Impact assessment of targeted wage subsidies using administrative data

Transcription

1 Impact assessment of targeted wage subsidies using administrative data Zsombor Cseres-Gergely 1 and Ágota Scharle* Preliminary report. Please do not quote. 20 July 2011 GDN-EI The project aims to assess the impact of wage subsidies for long term unemployed workers in Hungary, using administrative data. While such subsidies are often promoted as an efficient means to speed up recovery or to increase demand for low skilled workers, existing evidence on their employment effects is somewhat mixed, especially in the case of transition economies. We plan to apply discontinuity analysis and a combination of propensity score matching and regression analysis to estimate take-up and the employment and wage effects of two related subsidies and use these estimates to calculate their costs and benefits. In the first phase of the project we have prepared the dataset, identified the treatment and control groups and examined employment outcomes using a discrete-time duration model exploiting a discontinuity in the eligibility requirements (age). The original plan was that we use a dataset derived from the matched administrative records of four institutions which includes information on wages and allows us to reconstruct spells of unemployment and employment before, during and after the subsidy for up to 3.5 years after entering the scheme. This dataset is not yet available. Instead we have used a similar dataset which covers a period of up to 1.5 years after entering the scheme and does not include explicit information on take-up. The project focuses on the following five questions: 1. What is the level of take-up among the long term unemployed? 2. What is the effect of doubling the subsidy for the long term unemployed around the age of 50? 3. What is the effect of the subsidy on reemployment probabilities for the long term unemployed below 50? 4. What is the effect of the subsidy on earnings? 5. Given the estimated employment and earnings effects, is the subsidy cost-effective? The alternative dataset has no data on take-up, which we would need for answering questions 1 and 3, so we can only do a rough analysis of take-up using aggregate level administrative data. At this stage we can present preliminary results on take-up and on the effect of the extra subsidy for those aged over Budapest Institute and IE-HAS, *Budapest Institute 1

2 1. Introduction There is a clear shift in European employment policy towards active labour market programmes (ALMP) as opposed to cash transfers for the unemployed and the recent crisis has further increased the demand for well designed policies. 2 The meta-analyses of existing empirical evidence on ALMPs have shown that the impact of such policies is determined predominantly by their type, while the business cycle or labour market institutions are much less important, or not at all (Kluve 2006, Card et al 2009). Of the four general types defined by Kluve (2006), wage subsidies and services/sanctions proved to be most effective in increasing reemployment rates. The effectiveness of wage subsidies however has been questioned on several accounts. First, not all empirical studies found positive and significant effects. In fact, the few existing papers on transition countries have all shown a neutral or negative impact (Kluve 2006, Betcherman et al 2004). Second, wage subsidies are relatively expensive, which implies that the magnitude of their effect is as important as its size, i.e. only a relatively large impact can make such programmes cost effective. Third, the narrow targeting of subsidies may stigmatise recipients and reduce both take-up and effects (Katz 1996). Fourth, deadweight and substitution costs are likely to be high (Betcherman et al 2004). The policy relevance of wage subsidies has increased with the recent global financial crisis, as a means to preventing the rise of long term unemployment and speeding up recovery. Such subsidies are especially relevant for transition economies struggling to meet EU employment rate targets. Though their share in the labour force is relatively small, the transition has drastically reduced the employment rate of low-skilled workers in Eastern Europe, and the reduction has proved to be lasting in most countries. This has become one of the main causes behind the rise in long term unemployment, economic inactivity and persistent poverty. Wage-subsidies are often proposed as a solution to the unemployment problem among the low skilled. For Hungary, these subsidies may prove to be particularly useful as the uneducated labour force is relatively large and the minimum wage is relatively high (i.e. demand side incentives are likely to be effective). The project aims to contribute to resolving the first three of the above questions. (1) As very few studies have been done in transition countries, it is unclear if the disparity of earlier findings is due to the transition environment or poor policy design. We assess a wage subsidy introduced in Hungary in 2005 and extended in 2007, a relatively well-designed scheme. This scheme is very similar to the targeted payroll tax subsidies in Belgium and Finland, and a targeted tax credit in the US which earlier studies have shown to have some positive impact (Goos and Konings 2007, Huttunen et al 2010, Katz 1996). (2) We use administrative data, which include information on employment and wage as opposed to merely recording exit from unemployment register. This will allow us to compute the costs and benefits of the scheme. Eligibility for the subsidy is determined by observable characteristics of job seekers, so that the treatment group can be easily identified. The administrative data include the prior job search 2 For a recent initiative by influential economists supporting wage subsidies as means of speeding up the economic recovery see: 2

