At least upper secondary education -.2.2.5.6 -.5.61 -.8.9 -..2.1.3 Constant -11.51 8.57-14.36 19.57 8.1-39.13 9.71 3.79-7.17 8.19-8.55 1.69 Adjusted R2.3.247.37.47.6.269 F-statistsics 2.73 ** 2.31 * 3.41 ** 4.55 *** 1.24 2.47* P-.4.7.1..32.5 Number of observations 29 29 29 29 29 29 Note : *** significant at < 1% level (p<.1),** significant at < 5% level (p<.5),* significant at < 1% level (p<.1) The results of the regression analyses presented in table 1 suggest that working hours (Model 1), GDP (PPS) and Expenditure on (total) labour market policies (Model 2) are the significant predictors of unemployment rate for the total population. These variables are positively related to unemployment rate. Moreover, there is also a positive relationship between Expenditure on labour market and longterm unemployment rate. According to Eurostat (213g), the vast majority (63.4 %) of expenditures on labour market (LMP) interventions in 211 across the EU, financed passive interventions, which mostly cover unemployment benefits, while just over a quarter (25.7 %) was devoted to LMP measures (active interventions) and the remaining one ninth (1.9 %) was spent on LMP services (e.g. counselling). Furthermore, GDP has a negative effect on at risk of poverty and social exclusion rate, suggesting that along with an increase in productivity growth AROPE is decreasing (Model 4). Regional dispersion of GDP has a significant positive effect on AROPE. In other words, for an increase in unit of regional dispersion of GDP the AROPE will increase as well (Model 3 and 4). The adjusted R-squared are relatively high: 37% for Model 3 (27) and 47 % for Model 4 (21). Last, there is a negative relationship between GDP and long-term unemployment rate, suggesting that along with the increase in GDP, the long term unemployment rate decreases. 2 21

Chapter 3 Results 3.1.2 Youth Table 2 Ordinary least squares (OLS) multiply linear regression models of Labour market resilience for youth, in 29 European countries (27 and 21). Model variable 1. Unemployment rate 2. AROPE 3. Long term unemployment 4. Involountary part-time Model 1 (27) Model 2 (21) Model 3 (27) Model 4 (21) Model 5(27) Model 6 (21) Model 7 (27) Model 8 (21) B Std E B Std E B Std E B Std E B Std E B Std E B Std E B Std E Gross Domestic Product (PPS).8.9 -.28.19 -.17.19 -.11.17 -.34.21 -.32.2.3.27.18.31 Expenditure on LMP.59 1.58 4.84** 2.2.47 2.99 1.4 1.8-3.1 3.42 -.63 2.24 -.91 4.5 -.24 3.69 Working hours 1. *.49 2.16**.93.94 1.4 2.21 1.55 Share of temporary employment.23.14 2.16.28.9.4 -.6.48 Tax wedge.16.11 -.2.24.21.25.6.23.79 **.32.92 **.39 Dispersion of regional GDP.9.14 -..24.37.3.37.23.28.35.575 *.29.71.39.55.4 At least upper secondary education -.1*.6.4.11 -.2.13 -.7.11 -.7.14 -.16.13 -.56***.16 -.73***.19 Constant -3.28 21.14-54.45 4.14 13.89 16.54 21.86 14.41 34.85 * 17.58 31.61 * 15.48-18.38 6.27-5.57 67.21 Adjusted R2.29.2.2 -.2.17.26.36.43 F-statistsics 2.66 ** 2.21 * 1.12.91 2.43 * 3.41** 3.22 ** 3.96*** P-.4.1.4.5.7.2.2.1 Number of observations 29 29 29 29 29 29 29 29 Note : *** significant at < 1% level (p<.1), ** significant at < 5% level (p<.5), * significant at < 1% level (p<.1) In contrast to the pattern observed for the total population, none of the explanatory variables is the significantly related to the youth AROPE rates. Regarding the effect on youth unemployment rate, working hours appear to have a statistically significant positive effects both in Model 1 (27) and 2 (21). Given major policy concerns about rising unemployment levels among youth, it is worth exploring this finding further. Moreover, in