The E ects of Public Employment Programs on Equilibrium. Unemployment

Transcription

1 The E ects of Public Employment Programs on Equilibrium Unemployment Rafael Lalive, University of Lausanne Tanja Zehnder, University of Zurich y March 30, 2007 Abstract This paper introduces publicly provided employment programs into a standard search and matching model of equilibrium unemployment. Participation in an employment program is time consuming and is therefore less preferred than open unemployment. A calibration exercise indicates that introducing employment programs tends to reduce equilibrium unemployment. This is due to the ex ante e ect whereby job seekers not yet in a program try to evade the program by searching harder. We investigate the empirical relevance of this mechanism using data on Swiss regional employment agencies. Empirical results indicate that higher local program in ow rates induce unemployed workers to search harder for a job supporting the empirical importance of the ex ante e ect. Keywords: active labor market policy, employment program, general equilibrium evaluation JEL Classi cation: C14, C41, J64 CEPR, CESifo, IZA, and IFAU. Address: HEC / DEEP, Internef, CH-1015 Lausanne, Tel Rafael.Lalive@unil.ch y Corresponding author. Address: University of Zurich, IEW, Blümlisalpstr. 10, CH-8006 Zürich, Tel tzehnder@iew.unizh.ch 1

2 1 Introduction Understanding the role of employment programs for job search behavior is of high relevance for policy makers. Employment programs are an important element of active labor market policies in many OECD countries. In particular, public service employment programs make up about 0.1 percent in terms of gross domestic product and about 25 percent of the overall spending on labor market policies. These programs are especially intended to provide jobs to individuals who would otherwise have di culties to nd jobs in the regular labor market. A large literature has investigated the e ects of public service employment programs an participants nding adverse e ects for participants (Heckman et al. 1999). These ndings on the e ects of employment programs on the participating individuals are thus at odds with the wide-spread use of these programs in actual labor market programs. The purpose of this paper is to assess the e ectiveness of employment programs in reducing equilibrium unemployment. We develop an equilibrium search and matching model à la Mortensen and Pissarides where job seekers can participate in an employment program. There is an indisputable position that unemployed workers perceive a compulsory participation in an employment program as a punishment. This is why we model employment programs as time consuming. That means that participation requires work e ort that reduces the time available for other activities. Thus being in an employment programs has the same e ect as a tax on leisure. An unemployed worker is confronted with a possible assignment to an employment program. The probability to participate in employment programs depends negatively on the job seekers search intensity. Thus we capture the fact that in reality labor o ce sta decides on program participation. Our general equilibrium approach shows the e ects of employment programs on all job seekers including unemployed workers who do not participate in a program. Employment programs have two di erent e ects. On the one hand there is the ex ante e ect (e ect on non-participants). The ex ante e ect indicates the search behavior of unemployed workers before they become enrolled in a program. Job seekers not yet in a program try to evade the program by searching harder. This mechanism reduces non-participants unemployment duration. On the other hand there is the ex post e ect which indicates the search behavior of program participants. During program participation, job search intensity may be lowered, because there is less time to search for a job. This suggests that the ex post e ect tends to prolong participants unemployment duration. The two e ects are in the opposite direction and hence the net e ect on unemployment is unclear. Due to the focus on all job seekers we can show the net e ect of employment programs on the unemployment rate whereas recently microeconometric studies only can attest the partial e ect on participants. 1 1 For Switzerland there are three microeconometric studies which show the partial e ects on participants. Ger n and Lechner (2002), Ger n et al. (2003) as well as Lalive et al. (2007) nd that participation in an employment program leads to signi cant lower job nding rates for participants. 2

3 There are at least three innovations in this paper. First, this paper contains an in-depth theoretical analysis of the e ects of employment programs on equilibrium unemployment. Second, the paper uses disaggregated data for various regions within one OECD country to assess the empirical importance of the theoretical predictions. Third, the paper provides key evidence on the parameter that is most important in generating employment enhancing e ects of public service employment programs. We evaluate the empirical content of the model using two di erent complementary empirical exercises. First, we calibrate it to Swiss labor market data for the period Second, we use disaggregated data on unemployment out ow probabilities of job seekers in various public employment service units in Switzerland to test the central prediction that employment programs a ect the ex-ante job search strategies of workers. The following three points are the main ndings of our numerical analysis. First, the introduction of employment programs lowers the unemployment rate. The main mechanism is the threat e ect of participating which e ects the ex ante behavior of unemployed workers. Due to the threat of participating unemployed workers will increase ex ante their search e ort and lower wage claims to avoid an assignment. Thus unemployed workers will leave unemployment faster than without the threat of being assigned to an employment program. Provided that this threat mechanism is strong enough, unemployment is reduced although the program participants reduce ex post their search e orts and as a consequence have a lower job nding rate. Program participants lower their search because of the required work e ort in the program. Second, the employment e ects are higher the higher the threat. Policy makers have two possibilities to increase the threat. On the one hand, they can increase the work time in the program and on the other hand they can increase the assignment rate. More work time in the program increases the disutility of participating. A higher assignment rate boosts the threat e ect of the program. Both policies increase the ex ante search e orts and lower wage claims. As a consequence unemployed workers will faster nd a job. Calibration shows that doubling the assignment rate reduces unemployment duration by about two weeks. This quantitative prediction is con rmed by empirical analysis at the end. Third, employment programs induce less employment e ects compared to a policy that reduces unemployment bene ts. But the policy of employment programs is a good alternative because it is always feasible even in the low income sector. The policy of employment programs does not con ict distributive goals as securing a minimum amount of income for individuals without a job. To compare our quantitative predictions of the model we consider Swiss data for the period from 1998 to There are about 120 local public employment o ces in Switzerland. The federal nature of Switzerland implies that each of these public employment o ces has a great degree of freedom in setting its own rules regarding the probability of sending job seekers into employment programs. Thus we calculate 3

