Subsidized employment for young long-term unemployed workers - an evaluation

Transcription

1 Subsidized employment for young long-term unemployed workers - an evaluation Bart Cockx Christian Göbel Preliminary version Abstract In this paper we estimate the impact of subsidized employment for young long-term unemployed workers in Belgium on the transition rate from employment to non-employment. We account for selective participation on the basis of a multivariate duration model with correlated unobserved heterogeneity. We find that the policy decreases the transition rate from employment to non-employment in the first year of participation. There is no significant effect on the transition rates in the second year. After participation, the transition rates out of employment for men (but not for women) increases. The median impact on employment duration is positive. Keywords: evaluation, subsidized employment, MPH, simulation JEL-Classification: J64, J68, C41, H43 We acknowledge the Politique scientifique fédérale (Grant SSTC SO/10/039 "Cohésion sociale") for financial support. We also acknowledge Bernard Masuy for providing assistance with preparing the data and Bruno van der Linden for helpful comments. Department of Economics and Institut de Recherches Economiques et Sociales (IRES), Université Catholique de Louvain, Louvain-la-Neuve, Belgium and IZA, Bonn. cock@ires.ucl.ac.be Department of Economics and Institut de Recherches Economiques et Sociales (IRES), Université Catholique de Louvain. Address: IRES, Université Catholique de Louvain, Place Montesquieu 3, B-1348 Louvain-la-Neuve, Belgium. goebel@ires.ucl.ac.be

2 1 Introduction High unemployment rates persist for more then twenty-five years in the European Union (EU) and for a long time politicians worry about how to fight unemployment. At the latest from 1997 on employment questions are seen as important topic and explicit targets for unemployment rates in Europe have been adopted. To reach these targets the EU is in favor of active labor market policies (European Union (1997)). Active labor market measures include measures where member states of the EU should consider how reducing the cost of hiring less skilled workers through measure such as targeted reductions in non-wage labor costs. In this sense, subsidies in the form of reduction of social insurance contributions for long term unemployed people, considered in this paper, are an active labor market policy like advocated by the EU. Since active labor measures are often costly to the welfare states (OECD (2003)), there is a need for evaluation to asses and improve the design of existing measures. In this paper we estimate the impact of participation in subsidized employment for young long-term unemployed workers in Belgium on the transition rate from employment to non-employment. We find that the policy decreases the transition rate from employment to non-employment in the first year of participation. There is no significant effect on the transition rates in the second year. After participation, the transition rates out of employment for men (but not for women) increases. The median impact on employment duration is positive. The paper is structured like follows. Section 2 provides a description of the policymeasure we have evaluated. Section 3 gives a short description of the used database. In section 4 we will develop the econometric model which allows an evaluation of the reduction of social insurance contributions. Section 5 provides the key results of the estimation. Section 6 concludes. 2 Description of the policy In this evaluation we analyse the effect of the Belgian Employment Plan 1 which provides, under certain conditions, reduction in social insurance contributions in the case of employment. Since we limit our analysis to young long-term unemployed people who are unemployed for at least 12 months we restrict the description to the policy-modalities relevant for this group. 2 Employers who want to engage workers under the terms of the measure have to meet the following conditions: Being subject to social insurance contributions with exception of 1 This is a literal interpretation of the French name plan d embauche 2 Note that the Employment Plan provides different reduction schemes for other target groups. 1

3 the state, the language-communities, the regions, the public agencies and institutions of the former ones. Employers who benefit from a reduction of the employment plan are not allowed to benefit from other measure for the same individual. Entitled young employees have to be job-seekers and have to be unemployed with full unemployment benefits for at least 12 months without interruption. A period of unemployment with full unemployment benefits is considered to be without interruption when the period where the unemployed individual do not receive unemployment benefits is not longer than four months. The following periods are also counted like periods where unemployment benefits were paid: the waiting period which applies after leaving the school system, temporary work, certain subsidized employments, periods of diseases with payment of unemployment benefits, periods of unemployment which are covered by holidays (pécule de vacances), periods in prison within a unemployment period etc.. The employee or the employer have to let issue a employment card which proves that the employer can be engaged within the employment plan. This card has to be given to the authorities within sixty days after the first employment day. The social insurance contributions of the employer is reduced if he engages an entitled unemployed worker. The reduction is 75% of the base amount of social insurance contributions in the quarter of engagement and the 4 following quarters; and 50% of the base amount from the fifth to the eighth quarter. The base amount of social insurance contributions is between 32% and 34% of the gross wage. The employees have to be engaged at least for a half-time job. 3 Description of the database For the purpose of evaluation we used a dataset provided by the Belgian Crossroads Bank for Social Security. The database combines administrative data from different institutions of the Belgian social security system. 3 Since the database contains information about unemployment (ONEM), employment (ONSS and ONSS-APL), independence (INASTI) and inactivity (identified by absence in the other databases) we are able to determine individual labor market trajectories on a quarterly basis over the observation period. The original databases contain information about individuals who are years old and who received in 1998 for at least one day either waiting benefits 4 or transition 3 Our database contains information from the ONSS (National Office for Social Security), ONSS-APL (Local Agencies of the ONSS), ONEM (National Employment Office), ONAFTS (National office for family benefits for salaried persons), INASTI (Social insurance institute for independent workers) and the INS (National Institute of Statistics). 4 In Belgium individuals who are unemployed after finishing schooling have a waiting period before they are entitled to receive full unemployment benefits. For the age of the individuals, analyzed in this paper, the waiting period is 9 months. Waiting benefits, which are lower than unemployment benefits, are paid in this period, instead. 2