line with the pattern for total population, the Expenditure on labour market policies has a positive impact on unemployment rate. In other words, higher spending on LMPs is associated with higher unemployment rates. Also educational level appears to have marginally significant, negative effect (in Model 1). The dispersion of regional GDP appears to have a positive impact on long-term unemployment rate in Model 6 (21), which implies that stronger regional differentiation is associated with the higher long-term unemployment among youth. The reported R-squared is highest for the model where involuntary part-time employment is the dependent variable (43%). Both in Model 7 and 8, tax wedge and the educational level are significant predictors. There is a positive relationship between the tax wedge and the involuntary part-time employment, suggesting that higher tax on labour is associated with higher rates of involuntary part-time employment, whereas higher educational levels are associated with higher levels of involuntary part-time employment. 22 23

Chapter 3 Results 3.1.3 Older people Table 3 Ordinary least squares (OLS) multiply linear regression models of Labour market resilience for older people, in 29 European countries (27 and 21). Model variable 1. Unemployment rate 2. AROPE 3. Long term unemployment 4. Involountary part-time Model 1 (27) Model 2 (21) Model 3 (27) Model 4 (21) Model 5 (27) Model 6 (21) Model 7 (27) Model 8 (21) B Std E B Std E B Std E B Std E B Std E B Std E B Std E B Std E Gross Domestic Product (PPS) -.3.4 -.16.8 -.33.19 -.29.17 -.21.24 -.223.21 -.41 **.15 -.41 **.19 Expenditure on LMP.66.61 1.9.86 1.3 3.21.61 1.98 -.64 3.68 2.263 2.51 -.37 2.24 2.62 2.9 Working hours.5 1..64.83 1.77 1.26 2.62** 1.9 -.15 2.24 Share of temporary employment.8.6 -.7.13.1.28.9.26 -.36.35 -.7**.34.25.25 -.34.34 Tax wedge.3.5 -.4.11.12.28.6.21 Dispersion of regional GDP.7.6.1.11.67**.28.53**.22 At least upper secondary education.1.3.4.11 -.1.12 -.6.1 -.5.14 -.13.13 -.11.1 -.18.13 Constant -.61 3.74 5.51 7.5-2.98 42.94-7.76 36.1-27.12 56.41-52.9 49.5 36.85*** 9.46 4.4*** 12.81 Adjusted R2.3.6.4.42.2.35.19.1 F-statistsics 1.12 1.28 3.64 *** 3.85*** 2.43 * 3.97** 2.63* 1.78 P-.4.3.1.1.7.1.6.17 Number of observations 29 29 29 29 29 29 29 29 Note : *** significant at < 1% level (p<.1), ** significant at < 5% level (p<.5), * significant at < 1% level (p<.1) As illustrated in table 3, none of the explanatory variables is significantly related with the unemployment rate of older people. In line with the results for the total population the Dispersion of regional GDP appears to have significant positive effect on AROPE rates for older people. Furthermore, share of temporary employment is negatively associated with long-term unemployment rate among older people. Moreover, working hours have a positive effect on long term unemployment rate for older workers in Model 6 (21). Interestingly, working hours appeared to be also positively associated with unemployment rate for total population and youth. Finally, GDP (PPS) appears to have a significant negative effect on in-voluntary part-time employment of older workers. The reported R-squared is highest for the model where AROPE rate is the dependent variable (Model 3 = 4%, Model 4 = 42%). 24 25

