Wage Subsidies for the Unemployed: Does their Long-Run Effectiveness Change over Time?

Transcription

1 Wage Subsidies for the Unemployed: Does their Long-Run Effectiveness Change over Time? Marina Furdas University of Freiburg This version: February 2015 Abstract: This paper investigates the long-run effectiveness of a German labor market policy combining employer-side wage subsidies with training on the future labor market prospects of participants over time. In the empirical analysis we estimate calendar time-specific effects in a dynamic setting using discrete monthly data in the period from February 1980 to December We employ selection on observables assumptions for identification and estimate program effects conditional on the elapsed unemployment duration using inverse probability weighting. Our findings imply a reasonable time-specific variation in the long-run program effectiveness, which to some extent is decreasing over the years. Individuals receiving on-the-job training through the subsidy from 1993 onwards do not benefit from the program compared to similar unsubsidized workers. Further, we find only weak evidence for a positive relationship between program effectiveness and the economic conditions prevailing at program start. Finally, the program proves to be quite ineffective for participants with relatively short unemployment experience, thus suggesting potential for deadweight losses when the subsidy is granted to firms hiring newly unemployed individuals. Altogether, our findings strengthen the importance of sufficiently long post-treatment observation periods and well-constructed comparison groups in the evaluation of employer-side wage subsidies based on non-experimental data. Keywords: Wage subsidies combined with training, active labor market policy evaluation, administrative data, Germany JEL-Classification: J64, J68 I am grateful to Bernd Fitzenberger, Peter Mueser, Olga Orlanski, and Christoph Sajons for helpful suggestions and comments. Further, I thank Michael Rumert from the labor agency in Emmendingen for insightful information on the training measure evaluated in this study. This paper is part of the project Policy change, effect heterogeneity, and the long-run employment impacts of further training programs ( Politikänderung, Effektheterogenität und die längerfristigen Beschäftigungswirkungen von Fortbildung und Umschulung, IAB project number: ). Financial support by the IAB and BA is gratefully acknowledged. The responsibility for all errors is, of course, mine.

2 1 Introduction Active labor market policies(almps) play an important role in many developed economies. One key component within the toolbox of ALMP represents employer-side wage subsidies or hiring credits. These are labor demand-oriented measures that in general provide temporary payments to employers in order to stimulate hiring and facilitate reemployment among vulnerable groups of individuals, such as the low-skilled, unemployed, or disadvantaged workers. In the aftermath of the Great Recession, such job creation policies gained importance and have been implemented in many OECD countries. 1 In Germany, employer-side wage subsidies were implemented in the 1970s and nowadays represent an integral part of the German ALMP (see Wolff and Stephan (2013) for an overview of currently existing subsidized work schemes). The main rationale in favor of using temporary employer-side wage subsidies is derived from the traditional labor demand and supply framework. Wage subsidies reduce the relative costs of labor for an employer utilizing the subsidy by raising labor demand for the target group of workers and increasing employment and paid wage rates among this population. This effect depends, however, on labor demand and supply elasticities as well as on issues such as administrative costs and program awareness (Katz, 1998). In combination with training, wage subsidies may lead to a win-win situation for both firms and workers in so far as they compensate employers for the reduced worker productivity and provide employees the opportunity to improve human capital and increase future employability while learning and acquiring skills from the job (Jaenichen and Stephan, 2011). Further, based on previous findings from the existing literature, Neumark (2013) concludes that hiring subsidies with low targeting level and concentrating mainly on unemployed individuals, might be more likely to avoid stigma effects and to increase employment among the treated population. However, employer-side wage subsidies are often criticized due to the fact that they are likely to induce substantial adverse effects on the aggregate level including, among others, displacement effects, substitution effects, and deadweight losses (Calmfors, 1994; Heckman et al., 1999). The effectiveness of employer-side wage subsidies on the future labor market prospects of unemployed individuals is investigated empirically in a number of recent studies. 2 The 1 Prominentexamplesarethe 2010Hiring IncentivestoRestoreEmployment(HIRE) Act in the United States (Neumark (2013)) and the Zéro Charges Program in France (Cahuc et al., 2014). 2 SeeBelletal.(1999)andDorsett(2006)forUK;Katz(1998)andHamersma(2008)fortheUS;Sianesi (2004, 2008), Carling and Richardson(2004), Forslund et al.(2004), and Fredriksson and Johansson(2008) for Sweden; Huttunen et al. (2013) for Finnland; Gerfin and Lechner (2002) for Switzerland; Jaenichen 1