4 the assignment rate for each labor o ce. We then use these rates into employment program participation as an explanatory variable in the job nding rate. We control for demographic characteristics as well as for labor markets condition in di erent regions. This set-up generates two interesting results. First, we nd a signi cant positive relationship between the probability of starting an employment program and the unemployment out ow rate of job seekers who have not entered a program. That implies that unemployed workers who are reported in an o ce with relatively high assignment leave unemployment faster than those who are reported in an o ce with a moderate assignment policy before they have actually been assigned to the program. This positive correlation between the assignment rate and the job nding rate is consistent with a threat e ect of participating in an employment program as captured in the theoretical model. The second result is that our quantitative predictions of the calibration are con rmed. Econometric results indicate that doubling the assignment rate reduces unemployment duration by about two and a half weeks. This e ect is very much in line with a reduction of two weeks predicted by the calibration exercise. Our paper is related to Boone and van Ours (2006). They model how unemployment bene t sanctions a ect unemployment. They showed that bene t sanctions can substantially increase the transition from unemployment to work. 2 However, we analyze how time sanctions a ect unemployment. Time sanctions are indicated by means of employment program. Modeling program participation as a time punishment is consistent with the result of several econometric studies. Geerdsen (2006) found that compulsory programmes do motivate individuals to nd employment prior to participation. In the same way Rosholm and Svarer (2004) found a strong and signi cantly positive threat e ect. There are two other studies which nd that the imposition of additional activity leads to a faster exit of unemployment. Black et al. (2003) nd that claimants increase their exit from unemployment signi cantly when they are informed about their obligations to program participation. Richardson (2002) found that Australian claimants increase their exit rates on the verge of the imposition of the additional activity. This paper is organized as follows. In the next section we describe the Swiss labor market policy in more detail. Section 3 presents the model and section 4 discusses several calibration exercises. In particular, we show the e ects of an introduction of an employment program. Further we compare e ects of di erent designs of work programs. Section 5 presents the empirical results. Section 6 concludes. 2 Modeling Employment Programs Building on the Swiss system of active labor market policies, we nd that the typical features of a public service employment program are 2 There is empirical evidence from the Netherlands and Switzerland that suggest that the imposition of bene t sanctions substantially raises the transition rate to employment (Abbring et al., 2005; Lalive et al. 2005, van den Berg et al., 2004). 4

5 Employment programs are directly created by a government agency Jobs last typically for at least 6 months Participants earn about the same wage as when they are unemployed Programs require full-time presence on the job; they may provide some time for job search. Assignment to these programs is mandatory, refusal to participate can result in nancial sanctions Job seekers do not ask to be sent to an employment program even though they could (Lalive et al. 2006) The following sections introduce this kind of public employment service program into an equilibrium search and matching model. 3 3 The Model 3.1 The search for jobs Individuals have an in nite horizon and enjoy utility out of consumption minus the disutility of work and time spent for search. For simplicity, no savings are possible. When a worker consumes c units and spends t units of time for search he enjoys a utility ow of c1 1 (t). The parameter denotes relative risk aversion, and (t) with 0 (0) = 0; 0 0; 00 > 0 the disutility of search time. All workers are homogeneous in their productivity. There are three potential states: using the familiar Bellman approach, let V e denote the value of having a job, V u the value of being unemployed and V p the value of participating in an employment program. A worker receives wage w, works z units of time and the job gets dissolved at rate such that he gets unemployed: V e = w1 1 (z) + (V u V e ) (1) An unemployed worker receives an unemployment bene t of bw where b 2 (0; 1) is the replacement rate. The value of unemployment in equation (2) consists of three parts: the ow of utility during unemployment (utility of bene ts minus search costs), the expected value change when a job is found and the expected loss when a program assignment happens. Each unemployed has to decide on his search intensity s u. With more time devoted for search, the unemployed increases on the one hand his probability of nding a job, s u : The matching intensity is determined by the labor market conditions and are exogenous to the worker. On the other hand the job seeker decreases the probability that he will be transferred to the employment program with more time devoted for search. Thus we capture the 3 See Foellmi and Zehnder (2007) for the general equilibrium e ect of subsidizing bad jobs. 5