Chapter 3 Results 3.1.4 Migrants Table 4 Ordinary least squares (OLS) multiply linear regression models of Labour market resilience for migrants (27 and 21). 5) 1. Unemployment rate 2. AROPE Model variable Model 1 (27) Model 2 (21) Model 3 (27) Model 4 (21) B Std E B Std E B Std E B Std E Gross Domestic Product (PPS) -.11**.5 -.2.18 -.29**.12 -.33*.7 Expenditure on LMP 2.2**.78 2.56 2.9 3.16** 1.27 1.27 1.85 Share of temporary employment.1.8.4.2 Tax wedge.13*.6.5.16.3.23 Dispersion of regional GDP.3.8.27.29.7.16 -.9.24 At least upper secondary education -.4.3 -..8 Constant 7.5 4.79 2.76 15.79 11.68 11.12 32.3** 9.11 Adjusted R2.45.2.2.14 Gross domestic product (PPS) appears to be a significant predictor for the AROPE rate (Model 3 and 4), but also for unemployment rates (Model 1). In both cases there is negative relationship between variables. These results partly support the general view on hyper cyclic nature of migrant employment (Hogarth et al., 29), suggesting that migrant unemployment rates increase in recession periods and decrease quickly in expansion times. One of the possible explanations of this trend can be that migrant workers often occupy more low-skilled, flexible and seasonal type of jobs in contrast to native workers for example.. Furthermore, Expenditure on labour market policies and tax wedge are positively associated with unemployment rate in Model 1 (27). The reported R-squared is highest for the Model 1 (45%). Generally, the results of the regression analyses seem to be affected by the time frame as well. In other words, the factors that appear to be significant before crisis might have no effect after the crisis. F-statistsics 4.85 *** 1.12 2.19* 2.1 P-..37.8.11 Number of observations 25 25 28 28 Note : *** significant at < 1% level (p<.1),** significant at < 5% level (p<.5), * significant at < 1% level (p<.1) 5) As a result of lack of data : Bulgaria Lithuania, Poland, Romania and Slovakia are excluded from the Model 1 and 2. Romania is excluded from Model 3 and 4. 26 27

Chapter 3 Results 3.2 s This section presents the evaluation of the changes in unstandardized residuals based on the results presented in the regression models discussed in the previous section. These s have been calculated for total population, youth, older people and migrants in 29 European countries. We focus on two key dependent variables: unemployment rate and at risk of poverty and social exclusion rate. The change has been calculated, based on the difference between the residuals before the crisis (27) and during/ after (21). Consequently, we present four ranking tables per dependent variable: unemployment rates and AROPE, and for each of the groups: total population, youth, older persons and migrants. Ranking no 1 shows the change in the actual s before and after the crisis. This has been calculated by subtracting from the actual for 21 the for 27. Rankings no 2 and 3 are based on the residual score for 27 and 21, respectively. Ranking no 4 orders countries based on the changes in residuals. This has been calculated by subtracting from the score for 21 the for 27. All of the rankings are sorted starting from the smallest to the highest. The rankings based on the residuals provide an alternative for the understanding of countries labour market resilience. However, these rankings rely on regression models and not the actual s and therefore, should be treated rather as a supplement to the rankings based on the actual changes in unemployment rate and AROPE. 