3 existing findings differ mainly with respect to the primary parameter of interest (eligibility versus actual participation), the target group of the wage subsidy (youths, elderly, (longterm) unemployed, disadvantaged), and last but not least on the design of the program(in terms of size and subsidization period). In general, studies investigating the effect of actual participation find relatively large and positive employment effects, while empirical research on the impact of subsidy eligibility yields more pessimistic results. In a meta-analysis, Kluve (2010) examines the effectiveness of various active labor market policies from 19 European countries and finds positive effects for subsidies to private sector employment. In another meta-study, Card et al. (2010) find that classroom training and on-the-job training programs tend to be more effective in the medium run compared to subsidized public or private employment. Despite the large existing literature on the evaluation of active labor market policies in general and on employer-side wage subsidies in particular, relatively little is known on whether the effectiveness of training programs changes over time. To our knowledge, there are only two studies that directly address this research question. The influential study by Lechner and Wunsch (2009) investigates the effects of labor market training programs for unemployed individuals on a monthly basis over a 10-year period in West Germany. Their results suggest considerable variation in program effectiveness, which is significantly related to the initial labor market conditions at program start. Lechner and Wunsch (2009) find smaller lock-in effects and larger long-run positive effects of labor supply-oriented training in times of high unemployment. Forslund et al. (2011) use the within-variation of local unemployment rates over time in order to identify business cycle effects on the relative effectiveness of vocational training and on-the-job training programs in Sweden. Their findings suggest that training programs with larger lock-in effects are relatively more efficient in recessions than in booms. The present study aims to complement the existing evaluation literature for Germany by analysing the effectiveness of employer-side wage subsidies over time. With respect to the evaluation strategy and research design, our paper is closely related to the study by Lechner and Wunsch (2009). The main difference, however, is that we analyse a policy measure that primarily aims at stimulating labor demand rather than to activate labor supply. We pay particular attention to the long-run program effectiveness as most of the studies evaluating similar wage subsidy schemes for Germany suffer from relatively short and Stephan (2011), Bernhard et al. (2008), Stephan (2010), Neubäumer (2010), Schönemann et al. (2013) for Germany. 2

4 post-treatment periods(e.g. three years), thus suggesting a potential for an overestimation of the treatment effects of interest. The program we focus on is a labor market policy instrument in Germany which combines wage subsidies with training (hereinafter referred as to WST). Up until 1997, it had been regulated in 49 Employment Promotion Act (Einarbeitungszuschuss) and constituted a key component of the German ALMP. The program aims at encouraging firms to hire individuals with reduced working productivity who need on-the-job training to adapt to new skill requirements in order to be fully productive in present and/or future jobs. We evaluate the short-, medium, and long-run effectiveness of WST program participation on the future labor market prospects of unemployed individuals on a monthly basis over a 16-year period (1980 to 1995). The sufficiently long time coverage allows us to account for the institutional and economic environment at the time when individuals start the program, which is particularly important for at least two reasons. First, in the 1980s and 1990s, the legal framework for supporting eligible individuals with the wage subsidy program had been amended several times. The maximum amount of the subsidy was permanently reduced from 80% in the early 1980s to 30% in the mid-1990s. Further, from 1993 onwards the maximum program length was also restricted, from what was originally one year to six months. Such changes may alter the program assignment of individuals by caseworkers in the local employment offices on the one hand and the willingness of firms to make use of the wage subsidy program on the other. Second, the economic conditions during that time had changed as well. After the economic slowdown in the mid-1980s, the West German economy had begun to recover and had experienced an economic upswing in the first two years after the German reunification, followed by particularly high unemployment and sharply deteriorating labor market conditions in the mid-1990s. This paper takes advantage of large administrative records from the German Federal Employment Office based on a 50% random sample of all individuals that have participated in training at least once during the period 1980 to 1997 and a 3% random sample of all individuals without any training history up until The data is very similar to the data set used in Lechner and Wunsch (2009) with the exception that we rely on a much larger random sample of training participants, which allows ut to evaluate the effectiveness of the wage subsidy program with training, which is actually a small-scale training measure compared to further training and retraining, over time. In the empirical analysis we rely on selection on observables identification assumptions and use inverse probability 3

5 weighting to estimate calendar time-specific average treatment effects on the treated in a dynamic setting based on discrete monthly data. Further, we define treatment and comparison groups conditional on the elapsed unemployment duration by looking at the individual-specific treatment status each month during unemployment as long as individuals remain unemployed and are eligible for treatment. Similar to Jaenichen and Stephan (2011), we use two comparison groups in order to approximate the counterfactual outcome of WST participants in the non-participation state an unemployed comparison group (UCG) consisting of unemployed individuals who, according to the Sianesi (2004) concept, might be treated later or never and an employed comparison group (ECG) consisting of individuals who enter unsubsidized employment at the same time as participants start subsidized work. Our results show negative lock-in effects of WST participation and positive and for the most part significant effects in the medium and long run. However, in line with the existing literature, we find that program effects estimated with the unemployed comparison group are considerably larger than those estimated with the employed comparison group. With regard to our main research question, our findings reveal a reasonable time-specific variation in the (long-run) program effectiveness, which is to some extent decreasing over the years. It turns out that unemployed individuals entering the program during the most generous policy regime (1982 to 1983) benefit most from participation, whereas unemployed individuals entering the program from 1993 onwards do not experience any significant changes in their later labor market prospects compared to similar unsubsidized workers. Moreover, we find that employer-side wage subsidies, as evaluated in this paper, tend to be quite ineffective when reducing unemployment benefit recipiency in the future and even seem to harm the treated in the early 1990s. Following Lechner and Wunsch (2009), we examine whether the effectiveness of WST is correlated with the unemployment rate at program start and find only weak evidence for a positive relationship. However, our findings suggest that in the short run, program effects are mainly driven by the average planned subsidy duration. We further examine effect heterogeneity with respect to the time individuals spent in their current unemployment spell before they become treated. Our findings suggest that individuals with longer elapsed unemployment duration (4 to 12 months) are more likely to benefit from the subsidy. In other words, the wage subsidy program with training proves to be quite ineffective for participants with relatively short unemployment experience, thus suggesting potential for deadweight losses when the subsidy is granted to firms hiring newly unemployed individuals. 4