3 history and some information on the expected wages of the target group, which we can use to examine the take-up decision. The paper will examine a job subsidy for disadvantaged job seekers in Hungary. It is a quasivoucher scheme that offers a temporary reduction on payroll tax (social security contributions) to employers hiring the holder of the voucher. The scheme was first introduced in 2005 for school leavers and extended to various subgroups among the long term unemployed in The reduction is largest for job seekers aged over 50 and those with only primary education. The scheme covers other disadvantaged groups as well, but those are not examined in this project. The rest of this preliminary report is divided into six main sections. The next section briefly reviews existing relevant research on wage subsidies. Section 3 describes the Hungarian scheme in detail and summarises available data on take-up. Section 4 describes the originally planned dataset and, in more detail, the alternative dataset we use for the analysis presented in section 5. The last section outlines plans for completing the report. 2. Review of existing research on similar policy instruments Similar targeted payroll tax subsidies have been used in Belgium, Finland, the Netherlands and the US. The Maribel subsidies system in Belgium in the late 1990s offered a lump sum reduction in payroll taxes for employing manual workers, so that the relative size of the subsidy was highest for low wage workers. Goos and Konings (2007) use firm-level data to evaluate the effects of changes in this scheme and find significant positive effects on employment. Huttunen et al (2010) estimate the employment effects of a subsidy in Finland, which is targeted at the employers of low-wage older workers using a difference-in-difference-in-differences. They find that the subsidy has no effects on the employment rate but it increases the probability of part-time workers obtaining full-time employment (the scheme is only available to full time workers). Katz (1996) reviews evaluations of the Targeted Jobs Tax Credit (TJTC), a similar scheme operating in the US until 1994 and concludes that it was effective in improving the earnings and employment of disadvantaged groups, especially when combined with training elements. However, studies examining changes in the TJTC rules or its performance compared to more sophisticated schemes in experiments have pointed to a potential problem with narrow targeting. When TJTC was used without additional services (training or counseling), take up tended to be low and reemployment effects were small (or even negative in some experiments). These results were attributed to the stigmatisation of recipients. However, some of these concerns were called into question by later analysis showing that non-random selection into the treatment group could explain poor performance. Dubin and Rivers (1993) recalculate the effects of one such experiment in Illinois, where the treatment group were long term unemployed individuals who were offered a voucher, which their new employer could submit and receive a lump sum payment. They find that, once controlling for self-selection, the programme significantly increased the likelihood of re-employment. There are very few empirical studies on Hungarian ALMPs and as far as we know, the reemployment effect of the Hungarian START scheme has not yet been evaluated. There are two papers that evaluate the impact of traditional wage subsidies ( bértámogatás and bérköltség támogatás ). These provide funding for most of the wage costs for a maximum of one year and require continued employment for an extra period of time equivalent in length to the 3