6 fact that labor o ce decides on participation. If the labor o ce sta observe high search e ort of the unemployed worker they will send him with less probability to the program. The probability of being transferred to the employment program is modeled as a Poisson arrival rate ( s u ) with > 0 and 1. 4 V u = (bw) 1 max (s 0 s u 1 1 u ) + s u (V e V u ) + ( s u )(V p V u ) (2) The rst order condition for search of an unemployed agent is then given by 0 (s u ) = (V e V u ) (V p V u ): (3) The value of participation in the employment program is given in equation (4). A participant receives an unemployment bene t of bw: The value of participating consists of two parts: the ow of utility during participation (utility of bene ts minus time costs of searching and participating in an employment program) and the expected value change when a job is found. The participant may invest s p units of time in job search. Participation in the employment program costs p units of time. For simplicity, we assume that employment programs last until a new job is found. With more time devoted for search, the participant increases his probability of nding a job, s p : Higher presence time p in the program lowers the potential time to search of a person who participates. We get the Bellman equation rv p = max 0 s p 1 ((bw)1 (s 1 p + p) + s p (V e V p )): (4) and the corresponding rst order condition of search e ort 0 (s p + p) = (V e V p ): (5) The ows in Figure 1 require that jobless people have incentives to accept job o ers and that jobless people have an incentive to avoid assignment in the employment program. The reason for the refusal of participating in an employment program is the reduced time available for other activities as for example leisure when an unemployed worker enters the employment program. Participating in an employment program imposes only more time costs on unemployed workers. The following proposition establishes the ranking of utilities. Proposition 1 In an equilibrium, employment yields the highest utility, then being unemployed with no participation in an employment program and nally participating in the employment program: V e > V u > V p : 4 1 captures idea that system is imperfect. That means that an agent who search even at the optimal level can be transferred to the program. 6

7 Proof. See appendix. Figure Posting of jobs The value of a vacancy for a job is denoted by J v and satis es J v = cy + m v (J e J v ) (6) where m denotes the number of jobs matched and v the stock of job vacancies. The interest rate equals the discount rate. Per unit of time, a cost of cy accrues to open a vacancy which is proportional to output y. A rm can either occupy a vacancy with an unemployed or with a participant of an employment program. If a vacancy is lled the rm enjoys a value gain of J e J v : We assume free entry in posting vacancies, hence J v = 0. The ow value of an employed job J e then reads J e = y (1 + )w + (J v J e ): (7) where is the exogenous separation rate. Firms have to pay a tax w to nance unemployment bene ts. Using free-entry condition J v = 0 the value of a job can be expressed as follows: y (1 + )w + = v cy: (8) m 3.3 The matching-process on the labor market The pool of job seekers consists of unemployed job seekers and participants of the employment program. The number of contacts between job o ers and job seekers is given by the following matching technology with constant returns to scale: m = A(u u s u + u p s p ) 1 v where u u is the number of unemployed workers and u p the number of participants. The parameter A denotes the productivity of matching. Each stock of job seekers is multiplied with its corresponding search intensity. The search intensity of unemployed is denoted by s u, respectively by s p for the participants. The measure of the supply-side in the matching function is given by u u s u + u p s p, the aggregate amount of search measured in time units. The other input of the matching technology is the total number of vacancies which is given by the mass of vacancies v. The rate at which a job seeker is matched with a rm (per e ciency unit of search e ort) is given by = m u u s u + u p s p = A 7

8 where we de ne, the labor market tightness for jobs, as the ratio of total vacancies to the e ective resource of search: = v=(u u s u + u p s p ): Equivalently, we can express the probability that the rms job vacancy is matched with a worker in term of the labor market tightness m v = A 1 : Finally we state the equations of motion for the stock of unemployed, u u, and the employment program participants, u p (the equation of motion for the stock of workers, e; is redundant in steady state) _u u = e (A s u + ( s u ))u u _u p = ( s u )u u A s p u p : The labor force is normalized to 1 and stays constant. The in ow in unemployment consists of agents who lost their jobs, respectively; e: The stock of unemployed shrinks on the one hand because A s u u u agents found jobs and ( s u )u u agents are assigned to the employment program: The logic for in ows and out ows in the state of employment program is analogous. The in ow in employment program consists of unemployed ( s u )u u : The out ow consists of participants who found a job A s p u p. In steady state the number of workers in each state must remain constant. We get u u = A s p u p = ( s u) e = 1 u u u p where = ( s u ) + A s p ( + A s u + ( s u )): Figure 1 illustrates the ows between the three states, i.e., employment, employment program participation and unemployment. 3.4 Wage bargaining Workers and rms bargain over the wage. A worker and a rm split the joint surplus that their match generates. According to the standard literature we assume the wage is determined by a Nash bargaining. The wage rate of a good job w is then given by: w = arg max w ((V e V u ) + (1 )(V e V p )) 1 (J e J v ) (9) with 0 < < 1 denoting the bargaining power of the job advertiser. We assume for simplicity that a rm owner cannot di er whether a job seeker is unemployed or a participant of an employment program. This means that an unemployed cannot believably prove her position. For this reason both possible fall back positions of a job seeker must be taken into account in the Nash bargaining solution. Both di erent positions are weighted with their relative resource of search ( = s u u u =(s u u u + s p u p )). Note that the fall 8