6 The remainder of the paper is structured as follows. Section 2 describes the program and provides a short overview of the most important institutional changes related to the program. Section 3 describes the data and the research design. Section 4 presents the econometric approach. The empirical results are reported and discussed in section 5. Section 6 concludes. 2 Wage subsidies as part of German active labor market policy The most important objectives of Active Labor Market Policy (ALMP) include enhancing labor market mobility and improving labor market chances of individuals, especially of the unemployed and those at risk of unemployment. Until December 1997, ALMP in Germany was regulated in the Employment Promotion Act (Arbeitsförderungsgesetz, EPA) and executed by the Federal Employment Office (FEO) and its local authorities. On January 1 st 1998, the EPA was abandoned and replaced by the German Social Code as its Third Book (SGB III). In the 1980s and 1990s, training programs sponsored by the public sector (Fortbildung und Umschulung, FuU) were one of the most frequently used instrument in the German ALMP. Public-sponsored training consist of short- and long-term further training and retraining programs as well as employer-side wage subsidies with initial training requirements according to 49 EPA. While a number of recent papers have evaluated the effectiveness of further training and retraining 3, there is little empirical evidence on the impact of wage subsidies as part of the German ALMP. 4 This holds in particular for temporary employer-side wage subsidy schemes with a training component. 2.1 Program objectives This study focusses on the effectiveness of a temporary employer-side wage subsidy program for initial skill adaption training in Germany as regulated by 49 EPA (Einar- 3 See among others Fitzenberger and Speckesser (2007), Fitzenberger and Völter (2007), Fitzenberger et al. (2008), Wunsch and Lechner (2008), Lechner et al. (2011), Fitzenberger et al. (2013), Biewen et al. (2014). 4 An exception for Germany is the study by Jaenichen and Stephan (2011) evaluating an employer-side wage subsidy scheme without any training requirements and targeted to hard-to-place workers. 5

7 beitungszuschuss, EaZ). 5 A distinctive characteristic of the program compared to other public-sponsored training measures, such as further training and retraining, is that the provision of training activities and skills to the unemployed takes place in a subsidized employment relationship. The primary objective of the subsidy was to provide general and firm-specific skills facilitating the occupational mobility and the competitiveness of program participants in the labor market. Although this type of training did not explicitly target specific groups of individuals, as for instance long-term unemployed or disadvantaged, it had been designed for individuals with reduced productivity for whom on-the-job training was required in order to be fully productive in present and/or future jobs. For the rest of the paper, we refer to the program as to a wage subsidy program with training, in short WST. According to 49 EPA, the possibility of promoting new employment contracts with the subsidy was only possible when the productivity of the subsidized worker was expected to increase during the treatment period. 6 The group of eligible individuals included not only unemployed or individuals at risk of unemployment, but also currently employed workers. Moreover, participation in WST subsequent to other training measures such as further vocational training or retraining, was possible as well. In the yearly 1980s, the subsidy could also be granted if the potential program participant was regularly employed at the firm receiving the grant. Contrary to further training and retraining programs, where the selection of unemployed individuals into different programs was driven primarily by the discretionary power of the local caseworker and the supply of courses, the assignment mechanism regarding the provision of WST involves at least three parties: the caseworker, the potential participant, and the firm. Although there is no clear evidence on the exact selection process, there are three conceivable scenarios to be mentioned. First, employers with available vacancies can take the initiative and request suitable candidates from the local employment office. 7 Second, employers can demand support by the local authority in order to hire 5 Based on this, the FEO determined the conditions and the practical implementation on part of the local employment offices in supplementary guidelines for further training programs( 19 AFuU, Anordnung für Fortbildung und Umschulung). 6 This contrasted with the subsidy to regular employment according to 54 EPA (Eingliederungsbeihilfe), which did not have any training components and was targeted to individuals with permanent lower productivity. Other wage subsidy schemes that were directly paid to firms included the integration assistance scheme as regulated by 58 EPA and targeted to disabled (Eingliederungshilfe) as well as the employment subsidy as regulated by 97 EPA and targeted to older unemployed workers (Lonkostenzuschuss für ältere Arbeitslose). None of these instruments had any training component but primarily aimed at the permanent (re)integration of disabled and disadvantaged groups into the labor market. 7 According to officialsin the local laboragencyin Emmendingen, approximately50% ofthe subsidized 6