4 benefits 5. The observation period is from We are interested in a causal analysis of reduction of social insurance contributions by the means of duration analysis. More precisely we want to answer the following question: Is there a causal effect of participation at the Employment plan on the transition rate from employment to non-employment? Since we do not observe what happened before 1998 we might have an initial condition problem which can appear if we ignore the individual labor market histories, which are known to have a strong impact on the probability of leaving employment. Ignoring the labor market history may lead to spurious results in our analysis. To minimize the problem we have chosen a subsample which contains individuals with a homogeneous labor market history. For our analysis we used a subsample of individuals which are between 18 and 26 years old when they started there waiting period, who received unemployment benefits and who are without work experience before their selection in The resulting database contains information about individuals. Note that of these individuals have been the first time in full-unemployment in In the following full-unemployment is defined as receiving full unemployment benefits after the waiting period. For the other individuals, who enter full unemployment before 1998, we cannot be sure that they have homogeneous labor market histories, since for the time before 1998 we can only control for interruptions of unemployment when they are longer than 3 months. To be able to restrict our analysis to individuals without working experience, we only kept the individuals who are the first time in full unemployment in Note that for the estimation we deleted further observations either because of missing variables or because of contradictory values of different variables. Finally we kept a database with individuals, women and men. For a description of the explanatory variables see table 2 in the appendix. We concentrate the descriptive analysis on the subgroup of individuals who participate in subsidized employment and its control group, the individuals who have been in normal employment. Normal employment is defined as employment without labor market policy intervention. Note that the number of individuals which are concerned by the subsidy is relatively small compared to the individuals who have a transition to normal employment. We have 257 (246) individuals who participate in subsidized employment to 3152 (2773) control individuals for the women (men). Generally we can say that the statistics of the participants and the controls are close to each other. This may be due to the fact that the database contains already a specific subgroup of the population. We start with the individual characteristics of the individuals. The average age at the end 5 Allocation de transition or transition benefits are sometimes paid to young people who have a period of part time school within the period of compulsory schooling (15-18 years), that means people who leave full-time schooling before the age of 18 years. 3

5 of 1997 is around 20.5 years for all subgroups. The women (men) who have a transition to normal employment contain a slightly higher fraction of Belgians 91.4% (89.4%) than the subgroup of participants 87.9% (88.6%). We observe a bigger proportion of college degrees among women who have a transition to normal employment, 27%, compared to the participants where only 20.6% have a college degree. Whereas for the men the picture with respect to the school degree is ambiguous. The month of entrance to full unemployment corresponds to the month where the waiting period is finished and full unemployment benefits are paid. Note that for individuals who finish school education in June the waiting period starts the first August, if they are inscribed as job-seeking. These individuals enter full-unemployment in April of the year after the inscription. We can see that an important fraction (around 30%) of the individuals enter full unemployment in April. Since a lot of students finish their schooleducation in September or October there is also an important fraction of individuals who enter full-unemployment in May and June. 70% of the women and 66% of the men enter full-unemployment from April to June. There are two main explanations why individuals enter in other months - either they are not inscribed immediately as job-seekers or they left the education-system uncommonly in the middle of a school year 6. Our database contains also information about the characteristics of the household and relationship to the head of the household. The largest fraction of individuals is reported to be child of the head of the household. More than 80% (85%) of the women (men) are a child of the head. The remaining individuals are either oneself head of the household, its wife or have no family relationship at all to the head of the household. Since there are very few men who are the husband of the head of the household, these individuals have been removed from the database. The smallest fraction has an other family relationship to the head of the household than already mentioned. The database contains information about the number of individuals in different age classes in the household. Detailed statistics about these variables can again be found in table 2. On average there are 2.7 (2.8) persons in the household for women (men). Two variables concern the labor market conditions. First the local unemployment rate at the first quarter of the observation period. The unemployment rate is on the level of unemployment offices and allow to control for local labor market conditions. With 25.1% for the unemployed workers who have a transition to normal employment and 27.2% for those with a transition to subsidized employment, the unemployment rates for women are considerably higher than for men who have 17.6% and 19.5% for the respective groups. Second we have information about the region where the individual comes from. Note that around two third of the individuals came form the southern, French speaking (the Walloon) region of Belgium. Remarkable is that the individuals from the northern, Flemish 6 Note that in Belgium schools and colleges are organized in school years respectively academic years. 4

6 region who participate in subsidized employment are underrepresented compared to the Walloon individuals. This may be due to the fact that a relative higher fraction of the Flemish have a transition to normal employment in the observation period. We also provide some descriptive statistics concerning the labor market trajectories of the individuals. In average we observe 2.0 (2.2) spell for the women (men). Note that for the subgroups who have a transition to normal employment or to subsidized employment we observe more than 3 labor market spells which is considerably more than in the whole sample. This indicates that the subgroups of interest have a higher mobility on the labor market than the average of all observed individuals in the sample. The non-censored unemployment spells have a average duration of 9.8 (9.9) months for the women (men). The individuals who have a transition to normal employment stay slightly shorter in unemployment than the individuals who participate in the activating policy. The average non-censored employment duration is 7.3 months (women and men) for those who don t participate and 8.1 (9.2) months for the women (men) who participate at subsidized employment. In the description of the professional trajectory we limit ourselves to the transitions which are explicitly modeled and analysed in this paper. In our econometric model we allow for three transitions: Two competing transitions out of unemployment and one out of employment. It is remarkable that 60% of the male participants have a transition to full employment whereas only 49% of the non-participant have a transition to full employment. For the women only around 35% have a transition to to full employment independent of participation. This indicates that policy participation give access to full employment to a higher fraction of the men, but has no effect on the fraction of full time employment for the women. Remind that at the beginning of the analysed employment trajectories all individuals enter full unemployment after a waiting period of 9 months. For the first spell we analyse the transitions from unemployment to normal employment and from unemployment to subsidized employment. The unemployment spells of individuals who have other transitions are treated like right censored in the time period where this transitions occurs. Figure 1.1 shows the non-parametric survival rates in unemployment over time for men and women. We estimated the survival functions by estimating the grouped duration model of section 4 without explanatory variables. Since we have specified a piecewise constant hazard-rate for each time period the estimates result is an non-parametric Kaplan-Meier estimate of the duration distribution (Lancaster (1990)) 7. Figure 1.1 shows that the survival rate in unemployment is always higher for women than for men. As a consequence the median duration is longer for women: 16 months for the women compared to 13 months for men. Since for the first spell we model only transitions to (subsidized or normal) employment and all other transitions are treated as right censored, the second spell is necessarily an em- 7 We use the unique relation between the estimated hazard rate and the distribution of the durations to get the non-parametric survival rate. 5