3.2.1 population Apart from Germany and Austria, all European countries experienced increases in their unemployment rates for the total population in the period between 27 and 21. The effect in terms of increases in AROPE rates is not as pronounced. In fact, most of the countries managed to decrease their AROPE rates. Based on the rankings presented in Table 5 and 6, we can argue that in terms of mitigating the negative impact for the total population, in the period of between 27 and 21, there were several countries which appear to be particularly resilient. Thus, among the countries that either managed to decrease their unemployment rates or maintain the pre-crisis level were: Germany (-1,6 %), Austria, Poland, Malta, Belgium, Romania, and the Netherlands. These findings are mostly confirmed in the ranking of residual change. However, based on changes in residuals Bulgaria seems to perform the best, whereas Germany for instance is on the 7th position. Furthermore, among the countries that managed to considerably decrease their AROPE rates are: Bulgaria, Poland and Romania. The changes in residuals also support this conclusion. Among the countries that appear to be less resilient in terms of unemployment of total population are: Lithuania, Latvia, Estonia and Spain (above 1 % increase in unemployment rates). This is also confirmed by the scores in residuals, where these countries are at the bottom of the ranking. Regarding the AROPE rates, among all the countries only two countries recorded relatively high increases (above 5%): Lithuania and Ireland. However, residual change suggests that Ireland performed better than expected by our models. The differences among rankings based on 27 and 21, suggest that the crisis had a large impact on unemployment in several countries: e.g. Estonia, Lithuania. Furthermore, there are also small differences between rankings based on the actual s and the change in s (e.g. Luxemburg). Table 5 Unemployment rates (total population): 1. in actual s (21-27), 2. s (27); 3. s (21); 4. in residual s (RE 21 RE 27) Country 1 2 3 4 UR Country 27 Country 21 Country Res/UR Germany -1.6 Cyprus -2.85 Austria -4.62 Bulgaria -3.69 Austria Austria -2.18 Bulgaria -3.98 Malta -3.48 Poland.1 Slovenia -1.68 Czech Rep -3.9 UK -3.27 Luxembourg.4 Denmark -1.51 Cyprus -3.28 Slovakia -3.25 Malta.4 Lithuania -1.31 Belgium -3.24 Finland -3.5 Belgium.8 Czech Rep -1.12 Romania -3.16 Belgium -3.3 Romania.9 Estonia -.94 Denmark -3. Germany -2.84 Netherlands.9 Norway -.89 Switzerland -2.77 Czech Rep -2.78 Norway 1.1 Italy -.65 Finland -2.76 Romania -2.52 Switzerland 1.1 Romania -.64 Malta -2.7 Austria -2.43 France 1.3 Switzerland -.62 UK -2.26 Poland -2.36 Finland 1.5 Spain -.6 Slovenia -2.15 Switzerland -2.35 Czech Rep 2 Netherlands -.48 Poland -1.49 Denmark -1.49 Italy 2.3 Latvia -.42 Netherlands -.64 Hungary -.87 Slovenia 2.4 Bulgaria -.28 Hungary -.59 France -.84 Cyprus 2.4 Ireland -.27 Ireland -.39 Slovenia -.47 Sweden 2.5 Belgium -.21 Germany -.32 Cyprus -.43 UK 2.5 Hungary.28 Norway. Netherlands -.16 Portugal 3.1 Finland.29 France.1 Ireland -.12 Slovakia 3.3 Luxembourg.36 Italy.9 Greece.3 Bulgaria 3.4 Sweden.4 Greece 1.48 Norway.89 Denmark 3.7 Portugal.52 Slovakia 1.51 Sweden 1.19 Hungary 3.8 Malta.78 Sweden 1.59 Italy 1.55 Greece 4.3 Poland.87 Portugal 2.34 Portugal 1.81 Ireland 9.2 France.95 Luxembourg 3.25 Luxembourg 2.89 Spain 11.8 UK 1. Estonia 4.61 Estonia 5.54 Estonia 12.3 Greece 1.19 Spain 6.6 Spain 7.2 Latvia 13.3 Germany 2.52 Lithuania 6.83 Lithuania 8.14 Lithuania 14.2 Slovakia 4.76 Latvia 7.76 Latvia 8.19 Source : Actual s based on data from Eurostat, LFS. s based on own calculations. 28 29