8 a particular unemployed worker. Third, the caseworker can take the initiative by either offering a subsidy for a particular job-search match or by making a commitment to grant a subsidy in case the unemployed individual receives a job offer (Stephan, 2010). In all three cases, the employer as the recipient of the subsidy must turn in an application at the local employment office before the start of the program and explain the necessity of training. The application should also involve a detailed plan describing the training activities provided to the employee during the subsidization period. However, since the provision of WST was within the discretion of the FEO, the final approval was mainly up to the caseworker in the local employment office and had to be decided on a case-by-case assessment. The amount of the subsidy and the duration of the program were strongly related to the period that was necessary to fill the gap between actual and desired working productivity of the program participant. According to 49 EPA and 19 AFuU, the subsidy was granted for a minimum of four weeks and a maximum of twelve months. In the early 1980s, the subsidy amounted up to 80% and could be raised by 20 percentage points in special cases. The exact program length and amount were, however, negotiated between caseworker and employer. In particular, the size of the subsidy was determined on the basis of the collective bargaining wage or the prevailing local wage in the respective occupational field. One special feature of the program is the so called protection period, which bound the firm receiving the grant to employ the eligible individual for a follow-up period of usually the same length as the duration of the subsidy. In the early 1990s, the law additionally prescribed financial sanctions for employers terminating the employment relationship during the protection period for reasons not related to the subsidized worker. 2.2 Program changes since 1980 In the 1980s and 1990s, the legal framework for supporting eligible individuals with WST had been changed many times. 8 The first important policy change took place with the Law for the Consolidation of Employment Promotion (AFKG) in January The target group of potential participants was restricted mainly to unemployed individuals and individuals at risk of unemployment. Furthermore, WST by the previous employer employment relationships emerged in this way. 8 Table 2 provides an overview of important legislative modifications of the Employment Promotion Act regarding design and implementation of the program. See also Bohlen (1993) for a detailed discussion of the policy changes occurred in the 1980s. 7

9 could only be granted if the worker was at risk of loosing his/her job and the employeremployee relationship was already terminated. The promotion of new contracts by the previous employer was entirely abolished by the Budget Supplement Act (HaushBG) in Additionally, the maximum amount of the wage subsidy was reduced from 80% to 70%. In January 1986, the program design was further modified by the 7 th EPA Amendment (Siebte AFG Novelle), which explicitly allowed firms to make use of the wage subsidy program for temporary employment relationships. The intended aim of this modification should possibly reflect the primary subsidy objective of improving the occupational mobility among program participants. Temporary contracts were eligible for WST only until the end of The 9 th EPA Amendment (Neunte AFG Novelle), that became effective in January 1989, introduced two important changes. First, the maximum wage subsidy rate was reduced from 70% to 50%. Second, the law prescribed that this subsidy scheme had to be targeted mainly to individuals willing to return to the labor market after a period of child rearing or support and care of relatives. Until the end of 1992, the financial support for this specific group was within the discretion of the FEO and its local authorities ( Kann- Leistung ). It was then transformed into a non-discretionary measure ( Pflichtleistung ) in 1993 by the 10 th EPA Amendment (Zehnte AFG Novelle). The 1993 reform induced a further reduction of the maximum amount of the subsidy from 50% to 30%. The maximum program length was restricted as well, from one year to six months. In addition, 25 AFuU from May 1993 prescribed further conditions and restrictions on the provision of WST. Three important changes must thus be mentioned. First, the new job contract between the firm receiving the grant and the potential program participant had to be originated by the local employment office. Second, the protection period associated with the subsidy was explicitly stated in the law and a partial repayment obligation was introduced for those employers who did not comply with the legal requirement. Third, WST could not be assigned to individuals who have finished a vocational training within one year prior to the beginning of the subsidized work. In January 1998, the Employment Promotion Act was replaced by the Third Book of the German Social Book(SGB III). From 1998 to 2004, the wage subsidy scheme according to 49 EPA was regulated in SGB III and represented, together with the wage subsidies for older workers ( 97 EPA) and those for hard-to-place individuals ( 58 EPA), one of three possible versions of the Integration Supplement (Eingliederungszuschuss, 8