9 back position of a participant of an employment program is lower than that of an unemployed. Thus the higher the relative number of participants the lower is the claim for wages. To solve for the wage level in (9) we assume that the unemployment bene t level bw is de ned on the economy wide average wage w: Thus an individual worker s choice of wage w does not a ect the unemployment bene t level when he becomes unemployed or a participant of an employment program. That u =@w = 0 p =@w = 0 in the Nash bargaining function. Using equation (8), we get the following rst order condition for w w 1 1 where x = V u V p : = (1 )w cy (1 + )A 1 ( + ) + (z) + ( 1)( + )(V u V p ) + V u (10), w1 1 = (1 )w cy (1 + )A 1 ( + ) + (z) + ( 1)( + )(x) + V u 3.5 Government To close the model we impose the balanced budget constraint of the government we = bwu u + (1 + ')bwu p : (11) ' is the share of program costs on unemployment bene ts. For a given set of policy parameters b; ; ' and, the six equations (1), (3), (5), (8), (10) and (11) give us a solution for the six unknowns w; s p s u ; x; and : 4 Calibration We turn now to our question of primary interest. How does an introduction of an employment program a ect unemployment? What are the consequences of the introduction of an employment program on search intensities and the job nding rates for both kind of job seekers? To gain intuition about the robustness of the results we then discuss how the quantitative e ects change when the level of work time factor in the employment program and the assignment probability is varied. Thus we gain intuition about the possible threat e ect of the program. In a further step we compare the policy of employment program to a reduction in bene ts. In our baseline calibration we use parameter values that are in line with the Swiss labor market of To calibrate the model we have to parameterize the disutility of time (i): We assume that it takes an isoelastic form (i) = 1 (1 i)1 1 i; where 0 < < 1. The parameters are set to values that re ect a quarterly time dimension. The chosen parameter values in our baseline model are given in Table 1. Table 1 9

10 The value of the discount rate is set to which is 10.4% on an annual basis. 5 The value of the replacement rate b corresponds to the Swiss level. The search cost function parameter is set to 0.5. Simulations show that as long as this parameter is not too close to 0 or 1 results do not change substantial. The parameter of the matching function is set to its usual value of 0.5 which implies that the stock of job seekers and vacancies have similar e ect on the ow of lled vacancies. 6 The bargaining parameter is set to = 0:5. 7 The job destruction rate is set to 1.4%. Data shows that monthly ow into unemployment is on average 0.46% which is approximately 1.4% per quarter. The value of the assignment rate is set to corresponds to the data. Data shows that the assignment rate varies between 6% and 3% per quarter. We set the optimal search level to 1.5 ( = 1:5): Further we set the time spent in an employment program (p) to 0.2. This level of work time is equal to 40% of the work time in a regular job. Further, the output productivity y is set to one and the vacancy cost c is set to 3. This implies that in every period the costs of a vacancy are three times the value of production. Finally remember that the marginal wage tax adjusts endogenously to satisfy the government budget constraint (11). We set the share of program costs (') to 80% of the unemployment bene ts expenditure. That value corresponds to the Swiss level. 4.1 Introduction of an employment program First of all we want to discuss the e ects on labor market outcomes when a work progam is introduced. How do the unemployment rate, the search intensities and the job nding rates change when an employment program is introduced? As a basis serves a situation in which an employment program does not exist. What are the changes of the labor market outcomes when an employment program is introduced? A numerical example We compare a situation where no employment program exists with a situation where an employment program is introduced. Table 2 compares the labor market outcomes of the two situations. Table 2 For the situation with no employment program there are no participants of the program. The unemployment rate is 4.37%. This is in line with the average unemployment rate selected from Swiss data. The job nding rate for unemployed is about 31.19% which is in line with ow out of unemployment in reality. 5 This is quite common in these types of exercises. See for example Boone and van Ours (2006). 6 Broersma and van Ours (1999) nd that a value of = 0:5 is a reasonable approximation. 7 Thus we impose =. This assumption corresponds to the Hosios (1990) e ciency condition for the case where agents are risk neutral. Note that in our case workers are risk averse and hence = is not su cient e cient. 10