4 supported period. Both evaluations examined a wage subsidy scheme in comparison with other ALMPs using administrative and survey data. In O Leary (1998), the treatment group was drawn from official recorded outflows from ALMPs in the second quarter of 1996 and were interviewed 12 months later, in Using various modelling strategies (unadjusted-, matched pairs-, OLS regression adjusted- and OLS regression adjusted with full interaction comparison of outcomes), the subsidy was found to increase employment probability between 17 to 24 %points if unadjusted, -10 to 0 if adjusted or used matching. A significant increase was found in earnings on the first job, but a negative impact of the survey date in all specifications. In the interpretation of O Leary (1998), results suggest that employers are very selective in choosing who to hire with the subsidy, so that the impact of the scheme is plagued by skimming and deadweight loss. Csoba et al (2010) estimate the treatment effect of training, wage subsidy and public works schemes. Control and treatment groups are taken from the unemployment register, identifying participants and similar non-participants between September 2009 and February The sample is interviewed between August and September 2010 and the outcome variable is employment at the time of the survey. They use a logit model comparing control and participant groups with regressors include sex, age, education, Roma origin, legth of unemployment in pior work history and place of residence. Results show significant positive effects attached for program participation dummies and negative effects for indicators of past unemployment. However, the detailed regression output is not presented in the paper so the validity of this result cannot be assessed. 3. The policy instrument and the context of its introduction The new Hungarian wage subsidy scheme examined in this project was first introduced in October 2005 for school leavers (START) and extended to various subgroups among the long term unemployed in July 2007 (START plusz and extra). It is a quasi-voucher scheme that offers a temporary reduction on payroll tax (social security contributions) to employers hiring the holder of the voucher. As opposed to the traditional subsidies mentioned above, there is no obligation to continue the employment of subsidised workers once the subsidy expired. The amount of the subsidy varies across eligible groups, as summarised in the table below. 4

5 Table 1. Rules of the various START schemes at the time of introduction Name Eligibility Amount of subsidy (% of employers contribution)* Ceiling on subsidy *** Date of introduction 1st year 2nd year START START Plusz School leavers: below 25 (30 for graduates), no prior paid job On parental leave or care allowance, or registered unemployed for 12 months within preceding 16 months, not eligible for old age pension 53 % HUF / month 22 % HUF /month 1.5x min wage (2x for gradutes) Oct x min wage July 2007 START Extra Over 50 or primary education only, and registered unemployed for 12 months within preceding 16 months, not eligible for old age pension 100 % HUF /month 53 % HUF /month 2x min wage July 2007 Receiving unemployment assisstance, not eligible for old age pension Jan * In 2007, the employer s contribution was 32% of the gross wage, and this was waived in full during the first year of employing a person with a START extra voucher. The flat rate health contribution was waived in both years in all schemes, which was 1950 HUF a month (about 8 EUR), or around 3% of the minimum wage. Source: Cseres-Gergely and Scharle (2009) The subsidy is largest for job seekers aged over 50 and those with only primary education: 100 % of the employers social security contribution in the first year and 53 % in the second year, with a cap set at twice the minimum wage. The general subsidy available to all long term unemployed is 53 % in the first year and 22 % in the second year. The scheme (all three variants) has been administered by the tax authority who issue a plastic card to eligible persons which indicates the type and eligibility period of the subsidy. Cards are issued only if claimed, but the evaluation of claims is automatic, with no discretion or further conditions beyond age, education and long term unemployed status. Job centres have been actively encouraging job seekers to claim the card. The validity of the card and thus the period of eligibility starts on the day of issue. Job seekers are therefore advised to claim the card immediately before starting in their job, so that their employer may be eligible for the maximum length of the subsidy. Between July 2007 and March 2009, the START extra card was claimed by persons, and issued to 9660 persons. 3 Less than 2 % of the claims were declined by the tax authority, and some 3 We have not managed to obtain data on claims for the period ending December 2008 yet. 5

6 5 % was not issued for other, unknown reasons (e.g. the card holder withdrew the claim). Between July 2007 and December 2008, 6109 persons held a valid START extra card. During the same period, the number of persons employed with the subsidy started to grow more rapidly after February 2008 and peaked at 4998 in November 2008 (Figure 1). This suggests that most cards have been claimed once the job seekers had a job offer, as recommended by job centres. Figure 1 Number of persons employed with a START subsidy 2007 and December plusz extra Dec-08 Nov-08 Oct-08 Sep-08 Aug-08 Jul-08 Jun-08 May-08 Apr-08 Mar-08 Feb-08 Jan-08 Dec-07 Nov-07 Oct-07 Sep-07 Aug-07 Jul-07 Source: Aggregate administrative data by the tax authority, prepared on request from the Ministry for Labour The total cost of the scheme is relatively modest, compared e.g. to re-training or public works programmes. Between July 2007 and December 2008, the plusz and extra schemes cost a total of 2.8 billion HUF (or an annual 1.8 billion HUF, about 7.2 million EUR) in lost revenues to the social security fund. Of this, annual 0.8 billion went to START extra employers and 1 billion to START plus employers. 4. Data As noted in the introduction, we do not yet have the dataset we intended to use in the project. In this section, we briefly describe the data we are hoping to obtain and explain in more detail the alternative data we currently have access to (as this is what the ensuing analysis is based on). The dataset originally planned The project will use a purpose-built dataset derived from the administrative records of four institutions. It covers information on age, sex, dates of entering and exiting employment, earnings (pension insurance records); on type of subsidy claimed and date of submitting claim (tax authority issuing the voucher), prior employment and unemployment history (unemployment register), and finally, the branch of activity and size of the employer (company register). 6