10 EGZ). Differences among these instruments were mainly due to the target group and the program design, with WST being the most restrictive both in terms of subsidy rate and duration. As a consequence of the Hartz III reform in 2004, the three EGZ-versions were consolidated into one labor market instrument targeted to individuals with placement barriers(wolff and Stephan, 2013). The resulting integration subsidy is part of the current German ALMP and can cover up to 50% of the employee s wage for a fixed period of up to twelve months. 9 The aforementioned policy reforms led to changes in the development of WST as an active labor market policy instrument both in terms of utilization and resources spent for the program. Table 1 shows that on average program participation and expenditures on WST were quite low in the first half of the 1980s and changed only moderately over time. Over the period 1986 to 1988, however, participation in the WST program doubled compared to the two years before, remained almost unchanged up until 1993, and dropped to approximately 10 thousand participants per year thereafter. Expenditures, on the other hand, increased again at the beginning of the 1990s up to about 295 million Euro annually and declined sharply to about 33 million Euro per year from 1993 onwards. Program participation was relatively cheap during the 1980s ( Euro) compared to the early 1990s, in which average monthly training costs per participant amounted to approximately 1200 Euro costs per year. Table 1: Inflows into the WST program, expenditures (in 1000 Euro), and costs per participant by institutional framework (annual data) Time frame Inflows Expenditures Subsidy duration Costs per participant a) Note: Inflows and expenditures are taken from the official statistics by the Federal Employment Office for West Germany (Amtliche Nachrichten der Bundesanstalt für Arbeit, ). The subsidy duration is measured in months and is imputed from the data used for the empirical analysis (see next section). a) Training costs per participant are measured in Euro and calculated on a monthly basis as the average annual expenditures divided by the number of annual entries and then multiplied by the average subsidy duration. Figure 1 shows the development of expenditures on employer-side wage subsidies in 9 See Brussig et al. (2008) for a detailed description of the current integration subsidy scheme as well as Wolff and Stephan (2013) for an overview of evaluation results of major subsidized work programs before and after the Hartz reforms. 9

11 million Euro(solid line) and unemployment rate in percent (dashed line) in West Germany during the period 1980 to We plot the percentage share of WST spending on total ALMP expenditures (dotted line) as a crude measure of their relative importance among the instruments of ALMP based on the EPA. The graph reveals strong evidence of procyclical spending variation up until 1997, i.e. spending on wage subsidies declined in times of high unemployment and increased in periods of good economic conditions. Figure 1: Expenditures on wage subsidies and unemployment in West Germany Million Euro Last year of ALMP regulation under the Employment Promotion Act Percent Expenditures Unemployment rate WST/ALMP Note: Own calculations based on the official statistics by the Federal Employment Office (Amtliche Nachrichten der Bundesanstalt für Arbeit, ). Expenditure series refers to the wage subsidy program with training according to 49 EPA ( ), the integration supplement ( ), and the integration subsidy ( ). Apart from institutional changes, this spending profile might be explained by the following two reasons. First, during recessions general government expenditures on welfare transfers are likely to crowd out spending on active labor market measures facilitating either labor demand or labor supply. Second, since WST is a demand-side oriented measure, boosting employer-side wage subsidy spending might be difficult in times when firms are actually reducing their workforce and are thus reluctant to hire. 3 Data and evaluation sample 3.1 Data This study uses rich administrative data containing individual information on employment, unemployment, and further training participation in Germany. 10 The database is 10 We use the same database as in the previous two chapters. 10

12 constructed from two main sources: a special version of the IEB data (Integrated Employment Biographies) and the St35 (also known as FuU) data on program participation. The IEB dataset is based on daily records from the Employment Register Data (Beschäftigten-Historik, BEH) of the Federal Employment Office and the Benefit Payment Register (Leistungsempfänger-Historik, LeH) of the Institute for Employment Research. It covers periods of registered employment and/or periods of unemployment in which benefit payments were received by all individuals that were subject to social insurance contributions from 1971 to The IEB data provide information on individual, job-specific, and regional characteristics. Participation in further training or retraining programs is identified based on the receipt of income maintenance payments (Unterhaltsgeld). However, this data source does not allow to distinguish between subsidized and unsubsidized employment. The primary evidence on training program participation is drawn from the FuU data covering the period from 1976 to The dataset is based on questionnaires filled out by the caseworkers in the local employment offices and is collected by the Federal Employment Office for control and statistical purposes. 11 The FuU data consist of monthly based records on participation in further training ( 41 and 43 EPA), short-term training programs ( 41a EPA), retraining ( 47 EPA), temporary employer-side wage subsidies for initial skill adaption training (Einarbeitungszuschuss, 49 EPA), and German language courses ( 62 EPA). This dataset provides detailed information on the training spell (e.g. type and planned duration) as well as on the individual characteristics of program participants. For the purpose of the underlying project, a 50% random sample of all individuals that have been in training at least once until 1997 has been separately drawn from the FuU and the IEB data. In addition, we use a 3% random sample from the underlying IEB data without any program participation in the period from 1980 to Since the FuU training database provides only monthly spell information during participation, it has been extended with un(employment)-specific information from IEB by performing a merge procedure very similar to that described in Bender et al. (2005). 13 The resulting 11 See Bender et al. (2005) for more detailed information. 12 All datasourceshavebeen drawnaccordingtothe socalled birthdayconcept. Inboth50%samples, one draws 50% of all possible birthdays, starting on January 2 nd and then selecting every second day, which results in 182 birthdays. The draw of the 3% sample is based on 12 out of 182 birthdays and excludes records that have already been drawn before. See Fitzenberger et al. (2011) for further details on the data generation process. 13 The merge procedure is based on a personal identification number and additional criteria like consis- 11