11 Let us consider the implication for the structure of job seekers rst. Employment program is created and in steady state there are 0.36% job seekers engaged in the program. The simulation show that the unemployment rate decreases from 4.37% to 3.97%. However the job seeker rate decreases from 4.37% to 4.28%. What is the intuition behind this result? Due to the introduction of an employment program there is possible probability that an unemployed job seeker will be assigned to this program. Participating in an employment program imposes only time costs on unemployed workers. Thus the value of being unemployed is lower because a possible participation in the employment program is not attractive. That means that there is a threat e ect of the introduction of the program that a ects the behaviour of unemployed workers before they become enrolled to the program. Thus to avoid a participation unemployed workers increase ex ante their search intensities. Recall that assignment rate depends on the search intensity of the unemployed. Thus the higher the search intensity the lower is the probability to participate. Simulations shows that the search intensity of unemployed increases from 0.68 to A further consequence of the threat e ect is a lower bargaining position of employees in equilibrium and thus wage claims decrease. As a result more vacancies will be advertised and thus the job nding rates for a job seeker (being the sum of unemployed and program participants) per search intensity increase. The increase in the job nding rate per search intensity and the ex ante increase in the search intensity leads to higher job nding rate for unemployed. Thus unemployed will faster leave unemployment. The job nding rate for unemployed increases from 31.19% to 32.33%. Thus an introduction of an employment program increases the job nding rates for unemployed. But to show the overall e ect on the job seekers rate we have to consider all job seekers and not only the unemployed. Thus we turn now to the e ects on the participants. The reaction of the participants is a priori not clear. On the one hand their search intensity should increase because the situation in employment program is associated with less utility as in case of no participation. 8 On the other hand the work time in employment program a ects directly the disutility of time. That increases the search costs of participants and thus search intensity should decrease. Simulation shows that the search intensity of participants is lower compared to the situation with no employment program. The reduced search intensity leads to a lower job nding rate for participants. Thus the search intensity dominates the increase in vacancies. The job nding rate of participants is 26.17%. Thus the job nding rate for a non-participant is 20% higher than the one of a participant. This relation corresponds to the empirical treatment e ect found in microeconometric studies. The e ect on the average job nding rate for job seekers is not clear a priori because we detect a higher job nding rate for unemployed and a lower job nding rate for participants. Simulation shows that the average job seekers rate is higher as in case of no employment program. The reason for that result is the relative small share of program participants. The 8 As in case of non-participants the lower wage also tend to increase the search intensity. 11

12 higher average job nding rate leads to an increase in employment. Simulation shows that the tax rate is higher as in case of no employment program although the job seeker rate is lower. This result comes about because the introduction of program implicates substantial costs. To sum up, we show on the one hand that our model con rms the result found by microeconometric studies that employment programs lowers the job nding rates for participants. On the other hand we have shown that unemployed workers increase their search intensity to avoid program participation. Further the existence of an employment program decreases the wage claims. As a result of that more vacancies will be advertised. The ex ante increase in search intensity and the increase in vacancy leads to a higher job nding rate for unemployed. Simulations showed that the ex ante e ects dominates the ex post e ects. As a consequence employment increases even if the participants nd a job at a lower rate. 4.2 Change in the design of employment program Recall that two attributes determine the characteristics of the employment program. One element is the work time factor. The higher the time in the program the more inferior is the employment program compared to the state of unemployment with no participation. The other feature is the assignment rate to the program. A higher assignment rate boosts the threat e ect of the program E ect of a higher work time factor We examine a doubling in the work time factor. We increase the work time factor from 40% to 80% of a regular job. Table 3 summarizes the simulation results when the work time factor changes. With a higher work time factor a participation in an employment program becomes even less attractive. That leads to a stronger ex ante increase in the search intensity of unemployed. The search intensity of unemployed increases from 0.68 to 0.7 while the search intensity of a participants decrease from 0.55 to A higher work time factor reduces the bargaining power due to the higher disadvantage of participation in the program. Thus wage claims reduce and it becomes more attractive to open jobs. The increase in vacancy and in the ex ante search intensity increases the job nding rate for the unemployed from 32.33% to 34.06%. Thus the unemployment rate decreases from 3.97% to 3.79%. The job nding rate for participants decreases strongly from 26.17% to 21.22% due to the reduced search intensity of participants. Due to the small fraction of program participants the average job nding rate in the economy increases. Hence job seekers rate decreases from 4.28% to 4.15%. In other words employment increases with a higher work time factor in programs. Table 3 12