Chapter 3 Results Table 6 At risk of poverty and social exclusion rate (total population) : 1. in actual s (21-27); 2. s (27) ; 3. s (21) ; 4. in residual s (RE 21 RE 27) Country 1 2 3 4 AROPE Country 27 Country 21 Country Res/ AROPE Bulgaria -11.8 Estonia -1.44 Estonia -9.85 Bulgaria -12.91 Poland -6.5 Cyprus -8.36 Czech Rep -8.43 Romania -3.4 Romania -5 Czech Rep -7.65 Slovakia -7.6 Ireland -3.1 Cyprus -3 Slovakia -7.5 Malta -6.57 Netherlands -1.95 Belgium -2 Malta -6.89 Cyprus -6.45 Finland -1.83 Italy -1.7 Slovenia -4.86 Hungary -3.74 Greece -1.45 Greece -1.7 Hungary -4.84 France -3.63 Norway -1.41 Netherlands -1.5 Portugal -4.21 Belgium -2.52 Belgium -1.38 Austria -1.4 France -3.81 Portugal -2.23 Poland -1.27 Germany -1.2 Germany -2.54 Norway -1.55 UK -.89 Czech Rep -1 Austria -1.79 Slovenia -1.53 Denmark -.88 Norway -.9 Belgium -1.14 Germany -1.42 Czech Rep -.77 Finland -.8 Latvia -1.7 Austria -1.31 Switzerland -.1 Slovakia -.7 Sweden -.73 Finland -1.9 France.18 Estonia -.6 Switzerland -.48 Switzerland -.58 Malta.33 Hungary -.4 Norway -.13 UK.38 Luxembourg.43 France -.2 Italy.24 Poland.68 Slovakia.44 Portugal -.1 Finland.74 Denmark.75 Austria.48 UK.3 UK 1.27 Netherlands.79 Estonia.58 Switzerland.4 Denmark 1.64 Sweden.89 Hungary 1.1 Malta.9 Poland 1.96 Ireland 1.48 Germany 1.12 Slovenia.9 Spain 2.15 Italy 1.9 Sweden 1.62 Latvia 1 Lithuania 2.5 Greece 3.89 Italy 1.66 Luxembourg 1.3 Netherlands 2.74 Latvia 4.46 Cyprus 1.91 Denmark 1.4 Ireland 4.58 Spain 5.49 Portugal 1.98 Spain 1.6 Greece 5.33 Luxembourg 6.19 Slovenia 3.32 Sweden 2.1 Luxembourg 5.76 Romania 9.49 Spain 3.34 Lithuania 5 Romania 12.9 Lithuania 9.82 Latvia 5.54 Ireland 5.5 Bulgaria 24.65 Bulgaria 11.74 Lithuania 7.32 Source : Actual s based on data from Eurostat, EU-SILC. s based on own calculations. 3.2.2 Youth Generally speaking youth unemployment rates, as well as the AROPE rates increased on average twice as strongly, compared to older workers between 27 and 21. Based on the rankings presented in Table 7 and 8, we can argue that in terms of mitigating the negative impact of the crisis for the youth, in the period between 27 and 21, there were several countries which proved to be particularly resilient. Among the countries that either managed to decrease their unemployment rates or maintained the pre-crisis level are: Germany (-2 %) and Malta (-.8%), countries with almost no change: Austria, Luxemburg and Switzerland. The ranking based on the change in residuals suggest that the following countries performed better than predicted by our model: Malta, Bulgaria, Austria, Belgium and Switzerland. Interestingly, according to this ranking Germany is on the 8th position. Regarding the AROPE rates for youth both the changes in actual s and in the residuals suggest that in terms of At risk of poverty and Social exclusion rate for youth the most resilient were the following countries: Bulgaria, Poland, Norway, Belgium, Romania, Switzerland, Czech Republic. Among the countries that proved to be less resilient in terms of youth unemployment are: countries with more than 1% increase in unemployment rates (Greece, Slovakia, Ireland), countries with more than 2% increase (Estonia, Spain, Latvia, Lithuania, which are also at the bottom of the ranking based on the residual change). Regarding the youth AROPE rates, among the less resilient are: Netherlands, Malta, Latvia, Estonia, Spain, Lithuania, Ireland countries (increase above 5 % in actual s). This is also supported by the score ranking in residual changes. One of the striking observation is Greece which according to the ranking based on change in actual unemployment performs relatively poor, however, based on the residual change ranking it scores quite high, suggesting that it could have performed much worse. In contrast Luxemburg which is relatively high in the ranking based on actual s, scores low in changes in residual s, suggesting that it could have performed better according to our model. 3 31