13 dataset, henceforth merged FuU-IEB data, contains more precise information on a variety of public-sponsored training programs in comparison to the 50% IEB sample. It is also the only data that allows for identification of the wage subsidy scheme for initial skill adaption training as evaluated in this study. Using this information, we distinguish between jobs created with the wage subsidy (hereinafter referred to as subsidized employment) and jobs created without the wage subsidy (hereinafter referred as to unsubsidized or regular employment). 14 Based on the merged FuU-IEB data, table 3 shows the current state distribution of WST participants directly before program start. It turns out that approximately 55% of all participants received unemployment benefits or unemployment assistance, about 18% were regularly employed, 14% were non-employed, and only 10% started WST subsequently after having participated in some FuU training measure. Moreover, as indicated in table 4, those individuals who transition from unemployment into subsidized employment suffer only a short unemployment experience. About half of the entries in WST occurred during the first quarter of unemployment and almost 95% during the first unemployment year. Thus, the wage subsidy program with training was not targeted to individuals with placement difficulties in terms of their unemployment experience. The data for the empirical analysis is an integrated dataset, which combines the merged FuU-IEB data, the 50% IEB sample, and the 3% IEB sample. Since individuals are sampled from different data sources with different sampling probabilities, all subsequent empirical results are generated using sampling weights for estimation. 15 The integrated dataset is superior to the data generated by Bender et al. (2005) due to its large sample size that allows for the evaluation of rather small training measures such as wage subsidies for initial training according to 49 EPA. Further, we take advantage of a detailed preprogram history and a relatively long follow-up period (more than 10 years after program start). The sufficiently long observation period is particulary important for this training program seeing as subsidized workers will continue to be employed in the short run due tency in time and content of the corresponding spells. For a detailed description, see Fitzenberger et al. (2011). 14 In the same vein, we differentiate between subsidized and unsubsidized workers as well as between subsidized and unsubsidized firms, respectively. Note that except for the wage subsidy according to 49 EPA, our data does not allow to identify other types of subsidized employment, such as job creation schemes. 15 The sampling weight is the reciprocal of the probability with which an individual is selected from a given data set. Observations from the merged FuU-IEB data set and the 50% IEB sample receive a sampling weight of 365/182, whereas observations from the 3% IEB sample receive a sampling weight of 365/12, respectively. 12

14 to the protection period associated with the subsidy. 3.2 Evaluation sample The evaluation sample for the empirical analysis consists of unemployment cohorts that begin during the period from January 1980 to December We only consider individuals that become unemployed in a particular calendar month after being regularly employed for a period of at least six months. Unemployment spells are identified based on a permanent transfer payment information (unemployment benefit or unemployment assistance) observed within the third month of unemployment duration at the latest. Individuals who do not transition from unemployment to either regular employment or some public-sponsored training program, and therefore remain non-employed for more than 60 months, are excluded from our further empirical analysis. 16 The sample is further restricted to individuals living in West Germany and aged 25 to 50 years at unemployment start. 17 The focus of this paper is on a single treatment defined as the first participation in the wage subsidy program with a training component starting within the first unemployment year. Our treatment definition, however, has to take into account the dynamic assignment of individuals into training and the fact that exits out of unemployment towards training or regular employment may occur after different elapsed unemployment durations (see among others Sianesi (2004); Fredriksson and Johansson (2008); Van den Berg and Vikström (2014)). The variation in time until program participation mirrors the caseworker s decision to assign unemployed individuals to the WST program depending on the disposable financial resources as well as on the number of vacancies currently available (Crepon et al., 2009). From an individual perspective, varying starting dates reflect the decision of an unemployed individual to accept a job offer, if there is one, or to continue searching depending on the individual-specific reservation wage. We define treatment and comparison groups in a dynamic setting by looking at the individual-specific treatment status each month during unemployment as long as individuals remain unemployed and are eligible for treatment. This approach has been suggested 16 An important reasoningforthis condition is that WST is mostlikely to startearlierduringunemployment (see table 4), which would therefore make it inappropriate to retain individuals in the evaluation sample who do not have the potential to go through a WST program or even to find a regular, unsubsidized job. 17 The second restriction is required in order to avoid biased results due to some age-specific labor market programs, as for instance for the youth or the elderly. 13