13 4.2.2 E ect of a higher assignment rate Table 4 shows the simulation results for di erent assignment rates. We double the assignment rate from to 0.05 (we leave the work time factor level of 80% of a regular job) 9. Table 4 summarizes the e ects of doubling the assignment rate. A higher probability to be transferred to the program leads to an increase in the ex ante search intensity. This mechanism is straightforward because with a higher assignment rate the threat is higher and as a result of that unemployed increase search intensity to avoid assignment. There is another mechanism which boosts the increase in search intensity: the higher the higher is the utility of an additional increase of search intensity. The ex ante search intensity increases from 0.70 to The simulation shows that the ex post search intensity namely the search intensity of participants do not change. The higher assignment rate lowers the bargaining position of employees. Thus wage claims reduce and it becomes more attractive to open jobs. The higher ex ante search intensity, the unchanged search intensity of participants and the higher amount of vacancies increases job nding rates for both types of job seekers. The job nding rate of the unemployed increases by about 6.3% from 34.06% to 36.20%. That corresponds to a decrease in the unemployment spell by about 2.4 weeks. Simulation shows that the assignment rate almost doubles. Thus we conclude that a doubling of the assignment rate into employment programs reduces unemployment duration by about 2.4 weeks. The increase in the job nding rate for unemployed and the higher assignment rate leads to a lower stock of unemployed. Unemployment decreases from 3.79% to 3.41%. The e ect on the stock of participants is not clear a priori. On the one hand the in ow increases due to a higher assignment rate and on the other hand the out ow increases due to the higher job nding rate for participants. Simulation shows that program participant rate increases from 0.36% to 0.61%. Simulation shows that the reduction in unemployment rate dominates the increase of participants. Thus the job seeker rate decreases from 4.15% to 4.02%. Table Alternative policy: Reduction of the unemployment bene t A reduction in unemployment bene t is realized if the replacement rate is reduced. In this part we analyze the e ect of a lower replacement rate on the unemployment rate and the other variables in a world with an employment program. Work time in the program is set to 80% of a regular job. We further compare the impacts of the reduction in the replacement rate to the impacts of an increase in work time in the program. 9 Table 5 shows the result for doubling from to 0.05 for a work time factor level of 40% of a regular job. Results are qualitative the same as for a work time factor of 80%. But all e ects are weaker. 13

14 Table 6 shows the simulation results for di erent replacement rates. We reduce the replacement rate b from 0.7 to 0.6. A lower replacement rate decreases the value of being unemployed and increases the marginal utility of income. Since a decrease in the replacement rate increases the gain of nding employment, individuals respond by searching more when unemployment bene t becomes less generous. Further a lower replacement rate lowers the bargaining position of job seekers. As result of that wage claims reduce and it becomes more attractive to open jobs. The increase in vacancies and in the increase in both types of search intensities increases the job nding rate for both types of job seekers. Hence simulation shows that unemployment rate as well as the program participants rate decreases with a lower replacement rate. Thus total job seeker rate decrease from 4.15% to 3.14%. 10 We nd that a policy which focuses on the reduction of the replacement rate has stronger employment e ects than a policy which focus to extend the employment program. A reduction of the replacement rate from 70% to 60% reduces unemployment rate from 3.79% to 2.95%. Whereas a doubling of the assignment rate into a program re ects in a decrease in the unemployment rate from 3.79% to 3.41%. Thus with a focus on unemployment rate a reduction of the replacement rate is more e ective than the policy of extending the employment program. If we focus on the nance of the policy we also determine that the measure of a reduction in replacement rate is favored. A reduction in the replacement rate leads directly to big savings in the bene t payments. But to favor one measure one has to consider also welfare aspects. Politicians have to consider that a reduction of the replacement rate strikes all job seekers whereas policies within employment program do not strike all job seekers. That could be crucial in the sense of welfare aspects. In the end it is to say that reductions of replacement rates are limited in reality. In the class of low income a strong reduction in replacement rate is not feasible because then the bene t would be too low. Thus unemployed with low income would become to relief recipients. Thus a reduction in the replacement rate would be rst of all applied in the high income class. Whereas the policy of employment program can be apply for all income brackets. Table 6 5 Empirical evidence of the ex-ante e ect The theory argues that employment programs have a strong ex ante e ect. The threat of being enrolled to a program leads to an increase in the search intensity and lowers wage claims. Thus the threat of participation a ects the ex ante behavior of unemployed to avoid participation. This ex ante behavior 10 Another interesting point is that the share of program participants is reduced with a lower replacement rate. As already mentioned a lower replacement rate increases the out ow out of job seeker state. Thus employment is increased. At a rate an employee gets unemployed. An employee who loses his job does not enter directly in the program but at rate ( For this reason is the in ow in u p is relatively smaller than the in ow in u u: s u). 14