7 The dataset is based on a random sample of working age adults taken from national insurance records that cover the whole population. This is matched with the records of other institutions at the individual level using one-way irrecoverable identifiers in order to preserve anonymity. Employment spells are recorded for the period between June 2007 (launch of the extended subsidy) and December This allows us to observe entrants to the scheme for up to 3.5 years and post-treatment effects on job retention and earnings for up to 17 months. For those entering the scheme before July 2008, we can also observe post-treatment job retention for 6-17 months. The latter group will be relatively small (and increasingly so as we increase the months observed after expiry of subsidy), so there will be a trade-off between sample size and the capturing of longer term effects. The building of the dataset has been commissioned by the Ministry of Labour and Social Affairs responsible for the implementation of the scheme, and is managed under the Public Sector Information Programme of the von Neumann Digital Library and Multimedia Centre Kht. The panel covering was scheduled to be released in the second quarter of 2011 but this has been delayed due to the major reorganisation of the ministerial structure initiated by the new government in Ministry officials have confirmed that the data will at some point be constructed and released, but for the moment they cannot tell when. The alternative dataset The alternative dataset is also drawn from administrative records, constructed for research purposes by the Institute of Economics, Hungarian Academy of Sciences (IE-HAS). It contains the matched records of 50 % of the adult population including data from (1) health and (2) pension insurance, (3) the treasury, and the (4) unemployment register. 5 The sample was taken of the archives of January 2002 of the health insurance fund which include all people residing in Hungary who ever had been covered by public health insurance. This practically covers the total population. In the original dataset, data on employment, unemployment and transfer receipt are recorded in spells of varying length, with type of spell and dates of starting and ending of each spell. This was converted into a monthly structure. Labour market status is determined on the basis of all the available information. Information on current labour market status and transfer receipt is available for the period between January 2002 and December There is information on age, sex, dates of entering and exiting employment, earnings (pension insurance records), unemployment history and type and period of receiving various transfers (including disability benefits) and sick leave. There is no data on claiming the START cards (which is recorded by the tax authority), or on the employer. The information on employment and transfers is used to reconstruct labour market status in each month during the observed period. It must be noted that this reconstructed status may not 4 5 This involved merging the administration of economic, financial and employment policies in one super-ministry and has disrupted practically all the projects started in Figure x Death rate by sex and age cohort in the total population (secondary education) For a detailed desciption of the data collected by the Pension and the Health Insurance Funds and in the Treasury, see Elek et al (2008). 7

8 correspond to the actual situation in the economic sense. A person receiving unemployment benefit is coded as unemployed, even if not actively looking for a job, while a person receiving old-age pension is coded as inactive, even if available for work and actively looking for a job. Also, in some cases statuses are not mutually exclusive. For example, a person working on a community work scheme is typically recoded as employed in the official records, and thus will be coded as employed, but at the same time may appear in the unemployment register as well. In such cases, the cleaned dataset will indicate the second state in separate variable (as a secondary or a third state). Long term unemployed individuals potentially eligible for the START vouchers are identified on the basis of their records in the unemployment register. Though we have access to the raw data provided by the Employment Office, we used the dataset cleaned by IE-HAS, which is much easier to handle, but is slightly less accurate as changes within the month are not fully documented. If labour market status changed during the month, or the person had parallel states during the month (see example above), this is coded as secondary status in a separate variable. To take account for this, we used three definitions of varying accuracy. In the strictest one, eligible long-term unemployed are individuals with at least 12 pure months of registered unemployment with no secondary state recorded for any of the 12 months (but may have such mixed months in the rest of the period). The second definition allowed pure and mixed spells as well, while the third one only mixed spells. In June 2007, we could identify 5053 persons who would be considered long term unemployed by one of our definitions and 4436 (88%) of them would meet the strictest criteria as well. The rest of the analysis is restricted to this group. The outcome variable is constructed in a more inclusive manner: any paid job is considered as employment, whether recorded as a primary, secondary or third status in a month. However, casual jobs are excluded as such jobs are not eligible for the START subsidy. There is also some attrition in the dataset. Although the data are based on administrative records, the nature of the administrative sources itself may generate attrition: we lose information on people who leave the unemployment register and take up undeclared work, or become inactive without obtaining any form of social transfer. Finally, deaths are not recorded in the data, as for some reason this was considered to improve anonimity. Given all these potential sources of loss, attrition does not seem particularly large in the age groups we examine (see Figure 2 below) and it is not very different between the treatment and control groups. 8