7 Figure 1: Nonparametric estimates of the survival rates 1 women men 0.8 S(t u ) month 1.1: Unemployment 1 women men 0.8 S(t e ) month 1.2: Employment 6

8 Figure 2: Example for a labor market trajectory of an individual 1.SPELL: state normal employment with subsidies unemployment 2.SPELL: state t u years non-employment employment 1 2 years ployment spell. Figure 1.2 contains the survival rate in employment for the subpopulation of individuals who have a transition to employment. Eye-catching is that the survival rate for women and men is almost equal. The median-duration in employment is 11 months for both sexes. 4 Econometric Model The purpose of this paper is the evaluation of subsidized employment. More precisely we are interested in the effect of subsidized employment on the employment duration of the participants, where we consider the subsidized period already as employment. Note that our policy is not about providing a temporary period of employment but about reintroducing the unemployed worker into employment. Once individuals have found a subsidized job they are considered to be reinserted and assistance is not longer provided by the unemployment offices. 8 This approach is different from Abbring et al. (2000), van Ours (2002) and Lalive et al. (2002) who consider the time in participation as unemployment and who analyses the effect of subsidies on the transition to employment after participation. Figure 2 gives an example for a possible labor market trajectory of an individual. Like 8 The fading out of the subsidies, even though rough (in two steps), give further evidence to this point of view. 7

9 all individuals in our database the individual enters full unemployment at the start of his first spell. At a certain time t u the first spell ends and the individual in the example has a transition out of unemployment to employment. Note that he can either have a transition to normal employment or a transition to subsidized employment. The individual in our example enters the second (the employment) spell in the quarter where he left unemployment. After a certain duration he leaves employment to-wards non-employment. We ask how the transition to subsidized employment in the preceding spell influences the transition rates out of employment. We analyze the effect in different time intervals: the first year of participation, the second year of participation and afterwards. This is done by a time-varying dummy structure. All transitions which are not described here are treated as right censored, that means like individuals that leave the database before a transition occurs 9. We also take into account that some individual stay in a certain state longer than observed, by modelling explicit that the transition occurs beyond the observation period for those individuals. To get the causal effect of program participation we analyzed the differences in the transition rate of participants compared to non-participants. In such an approach we have to take a possible selection into account, we have to control for differences in observables and unobservable characteristics (Heckman et al. (1999), Blundell and Costa-Dias (2002)). To correct for selection in observables we introduce a set of explanatory variables which contain information about individual characteristics, the household of the individual and the local labor market. There may also be important variables which are not in our database. Neglecting such unobserved characteristics can lead to a number of problems. It is well known (Lancaster (1990)) that neglecting unobserved heterogeneity lead to inconsistent estimates of the parameters of the conditional hazards. More serious, in our case, selection in unobservables may lead to a serious bias of the estimates of the causal effects. To correct for selection in unobservables we follow a approach proposed in Abbring and van den Berg (2003) where it is shown that the causal effect of treatment is non-parametrically identified under certain conditions. The main idea to correct for selection in unobservables is to allow for dependencies of unobserved heterogeneity terms of different transitions. Like mentioned above our data contains information about the inflow into full unemployment on a monthly basis. The labor-market states in contrary are only available for each quarter. In our econometric model we have explicitly taken the monthly inflow into account. Note that by doing so we are able to identify the transition rates up to a month. Ignoring this information would lead to estimates which which are less exact, since we would not use all available information about the duration. Generally one has to take the initial condition problem into account which appears since 9 This included transitions to other policies than the analyzed one and transitions out of the labor market (i.e. inactivity, death). 8

10 the labor market history of an individual before the start of the observation period is not know. In our case we have young individual without any working experience. Therefore, we argue that the individuals are homogeneous with respect to their labor-market relevant history. Of course we control for differences in the the schooling level and the age of the individuals. The likelihood-function Note that our model is of the mixed proportional hazard (MPH) type (van den Berg (2000)). The equations for the different transition rates and the multivariate unobserved heterogeneity distribution are estimated simultaneously by the means of maximum likelihood. First we consider the state of unemployment. For this state we define two random durations: T up := the random duration until the subsidized employment T ue := the random duration until normal employment For the subsequent employment state we define: T en := the random duration until non-employment T pn := the random duration from program participation to non-employment We assume that all individual differences in the joint distribution T = (T up, T ue, T en, T pn ) can be characterized by explanatory variables X, V where X is observed and V is not. The joint distribution T X, V can be expressed in terms of the distributions (T up X = x, V ), (T ue X = x, V ), (T pn T up = t up, X = x, V ) (T en T up = t up, T ue = t ue, X = x, V ). The latter distributions are characterized by their hazard rates: θ up (t x, V ), θ ue (t x, V ), θ pn (t t up, x, V ), θ en (t t up, t ue, x, V ). Let V := (V p, V e, V n ) be a (3 1)-vector of unobserved covariates. Let T ue (V p, V n ) x, V e, implying that θ ue (t x, V ) = θ ue (t x, V e ) Let T up (V e, V n ) x, V p, implying that θ up (t x, V ) = θ up (t x, V p ) Let T pn (V e, V p ) t up, x, V n, implying that θ pn (t t up, x, V ) = θ pn (t t up, x, V n ) and T en (V e, V p ) t ue, x, V n, implying that θ en (t t ue, x, V ) = θ en (t t ue, x, V n ). Specifications of the hazards Like mentioned we specify a mixed proportional hazard model where explanatory variables and the unobservables shifts the estimated baseline hazard proportionally: K ue lnθ ue (t x, V e ) = k=1 lnθ up (t x, V p ) = α ue k ½ ue k + x β ue + V e (1) K up α up k ½up k=1 k + x β up + V p (2) 9