Chapter 3 Results Table 7 Unemployment rates (youth): 1. in actual s (21-27); 2. s (27); 3. s (21) ; 4. in residual s (RE 21 RE 27) Table 8 At risk of poverty and social exclusion rate (youth) : 1. in actual s (21-27); 2. s (27) ; 3. s (21) ; 4. in residual s (RE 21 RE 27) Country 1 2 3 4 UR Country 27 Country 21 Country Res/UR Germany -2 Austria -6.43 Austria -13.39 Malta -8.44 Malta -.8 Slovenia -6.36 Malta -9.25 Bulgaria -7.59 Austria.1 Czech Rep -4.19 Czech Rep -8.33 Austria -6.96 Luxembourg.2 Lithuania -4.1 Bulgaria -8.24 Belgium -6.43 Switzerland.6 Cyprus -3.87 Slovenia -8.17 Switzerland -6.19 Netherlands 1.7 Spain -3.67 Germany -7.23 Romania -6.12 Norway 2 Latvia -2.99 Denmark -7.12 Finland -5.75 Romania 2 Denmark -2.96 Cyprus -6.56 Germany -5.56 Poland 2.1 Norway -2.71 Switzerland -4.5 UK -5.9 Belgium 3.6 Estonia -1.8 Belgium -4.5 Poland -4.24 France 3.8 Ireland -1.75 Finland -3.41 Denmark -4.16 Slovenia 4.6 Germany -1.67 Romania -2.15 Czech Rep -4.14 Finland 4.9 Portugal -.99 Poland -1.7 Cyprus -2.69 UK 5.3 Netherlands -.98 UK -1.56 France -2.39 Sweden 5.6 Malta -.81 Netherlands -.63 Slovenia -1.81 Cyprus 6.4 Bulgaria -.65 Norway -.18 Slovakia -1.67 Denmark 6.5 Luxembourg.78 Ireland.17 Hungary -1.52 Portugal 7.3 Switzerland 1.69 France.18 Greece -.22 Italy 7.5 Hungary 1.95 Hungary.42 Netherlands.35 Czech Rep 7.6 Italy 2.34 Portugal 3.44 Ireland 1.92 Bulgaria 7.7 Finland 2.35 Greece 4.32 Sweden 1.96 Hungary 8.5 Belgium 2.38 Luxembourg 5.51 Norway 2.53 Greece 1 Poland 2.54 Slovakia 5.52 Portugal 4.43 Slovakia 13.3 France 2.57 Sweden 7.4 Luxembourg 4.74 Ireland 18.5 UK 3.54 Estonia 7.46 Italy 5.63 Estonia 22.8 Romania 3.97 Italy 7.97 Estonia 9.27 Spain 23.4 Greece 4.54 Spain 1.37 Spain 14.4 Latvia 25.3 Sweden 5.44 Lithuania 11.71 Latvia 15.8 Lithuania 28.5 Slovakia 7.19 Latvia 12.9 Lithuania 15.72 Source : Actual s based on data from Eurostat, LFS. s based on own calculations Country 1 2 3 4 AROPE Country 27 Country 21 Country Bulgaria -12.5 Estonia -12.36 Cyprus -11.57 Bulgaria -13.92 Poland -8.8 Slovenia -1.24 Switzerland -9.7 Poland -7.4 Norway -2.6 Austria -9.99 Slovenia -9.3 Norway -5.43 Belgium -2.1 Malta -9.73 Belgium -9.1 Switzerland -5.39 Romania -1.7 Latvia -8.8 Malta -9.1 Cyprus -4.7 Switzerland -1.5 Czech Rep -7.87 Austria -8.54 Belgium -2.62 Czech Rep -1.5 Cyprus -6.87 Czech Rep -8.43 Romania -1.38 Germany -1 Slovakia -6.66 Slovakia -5.68 Czech Rep -.56 Cyprus -.3 Belgium -6.4 Estonia -5.66 Denmark -.54 Slovakia.2 Switzerland -4.31 Latvia -5.6 Portugal -.17 Italy.3 Germany -4.11 Germany -4.28 Germany -.17 Slovenia.9 Portugal -2.52 Portugal -2.69 Italy.28 Luxembourg 1 France -2.16 France -1.2 Finland.29 Sweden 1.3 Lithuania -1.6 UK.31 Malta.72 Austria 1.4 UK -1.25 Poland 1.48 Sweden.9 Hungary 1.6 Spain -.42 Finland 1.5 Netherlands.93 France 2.1 Hungary.33 Netherlands 1.76 France.96 Denmark 2.4 Netherlands.83 Spain 1.86 Slovakia.97 Finland 2.4 Finland 1.21 Italy 2.37 Slovenia 1.21 Portugal 3.4 Italy 2.9 Hungary 2.61 Austria 1.45 UK 4.7 Luxembourg 2.6 Ireland 4.32 Ireland 1.55 Greece 4.8 Ireland 2.77 Luxembourg 5.8 UK 1.55 Netherlands 5 Sweden 4.8 Sweden 5.7 Spain 2.28 Malta 5.4 Greece 8.6 Norway 7.45 Hungary 2.28 Latvia 5.8 Poland 8.88 Denmark 8.82 Luxembourg 2.48 Estonia 6.1 Denmark 9.37 Lithuania 9.87 Greece 2.52 Spain 7.1 Romania 12.15 Greece 1.58 Latvia 3.74 Lithuania 9 Norway 12.87 Romania 1.77 Estonia 6.7 Ireland 11 Bulgaria 29.31 Bulgaria 15.39 Lithuania 11.47 Source : Actual s based on data from Eurostat, EU-SILC. s based on own calculations Res/ AROPE 32 33