15 leads to more employment than in the situation without the threat. We further showed that this ex ante behavior is stronger the higher the threat. In particular, our calibration results show that a doubling of the assignment rate reduces unemployment duration by about two weeks. As a criticism one can argue that the source of the ex ante behavior is the speci cation of the employment program. Participation of the program imposes only costs for the worker. Thus a criticism about the modeling of the speci cation of the program can be that there is no gain for the participants. To encourage our speci cation we consider the data. This allows us further to compare the prediction of the calibration exercise with the data. Switzerland has a long tradition of de-centralized political decision making. In the context of active labor market policies decentralization implies that each of the 26 cantons in Zurich decides on the structure and scope of the active labor market program. Within these cantons, there are several public employment service units or local labor o ces. The heads of these o ces have large scope in setting their own placement policy. 11. When we investigate the degree of local heterogeneity in terms of assignment to employment programs, we nd that the probability to participate in a program di ers strongly across the local labor o ces even within one canton. Thus we calculate the participation rate for each o ce. We then use these rates into employment program participation as an explanatory variable in the probability that job seekers leave registered unemployment before they start any active labor market program. The main prediction of the theoretical model is that there will be a positive correlation between the probability of leaving unemployment and the probability of being sent to an employment program. Such a positive correlation between the assignment rate and the job nding rate is consistent with the model. It emphasizes that employment program has a threat e ect which leads to an increase in the job nding rates for unemployed due to higher search and lower wage claims. 5.1 Data The data set used for our analysis covers the period January, December 31, We use the AVAM data set. 12 It contains information about all recorded unemployed workers and reported vacancies. The data structure is monthly. There are monthly records of a person so long as the person is recorded as a job seeker. Thus unemployment episodes which are shorter than a month are not captured in the data set. Job nding rates for unemployed correspond to the ratio of unemployed workers who leave unemployment in time t + 1 to the number of unemployed workers in time t: The logic for the assignment rate to the employment program is analogical. The assignment rate corresponds to the ratio of unemployed workers 11 In Switzerland the labor o ces are called RAV which means regional work agency. 12 The AVAM (System der Arbeitsvermittlung und Arbeitsmarktstatistik) data set serves as a job placement and all local labor o ces are connected to it. The data source has very good quality due to strict enquiry level. Further it is a census. 15

16 who enrol in an employment program in time t + 1 to the number of unemployed workers in time t: Note that we focus on the ex ante e ect of a program. Thus we only consider unemployment spells until the exit into employment or until the rst program participation. Thus the job nding rates for unemployed workers are not a ected by a possible e ect of any program participation. 5.2 Results In this section we present our main results. Our data set contains 119 local labor o ces. 13 In Table 7, we look at the contribution of the assignment rate to the job nding rate. We use the log speci cation for job nding rate as well as for the assignment rate. This speci cation is conventional to estimate individual hazard rates. We use six-year averages because we focus on steady state values and we want to abstract from cyclical uctuations. 14 We use robust standard errors. In column 1, we present the basic model where the average job nding rate is explained by the average assignment rate.we observe a signi cant positive coe cient of That implies that a 1% increase of the assignment rate increases the job nding rate by about 0.162%. A doubling of the assignment rate increases the job nding rate by about 16.2%. The average job nding rate over all local labor o ces is given by 14%. This value corresponds to an average length of unemployment of 7.14 months. 15 On the basis of this average level of the job nding rate and the average length of unemployment a doubling of the assignment rate increases the job nding rate from 14% to 16.3% and lowers the average duration of unemployment from 7.14 months to 6.13 months. In column 2, we control for other programs of active labor market policies. The other two program types are employment subsidies and further training. The assignment rates to these programs are calculated in an analogical sense as the assignment rate to the employment program. The inclusion of the assignment rates to the other two programs does not change our result. The coe cient for the employment program is still positive and signi cant. In column 3, we control for demographic characteristics as sex, nationality age and quali cation. QUALI corresponds to the quali cation of the unemployed worker. Quali cation is classi ed in three numbers. One is equal to skilled, two is equal to semi-skilled and three corresponds to unskilled. The inclusion of demographic characteristics variables leads to a 13 Today in Switzerland exists circa 130 local labor o ces. We consider only the local labor o ces which persist over the whole data period. Some of the o ces were closed and some were disclosed during the considered time period. Therefore we consider 119 local labor o ces. These o ces persist over the six years. 14 We also split the six-year period into two periods. None of the main results are a ected. Further not that even if we use annual data our results are still persisting. 15 For this calculation we assume that the exit of unemployment can be approximated by a Poisson process. denotes the job nding rate (probability to exit unemployment). If the exit process is Poisson then the duration of unemployment T has the following density function: exp( t); where t denotes the lapsed time in unemployment. The expected length 1R of unemployment is equal to E(T ) = tf(t)dt: Thus one can show that E(T ) = 1=: 0 16