11 K pn lnθ pn (t t up x, V n ) = α pn k ½pn k + γpn lnt up + x β pn + V n (3) k=1 lnθ en (t t ue, t up, x, V n ) = K en αk en ½ en k=1 k + γ en lnt ue + x β en + V n (4) here K lm is the number of periods 10 for the duration in origin state l and destination state m and ½ lm k = 1 if t (t lm k 1, tlm k ] and ½lm k = 0 otherwise. Note that t (t lm k 1, tlm k ] means that the transition from one state to another occurs in the k-th period, this type of model is referred to as grouped duration model with a piecewise constant hazard specification, in the literature. To capture the effect of participation in subsidized employment we will impose the following restriction: α pn k α pn k α pn k = αen k + δ 11 for k 5 (first year of participation) = αen k + δ 12 for k > 6 (second year of participation) = αen k + δ 2 for the time periods after the participation Individual contributions to the likelihood Now, we consider the individual contributions to the likelihood function conditional on unobservables. We can distinguish the following cases: 1. Right censored at unemployment duration t k : k l 1 (V ) = P(T ue > t k, T up > t k ) = exp [θ ue (t j x, V e ) + θ up (t j x, V p )] (5) j=1 The individual likelihood contribution in the case of right censoring at unemployment duration t k is the survival rate in a competing risk model at the end of the time period t k. Note that we do not assume independence between the different transitions in the competing risk specification of the model. 2. Leaving for normal employment within (t k 1, t k+s ] (for s {0, 1, 2}) and right censored in regular employment after t l quarters: l 2 (V ) = P(t k 1 < T ue t k+s, T ue > t l ) = tk+s θ ue (t x, V e )exp t k 1 t 0 [θ ue (τ x, V e ) + θ up (τ x, V p )] dt (6) 10 Note that for origin state u we have monthly periods. For the other origin state we have quarterly periods. Note also that t lm 0 0 and t lm K = + 10

12 = = l exp θ en (t j t ue = t k+s, t pe = 0, x, V n ) j=1 s k+j 1 θ ue (t k+j x, V e ) exp (θ ue (t i x, V e ) + θ up (t i x, V p )) θ j=0 ue (t k+j x, V e ) + θ up (t k+j x, V e ) i=1 k+j exp (θ ue (t i x, V e ) + θ up (t i x, V p )) i=1 l exp θ en (t j t ue = t k+s, t pe = 0, x, V n ) (7) j=1 { } s θ ue (t k+j ) j=0 m {e,p} θ um(t k+j ) [S u(t k+j 1 ) S u (t k+j )] S e (t l t ue = t k+s, t pe = 0) (8) where in the last line (and the sequel) the conditioning on x and V is implicit. Again we allow for possible dependencies between the different transitions of the model. See Cockx (1997) for derivation from equation (6) to equation (7). Note that we consider an interval wider than 1 month: (t k 1, t k+s ] for s {0, 1, 2} rather tan (t k 1, t k ]. The reason is that during unemployment we observe the elapsed duration in months, but the transitions only with a precision up to a quarter. In general, s = 2. However, if an individual entered within the last month of the quarter and has already left unemployment by the end of th quarter, then s = 0. Similarly, s = 1 if one enters in the second month and leaves by the end of the first quarter. 3. Leaving for normal employment within (t k 1, t k+s ] (for s {0, 1, 2}) and leaving for non-employment within (t l 1, t l ]. l 3 (V ) = P(t k 1 < T ue t k+s, t l 1 < T en t l ) { } s θ ue (t k+j ) = m {e,p} θ um(t k+j ) [S u(t k+j 1 ) S u (t k+j )] [S e (t l 1 ) S e (t l )] (9) j=0 4. Leaving for program participation within (t k 1, t k+s ] and right censored during programme participation at t l l 4 (V ) = P(t k 1 < T up t k+s, T pe > t l, T pn > t l ) { } s θ up (t k+j ) = m {e,p} θ um(t k+j ) [S u(t k+j 1 ) S u (t k+j )] S p (t l t ue = t k+s ) (10) j=0 11

13 5. Leaving for programme participation within (t k 1, t k+s ] and leaving to non-employment from programme participation within (t l 1, t l ]. l 5 (V ) = P(t k 1 < T up t k+s, t l 1 < T pn t l ) { } s θ up (t k+j ) = m {e,p} θ um(t k+j ) [S u(t k+j 1 ) S u (t k+j )] [S p (t l 1 ) S p (t l )] (11) j=0 6. Leaving for programme participation within (t k 1, t k+s ], leaving to normal employment within (t l 1, t l ] and leaving for non-employment in (t m 1, t m ]. l 6 (V ) = P(t k 1 < T up t k+s, t l 1 < T pe t l, t m 1 < T en t m ) { } s θ up (t k+j ) = m {e,p} θ um(t k+j ) [S u(t k+j 1 ) S u (t k+j )] [S p (t l 1 ) S p (t l )] j=0 [S e (t m 1 ) S e (t m )] (12) The likelihood function in the case of unobserved heterogeneity: In order to obtain the unconditional likelihood contributions, we need to integrate out the unobserved heterogeneity terms. For individuals in the flow sample the likelihood contribution becomes: l m = for m {1, 2,...,6} where G(V ) is the joint distribution of the heterogeneity terms. v l m (V )dg(v ) (13) Specification of the heterogeneity distribution: A flexible 3 factor specification Suppose that v m (m {e, p, n}) can take two values v m1 and v m2 for each possible destination state m. This results in a joint heterogeneity distribution with 8 points of support and the following probabilities associated to them: P 111 = P(V e = v e1, V p = v p1, V n = v n1 ) = p 1 P 112 = P(V e = v e1, V p = v p1, V n = v n2 ) = p 2 P 121 = P(V e = v e1, V p = v p2, V n = v n1 ) = p 3 P 122 = P(V e = v e1, V p = v p2, V n = v n2 ) = p 4 P 211 = P(V e = v e2, V p = v p1, V n = v n1 ) = p 5 P 212 = P(V e = v e2, V p = v p1, V n = v n2 ) = p 6 P 221 = P(V e = v e2, V p = v p2, V n = v n1 ) = p 7 12