15 by Sianesi (2004) and extensively applied in the treatment evaluation literature (Sianesi, 2008; Fitzenberger and Speckesser, 2007; Fitzenberger et al., 2008; Biewen et al., 2014). Eligibility for treatment refers to the population of unemployed individuals who, at a given point in time, are neither employed nor participating in a public-sponsored training program. Conditional on being unemployed in the previous month, an individual is currently observed to be in one out of four mutually exclusive states: he/she remains unemployed and continues searching, starts a regular unsubsidized employment, starts the wage subsidy program with training, or takes part in another public-sponsored training program assigned by the caseworker in the local employment office. Thus, for each month of elapsed unemployment, program participants are in a subsidized employment, while non-participants might be either regularly employed or remain unemployed and become treated later or never. 18 This setting suggests two alternative comparison groups that are constructed conditional on the calendar time and the elapsed unemployment duration: an unemployed comparison group (UCG) that consists of openly unemployed individuals and an employed comparison group (ECG) including only individuals who find a regular, unsubsidized job at the same time as participants enter subsidized employment. 19 The choice of the comparison group and its appropriateness with regard to the evaluation of WST is briefly discussed in the next section. The sample size for the treatment and both comparison groups used in the empirical analysis are given in table 5. Based on the evaluation sample, figure 2 shows the unconditional probability of participating in the wage subsidy program with training and the program duration over the period from February 1980 to December The unconditional treatment probability is calculated as the number of individuals starting WST in a given calendar month during their first unemployment year relative to the number of unemployed individuals eligible for treatment at the same point in time. In general, the probability to start WST does change over time and is strongly affected by the policy reforms described in the last section. In the early 1980s, treatment probability declines sharply, reaching its minimum 18 Individuals that start another FuU training program, like further training or retraining, at the same time as participants transition into a subsidized employment might be regarded as another comparison group. For reasons of simplicity, however, this paper refrains from comparing the effectiveness of different types of training. 19 Jaenichen and Stephan (2011) apply both comparison groups in the evaluation of an integration wage subsidy for hard-to-place individuals. The main difference compared to our treatment definition is that they use hypothetical starting dates for those individuals that remain unemployed, whereas we define treatment status conditional on the exact elapsed unemployment duration and the calendar month of potential program start for either groups. 20 Note that due to data construction individuals cannot start training in the same month as they enter unemployment. This is the reason why program starts are observed only from February 1980 onwards. 14

16 in the second half of 1984, along with the 1984 reform that prohibited WST by previous employers. After a notable increase during the period in which firms could make use of the program for temporary employment relationships, the probability to start WST slightly declines as a result of the reduction of the maximum subsidy rate from 70% to 50% in The most restrictive policy reform that came into force in 1993 leads to a pronounced drop in the treatment probability that persists up to the abolishment of EPA at the end of Figure 2: Program participation and program duration 1980 to 1997 Program participation Program duration Percent Months m1 1983m1 1986m1 1989m1 1992m1 1995m1 1998m1 1980m1 1983m1 1986m1 1989m1 1992m1 1995m1 1998m1 Unconditional treatment probability Planned duration Actual duration Note: refers to the (potential) program start. Gray vertical lines display the reform years in which the program design has been modified. With regard to program duration, it turns out that wage subsidy had been granted for about 6.5 months on average up until 1992 and for about 4 months thereafter, respectively. The actual subsidy duration behaves in the same way as the planned program duration except for some discrepancies, e.g. at the end of 1982 and 1988, which might be due to some data generation reasons or program participant drop outs. 4 Econometric approach 4.1 Parameter of interest and the choice of comparison group The aim of this paper is to analyze the effectiveness of WST participation on the future labor prospects of participants over time. In particular, our interest is on the calendar time-specific average treatment effect on the treated, which we identify and estimate in a dynamic setting using discrete monthly data. ThetreatmentisdefinedasstartingtheWSTprograminperiodt = {2/1980,...,12/1995}, 15

17 with calendar time t being a deterministic function of month of unemployment start and elapsed unemployment duration m. 21 The binary treatment indicator is denoted by D t (m), where D t (m) = 1 indicates participation in WST starting at time t conditional on mmonthsofunemployment. D t (m) = 0indicatesnon-participationthatencompasses two alternative comparison groups, g = {UCG,ECG}, with g = UCG being the group of unemployed individuals that might be treated later or never and g = ECG being the group of employed individuals that could not be treated at all since they exit the unemployment pool at time t and are no longer considered as eligible for later treatment. Further notation suppresses the index g for simplicity, but everything that follows is conditional on using one comparison group instead of the other. Let Y st (m) denote observed outcome at post-program time s, which, following the potential outcome framework by (Roy, 1951; Rubin, 1974), can be expressed in terms of potential outcomes {Yst 0(m),Y st 1 (m)} as Y st (m) = D t (m)y 1 st (m)+(1 D t(m))y 0 st (m). The parameter of interest is the average treatment effect on the treated of treatment in period t at elapsed unemployment duration m (1) Θ st (m) = E [ Y 1 st (m) D t(m) = 1 ] E [ Y 0 st (m) D t(m) = 1 ]. For a given post-program period, Θ st (m) amounts to the expected difference in the average outcome measure of participants and the average outcome measure they would have if they had not participated in the WST program at calendar time t. Evaluating the impact of WST program participation in a dynamic setting is beneficial compared to a static approach in the sense that we do not condition on future outcomes and future treatments that are likely to confound the true treatment parameter of interest (Sianesi, 2004; Fredriksson and Johansson, 2008). Further, conditioning on t and m simultaneously ensures that we perfectly account for cohort-specific effects and for the fact that individuals might have experienced different event histories between unemployment and potential treatment start. 22 Thus, although similar at the beginning of the unemployment spell, at a given point in time, individuals might be systematically different in terms of both job opportunities and job search efforts (Forslund et al., 2011). A particulary important issue in the context of employer-side wage subsidy evaluation is that of the selected comparison group. The problem, as discussed by Grogger (2008) 21 The sample is restricted by the end of 1995 because of particulary small participant groups in the last two years of the initial observation period (see figure 2). 22 This point has been made by Van den Berg et al. (2014) under a different context. 16