17 coe cient for the employment program which is still positive and signi cant. Finally, we use in column 4 canton dummies. Thus we can control for the fact that the supply of labor market policies is arranged on a cantonal level. Thus we can eliminate the correlation that cantons with bad labor market chances provide less employment programs and vice versa. We observe a signi cant coe cient of for the employment program. That implies that a 1% increase of the assignment rate increases the job nding rate by about 0.093%. A doubling of the assignment rate increases the job nding rate by about 9.3%. 16 As already mentioned the average job nding rate over all local labor o ces is given by 14%. Thus a doubling of the assignment rate increases the job nding rate from 14% to 15.3%. In other words the length of unemployment decreases from 7.14 month to 6.54 month. That corresponds to a reduction of the unemployment spell by about 2.4 weeks. Finally, we consider the four explanatory variables which characterize the unemployed workers. None of them is signi cant. A reason for the insigni cance is that the average characteristics of an unemployed do not di er across the local labor o ces. To sum up, there is empirical evidence for a threat e ect of participation in an employment program. Furthermore, the prediction of the calibration is con rmed. Data show that a doubling of the assignment rate leads to a reduction of the unemployment spell by about 2.4 weeks whereas the calibration exercise predicts a reduction of 2 weeks. Table 7 6 Conclusion This paper discusses the general equilibrium implications of publicly provided employment programs. A search and matching model of unemployment predicts that equilibrium unemployment will be lower in a situation where these programs are used than in the counterfactual situation without the program. This e ect arises due to the threat of participation that a ects the ex ante behavior of unemployed workers. Program participation requires work e ort that reduces the time available for other activities. Thus an unemployed worker who he is confronted with a possible assignment to an employment program increases his search intensity to avoid participation. This leads to a further reduction in the wage demanded when bargaining at the beginning of a new job. As a result of lower wages more vacancies will be advertised. The ex ante increase in search intensity and the increase in vacancy leads to a higher job nding rate for unemployed workers who not participate in a program. Further we showed that participants have a lower job nding rate compared to a situation with no program. This result is in line with the microeconometric evaluation literature nding negative e ects of 16 Simulations (see table 4) show that a doubling in the assignment rate for a work time factor of 80% increases the job nding rate by about 6.32%. Thus the simulation result is in line with the empirical result. 17

18 employment programs on the employment chances of treated workers. In spite of the lower job nding rate of participants employment programs lead to more employment than in the situation with no program because the ex ante e ect dominates the ex post e ect. Thus employment programs do not shorten unemployment duration due to a positive treatment e ect but rather due to the threat e ect of the program. Furthermore employment e ects are higher the higher the threat. Policy makers have two possibilities to increase the threat. On the one hand they can increase the work time in the program and on the other hand they can increase the assignment rate. With more work time in the program the disutility of participating is higher. A higher assignment rate boosts the threat e ect of the program. Both policies lead to a stronger ex ante e ect and as a consequence to more employment gains. Calibration results show that a doubling of the assignment rate decrease the duration of unemployment by about 2 weeks. Compared to a policy which focuses on reducing replacement rate the policy of employment program is not favored if the focus is the e ectiveness of reducing unemployment rate. But the policy of employment programs is always feasible in the sense that it can be applied for all unemployed because there is no drop of income due to this policy. Based on Swiss data ( ) we give an empirical evidence of the threat e ect of employment programs and con rm the prediction of the calibration. We nd that a doubling of the assignment rate reduces unemployment spell by about 2.4 weeks. The quantitative results of the calibration match this nding well. 18

19 References [1] Abbring, Jaap, van den Berg, Gerard J., and van Ours, Jan (2005). "The e ect of unemployment insurance on the transition rate from unemployment to employment", Economic Journal, vol. 115 (June), pp [2] Black, Dan, Smith, Je rey A., Berger, Mark and Noel, Brett J. (2003), "Is the threat of reemployment services more e ective than the services themselves? Evidence from random assignment in the UI System", American Economic Review, 93: [3] Boone, Jan and van Ours, Jan 2006, Modeling nancial incentives to get unemployed back to work, Journal of Institutional and Theoretical Economics, Vol. 162 (2), pp [4] Broersma, Lourens and van Ours, Jan (1999). Job Searchers, Job Matches and the Elasticity of Matching, Labour Economics 6, [5] Foellmi, Reto, and Zehnder, Tanja (2006). "Is it good to subsidize bad jobs?", mimeo, University of Zurich. [6] Geerdsen, Lars Pico (2006). "Is there a Threat E ect of Labour Market Programmes? A Study of ALMP in the Danish UI System," Economic Journal, Royal Economic Society, vol. 116(513), pages , 07 [7] Ger n, Michael and Lechner, Michael (2002). Microeconometric Evaluation of the Active Labor Market policy in Switzerland, The Economic Journal 112, [8] Ger n, Michael, Lechner, Michael und Steiger, Heidi (2003). "Does subsidised temporary employment get the unemployed back to work? An econometric analysis of two di erent schemes", forthcoming in: Labour Economics. [9] Heckman, James J., LaLonde, Robert J. and Je rey A. Smith (1999). "The Economics and Econometrics of Active Labor Market Programs", Handbook of Labor Economics, Vol. IIIA, ed. by O. Ashenfelter and D. Card. [10] Hosios, Arthur J.(1990). "On the e ciency of matching and related models of search and unemployment", Review of Economic Studies, vol. 57(2), pp [11] Lalive, Rafael, van Ours, Jan and Zweimüller, Josef (2005). "The e ect of bene t sanctions on the duration of unemployment", The Journal of the European Economic Association, 3(6):1-32. [12] Lalive, Rafael, Zweimüller Josef, and Tanja Zehnder (2006). "Makroökonomische Evaluation der Aktiven Arbeitsmarktpolitik der Schweiz", Ministry of Economic A airs, Seco, Berne. 19