14 P 222 = P(V e = v e2, V p = v p2, V n = v n2 ) = p 8 (14) where we specify the probabilities p j by a multinomial logit model: for j = 1,...,7 and p j = 7 p 8 = 1 p j = j=1 exp(λ j ) i=1 exp(λ i ) i=1 exp(λ i ) (15) (16) and where we impose the following normalization: α ue 1 = α up 1 = α en 1 = 0 (17) One may argue that our distributional assumption for the heterogeneity distribution is too restrictive if, for example, the real distribution would be continuous. However Baker and Melino (2000) have shown in a Monte Carlo study that a specification with two mass-points performs well, even in the case of continuous distributions. For a nonparametric maximum likelihood estimator of the unobserved heterogeneity distribution see Heckman and Singer (1984). The corresponding likelihood components in (13) now become: 5 Estimation Results l m = P abc l m (v ea, v pb, v nc ) (13 ) a=1 b=1 c=1 In this paper we concentrate to the key results of our estimation. For the complete estimation results see table 3 and table 4 in the appendix. Note that the model without unobserved heterogeneity (UH) is rejected against the model with UH on the basis of a simple LR-test. Therefore we base the following summary of the results on the estimations with unobserved heterogeneity. However, for the men there seems to be no selection in unobservables. This is since we don t find unobserved heterogeneity in the transition out of employment. For the women we find a slight selection in unobservables. In general the parameters with and without UH are close to each other, in our estimates. For the effect of participation in the policy on the transition rate from employment to non-employment we can take a look at table 1, which summarizes the main results. We find that policy participation decreases the transition rate from employment to non-employment in the first year of participation by 36% (31%) for the men (women). There is no significant effect on the transition rates in the second year. After participation, the transition rates out 13

15 Table 1: Estimated effect of policy participation 1.year 2.year after men 36% +35% +60% women 31% 18% 2% Effect on the transition rate out of employment. Only the estimates which are market with a are significant different from zero. of employment for men (but not for women) increases significantly by 60%. There seems to be an important layoff effect after participation for the men. Two possible explanations for that are the existence of stigmatization of program participants or replacement effects for the men. Since a change of transition rate is a quite abstract measure we also simulated the effects of participation on the duration in employment. The effect on the duration was computed by taking the difference between the survival-time in employment for participants and their survival time in (hypothetical) non participation. Since the effect on the duration differs with the participation time and with the time in employment we have simulated the distribution of the effects. For the simulation of the distribution of the effects we proceeded like follows: 1. draw participation times for each individual 2. compute of the survival rate for participants, conditioned on his/her characteristics, given the estimated participation time and taking selection in unobservables into account 3. simulate the survival rate for participants for the hypothetical case of non-participation 4. compute the difference between the survival rate for participation and the survival rate in hypothetical non-participation at certain percentiles. The survivor rate in employment at the end of a time interval t k for an individual who had a transition to destination state d in the first spell is estimated by S ed (t k ) = v e v p v n P(v e, v p, v n ) S u (t u v e, v p ) h ud (t u v d ) S e (t e v e ) v e v p P(v e, v p ) S u (t u v e, v p ) h ud (t u v d ) where t u is the duration the individual stayed in unemployment (in the first spell), h ud is the transition rate from unemployment to the realized destination state d with d {e, p} [ and S u (t k v e, v p ) = exp ] k j=1 (exp(x β e + γ ej + v e ) + exp(x β p + γ pj + v p )) (18) 14

16 Figure 3: Distribution of the simulated effects - women fraction ( #obs < obs ) increase of employment duration caused by participation (in months) Note: The large positive effects is mainly due to the assumption of persisting effects. is the survival rate in unemployment at the end of the time-interval t k for a model with discrete heterogeneity distribution in a competing risk framework. For the distribution of the effects for women see figure 3. Since for the simulation of low percentiles we are sometimes beyond the observation period we cannot compute the effect at this percentile. To deal with this we have assumed that the effect after participation persists beyond the end of the observation period, which is a reasonable approximation for the quarters close to the end of the observation period. However, we should not place emphasis to the extremities of the distributions since their extreme natures are mainly due to the mentioned assumption of persisting effects. All simulated effects for the women are positive. The increase of the employment duration at the median position is 8.7 months for the women. For the distribution of the effects for men see figure 4. Here again we made the assumption of persisting effects, which explains partly the large negative effects. The increase of the employment duration at the median is 3.1 months. To summarize we can say that the median impact of participation in subsidized employment of young long-term unemployed people on the employment duration is positive. Remind, however, that we analyzed only the effects of participation in subsidized employment for the specific group of young long-term unemployed workers. 15

17 Figure 4: Distribution of the simulated effects - men fraction ( #obs < obs ) increase of employment duration caused by participation (in months) Note: Here the large negative effects are due to the assumption of persisting effects. 6 Conclusion In this paper we evaluated the effect of participation in subsidized employment for young long-term unemployed workers. The analyzed subsidy is in the form of a reduction of social insurance contributions which is decreasing in two steps over two years. After the description of the policy and our data, we applied duration models to estimate the effect of the subsidy on the duration in employment. To take possible selection into account we applied a mixed proportional hazard (MPH) model. We control for selection in observables by a large set of explanatory variables. The selection in unobservables is taken into account by allowing for correlation between unobserved heterogeneity terms of the transition to participation and the transition out of employment. To give a notion of its magnitude, we simulated the distribution of the treatment effects. In the simulation we took the multivariate distribution of the heterogeneity distribution into account. We find that the policy decreases the transition rate from employment to non-employment in the first year of participation. There is no significant effect on the transition rates in the second year. After participation, the transition rates out of employment for men (but not for women) increases. The median impact on employment duration is positive. 16