18 and Jaenichen and Stephan (2011), is that starting subsidized employment out of unemployment is a two-part treatment and reflects both getting a job offer and getting the subsidy. As a consequence, using unemployed individuals as a comparison group for WST participants is likely to overestimate the effect of the subsidy on the future labor market prospects of participants (Schönemann et al., 2013). Moreover, results from a placebo experiment carried out by Grogger (2008) suggest huge effects of solely gaining employment on future employment of approximately 35 percentage points. The group of employed comparison individuals, ECG, is very likely to overcome this problem unless WST participants and non-participants gain employment at the same point in time t after having spent m months in unemployment. The interpretation of the treatment parameter of interest, however, does change depending on the selected comparison group. Using similar unemployed individuals as a comparison group refers to the effect of participation in the WST program at time t versus waiting for later treatment, whereas using the employed comparison group simply measures the duration-specific effect of the subsidy conditional on gaining employment (see also Jaenichen and Stephan (2011)) Identification Evaluating the effectiveness of WST program participation is concerned with the identification problem that at time t an individual cannot be observed in both the participation and non-participation state. While the first quantity in equation (1) is directly identified from the data as E[Y 1 st (m) D t(m) = 1] = E[Dt(m)Yst(m)] E[D t(m)=1], the second term represents a counterfactual situation and is not identified without further assumptions. Approximating the outcome of treated individuals in the non-participation state based on the observed outcome of non-participants leads to biased results in general, since observed and unobserved characteristics are likely to affect treatment selection as well as reemployment. This paper builds upon a selection on observables strategy for identification in a dynamic evaluation framework. The underlying idea is that conditional on all relevant factors that jointly determine program selection and potential outcomes up to a certain time period, treatment assignment in t is random and independent of future treatment and labor market outcomes (Sianesi, 2004). More formally, to identify Θ st (m) the following 23 Wolff and Stephan (2013)argue that comparing subsidized workerswith similar unsubsidized workers provides one alternative to measure potential deadweight effects, insofar as subsidized workers could have found their way back into regular employment without the subsidy, but their outcome would then be different. 17

19 two assumptions must hold: (A1) Dynamic Conditional Mean Independence (DCMI): E[Y 0 st(m) D t (m) = 1,X t = x t ] = E[Y 0 st(m) D t (m) = 0,X t = x t ]; (A2) Overlap: p t (x t ) = Pr(D t (m) = 1 X t = x t ) < 1. Assumption (A1) requires that conditional on an appropriate set of observed calendartimespecificcovariatesx t aswell asontheelapsedunemployment durationuptoprogram start, the distribution of the potential non-participation outcome Yst 0 for individuals entering the WST program at time t is the same as the observed distribution of the potential outcome Yst 0 in the group of individuals not entering the WST program at the same point in time. Note that in comparison to Sianesi (2004) or Fitzenberger et al. (2008), the version of the dynamic conditional independence assumption employed in this paper imposes perfect alignment between treated and comparison individuals not only with respect to the amount of time spent in unemployment, but also based on the potential calendar time-specific program start in t. Assumption (A2) prevents X t from being a perfect predictor of treatment assignment and ensures that for any set of the covariates, all treated individuals have a similar counterpart in the non-treated population. A central issue in the evaluation of employer-side wage subsidies relates to the question of whether selection on observables is a credible strategy for identification. While we think that the common support assumption is unlikely to be violated in the empirical analysis due to rich data and particulary large sample sizes of comparison individuals (see table 5), there is a concern with regard to the plausibility of the DCMI assumption (see also the discussion in Jaenichen and Stephan (2011)). The problem is that, other than labor market training programs such as further training and retraining, selection into WST results not only from the interplay between potential participant and caseworker, but requires an existing job opening and a firm willing to hire the particular worker. In such a setting, firm heterogeneity is likely to become an issue once systematic differences in both observed and unobserved factors exist between subsidized and unsubsidized firms. In contrast to Jaenichen and Stephan (2011), however, our data allow us to partly account for observed firm heterogeneity in two ways. First, when estimating the conditional treatment propensity based on using unsubsidized workers as a comparison group for WST participants, we control, among other factors, for firm size and industry structure from the new job. Second, we use indicators on past employer-side experience with the 18