9 Figure 2. Attrition in the treatment and control group, July 2007-Nov treatment control However, mortality starts to increase relatively fast over the age of 40, especially for men, which would generate faster attrition in the treatment group and a difference in the composition of the control and treatment groups over time (Figure 3). Also, as death is likely to be preceded by illness, it is not unrelated to the probability of employment, so we plan to explore this further. Figure 3 Mortality rate by sex and age-group men women Source: National Statistical Office, Hungary. Averages for

10 5. Identification strategy, estimation method and preliminary results As explained above, the START scheme offers a 53 % subsidy ( plusz ) on hiring any long term unemployed job seeker and a 100 % subsidy ( extra ) on hiring a long term unemployed job seeker aged over 50 in the first year of employment. The first scheme envelopes the second in the sense that everyone eligible for the second one is eligible for the (less generous) first one as well. Therefore, the cutoff in the eligibility conditions can only be exploited to identify the additional effect of the START extra scheme. Our strategy is based on the assumption that individuals slightly younger than 50 are not eligible to the extra subsidy, but are not different in any unobserved way affecting employment chances. Given this setup, the treatment effect is identified as T = E(I P=1) E(I P=0) that is as a simple difference between the outcomes of the treatment and the control group, where I is the outcome indicator and P is the indicator of programme participation. In order to control for the observable differences between the two groups, we can use the conditional version of these statistics to obtain T' = E(I P=1,X=x1) E(I P=0,X=x0) To implement this idea, we use regression-discontinuity analysis to identify the extra effect. Although the discontinuity makes the effect identifiable, the reliability of our estimates depends on the definition of the treatment and control groups, which takes some discretion. When defining the groups slightly below and slightly above 50, we are facing a trade-off. Opening up the age-windows and making the groups larger, we are obtaining more observations and hence more precise estimated effects. At the same times, the estimates become more prone to agerelated effects that might be correlated with our outcome a familiar trade-off between consistency and variance. Based on trial and error, we settled at using a 3 year window for the control group and 2.5 years for the treatment group with an 18 month gap between them in order to ensure that no member of the control group becomes eligible for participation during the observed period. As jobs seekers with primary education are eligible for START extra irrespective of their age, they are excluded from the analysis. Descriptive statistics for the treament and control groups are presented in Table 1 in the Appendix. We define the control group as those eligible for the START extra voucher except for the age condition and treatment as being eligible for the voucher, rather than as holding the voucher. This is not only a pragmatic decision taken for the lack of better data, but can also be justified on the basis of the official information on claims, take-up and subsequent employment presented in section 3 above. As we have seen, the issuing of vouchers is almost automatic and most claims are shortly followed by employment. There is a strong incentive for that, as the eligibility period starts immediately on the date of issuing the voucher. This suggests that claiming the voucher is a formal exercise, which in reality may happen after the outcome, that is, after the employer decided to hire the job seeker. True take-up is knowledge of the scheme and of the age condition by either the employer or the job seeker. As the scheme was actively advertised in job centres, it does not seem far-fetched to assume take-up to be close to full and to consider all job seekers above 50 as treated. The assumption can be cross-checked by comparing job exits of card holders to eligible non-holders when we obtain the originally planned dataset, or by comparing 10