18 Table 2: Descriptive statistics of explanatory variables women men normal activation all normal activation all variables employm. employm. nobs individual characteristics: age at the end of ,8 20,6 20,4 20,7 20,7 20,5 Belgian 91,4% 87,9% 87,8% 89,4% 88,6% 89,2% EU not Belgian 4,6% 7,4% 5,4% 5,6% 7,3% 5,1% Not EU 4,0% 4,7% 6,7% 5,0% 4,1% 5,8% primary school 3,6% 5,1% 7,9% 9,1% 7,7% 12,0% secondary school - base 13,4% 14,0% 22,1% 23,3% 23,6% 27,5% secondary school - high 49,2% 53,3% 48,0% 45,2% 49,2% 42,4% college - non-university 19,3% 14,4% 12,4% 10,3% 10,6% 8,5% college - university 7,7% 6,2% 5,0% 5,8% 3,7% 4,4% other education 0,5% 0,8% 0,8% 0,7% 1,2% 0,9% unknown education 6,4% 6,2% 3,7% 5,7% 4,1% 4,3% month of entrance in full unemployment, January 1,9% 2,0% 2,1% 2,4% 1,2% 2,5% month of entrance in full unemployment, February 1,3% 0,8% 1,8% 2,7% 1,6% 2,3% month of entrance in full unemployment, March 2,8% 3,5% 3,1% 3,0% 2,4% 3,1% month of entrance in full unemployment, April 37,3% 35,8% 32,2% 28,4% 35,8% 27,2% month of entrance in full unemployment, May 12,5% 12,8% 11,5% 11,5% 11,8% 10,5% month of entrance in full unemployment, June 26,6% 23,7% 25,9% 29,0% 24,4% 28,8% month of entrance in full unemployment, July 7,1% 9,0% 7,7% 7,5% 7,3% 7,8% month of entrance in full unemployment, August 3,0% 4,3% 4,8% 4,4% 4,9% 4,9% month of entrance in full unemployment, September 2,2% 2,3% 3,5% 2,7% 3,3% 3,7% month of entrance in full unemployment, October 2,1% 3,9% 2,7% 3,3% 2,9% 3,7% month of entrance in full unemployment, November 1,5% 1,2% 2,3% 2,6% 0,8% 2,7% month of entrance in full unemployment, December 1,7% 0,8% 2,4% 2,6% 3,7% 3,0% characteristics of the household: head of the household 6,6% 8,2% 9,6% 7,0% 7,3% 8,3% husband/wife of the head o.t.h. 3,3% 3,9% 5,5% son/daughter of the head o.t.h. 80,4% 82,1% 72,5% 86,7% 87,8% 84,9% other family relationships to the head o.t.h. 1,7% 1,2% 2,3% 2,9% 2,4% 3,0% no family relationship to the head o.t.h. 8,0% 4,7% 10,1% 3,4% 2,4% 3,7% indicator (# of persons in the household [0-1) > 0 ) 2,9% 3,9% 6,0% 1,2% 1,2% 1,4% mean (# of persons in the household [0-1) 1,03 1 1,03 1,06 1,33 1,06 indicator (# of persons in the household [1-3) > 0 ) 2,3% 3,9% 4,7% 1,7% 1,2% 2,0% mean (# of persons in the household [1-3) 1,03 1 1,06 1,06 1 1,07 indicator (# of persons in the household [3-6) > 0 ) 3,6% 3,5% 4,7% 3,0% 1,6% 3,8% mean (# of persons in the household [3-6) 1,11 1 1,13 1,07 1,25 1,11 indicator (# of persons in the household [6-12) > 0 ) 11,0% 12,5% 12,8% 13,6% 10,6% 13,9% mean (# of persons in the household [6-12) 1,23 1,25 1,3 1,25 1,31 1,29 indicator (# of persons in the household [12-18) > 0 ) 24,6% 26,8% 26,1% 27,8% 29,7% 28,9% mean (# of persons in the household [12-18) 1,3 1,33 1,39 1,31 1,37 1,33 indicator (# of persons in the household [18-30) > 0 ) 48,8% 53,7% 48,5% 43,8% 47,2% 42,5% mean (# of persons in the household [18-30) 1,27 1,32 1,29 1,28 1,32 1,33 indicator (# of persons in the household [30-50) > 0 ) 62,8% 64,2% 59,1% 68,2% 69,5% 66,8% mean (# of persons in the household [30-50) 1,56 1,46 1,54 1,52 1,59 1,52 indicator (# of persons in the household [50-65) > 0 ) 36,8% 38,9% 32,5% 40,6% 35,8% 38,5% mean (# of persons in the household [50-65) 1,45 1,34 1,42 1,39 1,4 1,39 indicator (# of persons in the household [65-75) > 0 ) 3,3% 3,9% 3,5% 3,9% 5,7% 4,6% mean (# of persons in the household [65-75) 1,16 1 1,13 1,14 1,07 1,11 indicator (# of persons in the household 75+ years > 0 ) 1,8% 3,1% 1,6% 2,6% 1,2% 2,3% 17

19 Table 2: Descriptive statistics of explanatory variables (continued) women men normal activation all normal activation all variables employm. employm. mean (# of persons in the household 75+ years 1,12 1,38 1,09 1,14 1 1,08 # of persons in the household 2,7 2,9 2,7 2,8 2,9 2,8 labor market conditions: unemployment rate - 1. quarter ,1 27,2 26,4 17,6 19,5 18,3 Flemish region 30,9% 21,4% 24,7% 24,1% 12,6% 19,6% Walloon region 59,2% 66,5% 63,8% 64,8% 73,6% 67,7% Brussels region 9,9% 12,1% 11,4% 11,0% 13,8% 12,7% professional trajectory: # spells in total 3,4 3,0 2,0 3,3 3,1 2,2 # employment spells 1,5 1,4 0,7 1,5 1,4 0,8 # unemployment spells 1,8 1,6 1,3 1,8 1,7 1,4 unemployment unemployment duration in month (non-censored) 9,3 10,3 9,8 9,4 9,7 9,9 unemployment duration in month (censored) 15,9 14,9 non-censored 100,0% 100,0% 49,6% 100,0% 100,0% 57,5% employment indicator (employment duration > 0) 98,2% 99,2% 39,6% 98,4% 98,4% 46,9% employment duration in month (non-censored) 7,3 8,1 7,4 7,3 9,2 7,4 employment duration in months (censored) 22,0 20,0 20,8 21,4 20,3 20,5 non-censored 61,6% 51,4% 59,1% 60,1% 52,5% 58,1% full time employment 35,8% 35,4% 35,2% 48,6% 60,2% 48,6% Explanatory Note: Normal employment refers to the individuals who have a transition from unemployment to normal employment. Activation refers to the individuals who have a transition from unemployment to subsidized employment. All refers to all individuals at the beginning of the observation period. Note that we have not displayed the statistics for the remaining subgroups (individuals who have a transition to another labor-market-policy, individuals who stay in unemployment over the whole observation period) for lack of space

20 Table 3: Estimation results - women with UH no UH women b exp(b) p-value b exp(b) p-value unemployment to activating policy (PAE12) constant -5,010 0,007 0,000 age at the end of ,064 0,938 0,374-0,018 0,982 0,695 Belgian EU not Belgian 0,386 1,471 0,198 0,179 1,196 0,467 not EU -1,012 0,363 0,059-0,666 0,514 0,051 primary school -1,494 0,224 0,001-0,744 0,475 0,035 secondary school - base -0,960 0,383 0,002-0,684 0,505 0,001 secondary school - high college - non-university 0,525 1,691 0,081 0,491 1,634 0,021 college - university 0,526 1,692 0,208 0,790 2,204 0,010 other education -1,271 0,281 0,476-0,314 0,730 0,690 unknown education 1,246 3,475 0,005 1,805 6,082 0,000 head of the household 0,036 1,036 0,939-0,233 0,792 0,471 husband/wife of the head o.t.h. -0,968 0,380 0,133-0,400 0,671 0,324 son/daughter of the head o.t.h. other family relationships to the head o.t.h. -1,391 0,249 0,155-0,709 0,492 0,275 no family relationship to the head o.t.h. -1,396 0,248 0,002-0,775 0,461 0,021 # of persons in the household [0-1) -0,718 0,488 0,216-0,386 0,680 0,273 # of persons in the household [1-3) 1,430 4,179 0,014-0,087 0,916 0,812 # of persons in the household [3-30) 0,020 1,020 0,805 0,025 1,025 0,637 # of persons in the household [30+) 0,019 1,019 0,914-0,024 0,977 0,839 unemployment rate - 1. quarter ,000 1,000 0,986-0,002 0,999 0,908 Walloon region Flemish region -0,087 0,917 0,821 0,078 1,081 0,776 Brussels region 0,564 1,757 0,124 0,201 1,222 0,381 baseline hazard unemployment to activation: quarter 1 quarter 2 quarter 3-0,076 0,927 0,735-0,376 0,687 0,077 quarter 4-0,226 0,798 0,392-0,647 0,524 0,010 quarter 5 1,152 3,165 0,000 0,163 1,176 0,399 quarter 6,7,8,9,10,10-0,087 0,917 0,832-1,859 0,156 0,000 unemployment to normal employment constant -2,421 0,089 0,000 Belgian EU not Belgian -0,124 0,883 0,106-0,118 0,889 0,113 not EU -0,681 0,506 0,000-0,661 0,516 0,000 primary school -0,934 0,393 0,000-0,917 0,400 0,000 secondary school - base -0,624 0,536 0,000-0,607 0,545 0,000 secondary school - high college - non-university 0,489 1,631 0,000 0,469 1,599 0,000 college - university 0,816 2,260 0,000 0,765 2,149 0,000 other education -0,693 0,500 0,002-0,692 0,500 0,002 unknown education 0,976 2,653 0,000 0,912 2,490 0,000 age*college - non-university 0,106 1,112 0,000 0,090 1,094 0,000 age* all other educations -0,020 0,980 0,351-0,019 0,981 0,330 month of entrance 1,2,7,8,9,10,11,12-0,154 0,857 0,001-0,154 0,857 0,001 month of entrance 3,5,6 0,060 1,062 0,111 0,054 1,055 0,130 month of entrance 4 head of the household -0,094 0,911 0,260-0,101 0,904 0,209 19

21 Table 3: Estimation results - women (continued) with UH no UH women b exp(b) p-value b exp(b) p-value husband/wife of the head o.t.h. -0,305 0,737 0,004-0,314 0,731 0,002 son/daughter of the head o.t.h. other family relationships to the head o.t.h. -0,224 0,799 0,086-0,228 0,796 0,064 no family relationship to the head o.t.h. -0,021 0,979 0,771-0,047 0,954 0,485 # of persons in the household [0-3) -0,553 0,575 0,000-0,536 0,585 0,000 # of persons in the household [3-6) 0,068 1,070 0,398 0,077 1,080 0,325 # of persons in the household [6-18) -0,081 0,922 0,000-0,074 0,929 0,000 # of persons in the household [18-30) 0,029 1,030 0,190 0,016 1,016 0,458 # of persons in the household [30-50) 0,112 1,119 0,000 0,101 1,106 0,001 # of persons in the household [50+) 0,044 1,045 0,176 0,034 1,034 0,264 unemployment rate - 1. quarter ,023 0,978 0,000-0,022 0,978 0,000 Walloon region Flemish region 0,316 1,371 0,000 0,274 1,316 0,000 Brussels region 0,045 1,046 0,475 0,050 1,051 0,414 baseline hazard unemployment to normal employment: quarter 1 quarter 2-0,095 0,909 0,082-0,120 0,887 0,027 quarter 3-0,220 0,802 0,000-0,251 0,778 0,000 quarter 4,5-0,262 0,769 0,000-0,301 0,740 0,000 quarter 6,7-0,254 0,775 0,000-0,301 0,740 0,000 quarter 8,9,10,11-0,402 0,669 0,000-0,490 0,613 0,000 employment to non-employment constant -0,886 0,412 0,000 age at the end of ,037 0,964 0,022-0,037 0,964 0,022 Belgian EU not Belgian 0,139 1,149 0,210 0,140 1,150 0,208 not EU -0,054 0,947 0,664-0,053 0,948 0,668 primary school 0,616 1,852 0,000 0,616 1,852 0,000 secondary school - base 0,362 1,436 0,000 0,362 1,436 0,000 secondary school - high college - non-university -0,147 0,863 0,039-0,147 0,863 0,039 college - university -0,374 0,688 0,001-0,374 0,688 0,001 other education 0,320 1,378 0,287 0,321 1,379 0,287 unknown education -1,349 0,259 0,000-1,349 0,260 0,000 month of entrance 1,2,7,8,9,10,11,12-0,079 0,924 0,206-0,080 0,923 0,202 month of entrance 3,5,6-0,104 0,901 0,045-0,104 0,901 0,044 month of entrance 4 head of the household -0,018 0,982 0,878-0,018 0,983 0,883 husband/wife of the head o.t.h. -0,109 0,897 0,430-0,109 0,897 0,431 son/daughter of the head o.t.h. other family relationships to the head o.t.h. 0,091 1,095 0,605 0,091 1,095 0,603 no family relationship to the head o.t.h. -0,001 0,999 0,993-0,001 0,999 0,991 # of persons in the household [0-18) 0,026 1,026 0,292 0,026 1,026 0,294 # of persons in the household [18-30) 0,084 1,087 0,007 0,084 1,087 0,007 # of persons in the household [30-65) -0,121 0,886 0,007-0,121 0,886 0,007 # of persons in the household [65+) -0,046 0,955 0,606-0,045 0,956 0,617 unemployment rate - 1. quarter ,005 1,005 0,246 0,005 1,005 0,265 Walloon region Flemish region 0,120 1,128 0,167 0,117 1,124 0,180 Brussels region -0,084 0,920 0,333-0,086 0,918 0,322 log(unemployment-duration) -0,068 0,934 0,020-0,068 0,934 0,020 20

22 Table 3: Estimation results - women (continued) with UH no UH women b exp(b) p-value b exp(b) p-value participation 1.year -0,377 0,686 0,001-0,378 0,686 0,001 participation 2.year -0,197 0,821 0,523-0,199 0,820 0,517 after participation -0,023 0,977 0,921-0,019 0,981 0,935 baseline hazard employment to non-employment: quarter 1 quarter 2-0,732 0,481 0,000-0,732 0,481 0,000 quarter 3-1,062 0,346 0,000-1,062 0,346 0,000 quarter 4-0,733 0,481 0,000-0,733 0,481 0,000 quarter 5,6-1,485 0,227 0,000-1,485 0,226 0,000 quarter 7,8,9,10-1,933 0,145 0,000-1,934 0,145 0,000 distribution of the unobserved heterogeneity points of support: trans. to normal employment 1 -Inf 0,000 trans. to normal employment 2-2,377 0,093 0,000 trans. to activating policy 1 -Inf 0,000 trans. activating policy 2-2,129 0,119 0,001 trans. to non-employment 1 -Inf 0,000 trans. to non-employment 2-0,894 0,409 0,000 probability mass: lam111 = lam112-4,026 0,018 0,000 lam121 lam122-2,969 0,051 0,000 lam211 lam212 lam221 -log (likelihood) # variables # observations

23 Table 4: Estimation results - men with UH no UH men b exp(b) p-value b exp(b) p-value unemployment to activating policy (PAE12) constant -4,752 0,009 0,000 age at the end of ,135 0,873 0,026-0,026 0,975 0,546 Belgian EU not Belgian 0,189 1,208 0,643 0,233 1,263 0,376 not EU -0,211 0,810 0,693-0,441 0,644 0,216 primary school -0,950 0,387 0,023-0,837 0,433 0,003 secondary school - base -0,645 0,525 0,015-0,492 0,612 0,006 secondary school - high college - non-university 0,602 1,826 0,086 0,336 1,399 0,160 college - university 0,117 1,124 0,813 0,123 1,131 0,741 other education 1,424 4,152 0,503-0,008 0,993 0,991 unknown education 2,764 15,857 0,000 1,224 3,402 0,002 head of the household -0,645 0,525 0,163-0,002 0,998 0,996 son/daughter of the head o.t.h. other family relationships to the head o.t.h. -1,861 0,156 0,003-0,253 0,776 0,577 no family relationship to the head o.t.h. -0,430 0,651 0,495-0,250 0,779 0,586 # of persons in the household [0-3) 1,315 3,723 0,025-0,241 0,786 0,468 # of persons in the household [3-12) -0,512 0,600 0,019-0,301 0,740 0,050 # of persons in the household [12-18) -0,201 0,818 0,167 0,152 1,165 0,149 # of persons in the household [18-65) 0,095 1,100 0,360 0,096 1,100 0,177 # of persons in the household [65+) 1,542 4,674 0,000 0,131 1,140 0,606 unemployment rate - 1. quarter ,003 1,003 0,907-0,004 0,996 0,779 Walloon region Flemish region -1,086 0,337 0,015-0,345 0,708 0,283 Brussels region 0,807 2,241 0,012 0,202 1,223 0,350 baseline hazard unemployment to activation: quarter 1 quarter 2 quarter 3,4-0,138 0,872 0,476-0,524 0,592 0,003 quarter 5 0,792 2,209 0,005 0,027 1,028 0,888 quarter 6,7,8,9,10,11-0,843 0,430 0,050-1,836 0,159 0,000 unemployment to normal employment constant -2,684 0,068 0,000 Belgian EU not Belgian 0,012 1,012 0,880 0,000 1,000 0,998 not EU -0,075 0,928 0,403-0,075 0,928 0,387 primary school -0,586 0,556 0,000-0,572 0,564 0,000 secondary school - base -0,414 0,661 0,000-0,406 0,666 0,000 secondary school - high college - non-university 0,065 1,067 0,541 0,091 1,096 0,385 college - university 0,439 1,551 0,000 0,384 1,468 0,000 other education -0,668 0,513 0,002-0,629 0,533 0,004 unknown education 0,655 1,926 0,000 0,721 2,057 0,000 age*college - non-university 0,099 1,104 0,009 0,087 1,091 0,020 age* all other educations -0,028 0,972 0,149-0,027 0,974 0,154 month of entrance 1,2,7,8,9,10,11,12 0,013 1,013 0,785 0,029 1,029 0,539 month of entrance 3,5,6 0,138 1,147 0,001 0,149 1,161 0,000 month of entrance 4 head of the household 0,019 1,019 0,831-0,002 0,998 0,979 son/daughter of the head o.t.h. 22