11 the aggregate statistics on employment with the subsidy to the total of observed job exits in the relevant age groups in our data. The effect of the scheme on employment may take several forms. We have considered the timing of the first exit to job, the total number of employment spells in the 18 month after the scheme was introduced, and the share of persons employed at various points in time. Raw data presented in figures 4 and 5 suggest a modest treatment effect, especially for women. Moreover, the difference in exit rates for the total of the period (table 2) is zero for men and negative for women. At first glance the only hopeful sign of a positive treatment effect is that men in the treatment group spent slightly more time (2.93 months) employed during the 18 months of the treatment period than the control group (2.63 months). We also have information on wages, which we plan to examine in the next phase. Figure 4. % employed in treatment vs control group from Feb 2006 to Dec 2008, by sex (vertical line marks July 2007, when the treatment period started) MEN observed period (months) treatment control 11

12 WOMEN observed period (months) treatment control Note: t=17: July 2007, t=22: Dec 2007 Figure 5 Job exits by age cohort and sex (six-month cohorts, age 45-52) women men Note: All exits during observed treatment period, with no adjustment for attrition. Cohort 2 is aged between exactly 45 and 45 and 6 months, etc. The three middle cohorts are excluded because they gained eligibility for the subsidy during the observed treatment period. 12

13 Table 2. Exits to job / survival in unemployment and employed spells after July 2007 Men Stays (N) Exits (N) Exits (%) Total (N) employed spells (month) Control Treatment Women Control Treatment However, the treatment and control groups span across seven years of age, which may blur the treatment effect if years of age have a large negative impact on re-employment probabilities, as has been found by most studies on the reemployment of long term unemployed. As Figure 6 below shows, this is likely to apply to men. In the top right panel of the figure we compare the nearist two age cohorts and find visibly higher exit rates in the treatment group (aged 50) than in the control group (aged 48) after January In the top left panel, we compare the two cohorts widest apart and find that the control group (aged 45) has higher exit rates for all of the period than the treatment group (aged 52). There is no such clear cut pattern in the case of women. Figure 5. Cumulative exits for age cohorts apart and near MEN time time aged 52 (t) aged 45 (c) aged 50 (t) aged 48 (c) 13

14 WOMEN time time aged 52 (t) aged 45 (c) aged 50 (t) aged 48 (c) Note: (c) control group (t) treatment group Next we use regression analysis to test if the treatment speeds up exits to job in the treated group, controlling for observed differences between the treatment and the control group. Given the monthly information on labour market status, we examine job exits (survival in unemployment) in a discrete time duration model 6 following Jenkins (1995), by estimating a regression model for a binary dependent variable defined as follows. The unit of analysis here is the spell month and a dummy variable is defined for each observation indicating continued non-employment (0), or exit to a job (1). The hazard function is logistic in this case and the model likelihood has the same form as that for a standard binary logit regression model. Note that the separability in the loglikelihood function ensures correct estimates even in the case of a competing risk, such as death in our case. Regressors include personal characteristics and labour market history. Table 3 below summarises the results of various specifications. Table 3. Probability of survival in unemployment / exit to job (logit) for men Coef. dy/dx Coef. dy/dx Coef. dy/dx Coef. dy/dx (1) (2) (3) (4) treat 0.491** * * * age age other vars: all edu, history edu Notes: **significant at 1% * significant at 5%, unit of observation = month*person For details see table A.2. in the Appendix 6 If t denotes the passage of calendar time (quarters), the distribution of durations is modelled via the probabilities of ending a spell at each value of t. There is a one-to-one relationship between these probabilities and the probabilities of having completed spell durations of different lengths. 14

15 We find a significant positive effect for men (positive but not significant for women) after controlling for age, education and participation in training (in the 2 years preceding the treatment). As expected, age is important to control for there is no treatment effect when age is ommitted from the model. 6. ext steps In the second phase of the project we plan to (1) refine the analysis presented above, using the data we currently have and (2) complete the original research plan using the data originally planned. We shall ask for a six month extension of the project deadline and hope to be able to confirm if and when the originally planned data will be available in time for us to do the analysis before the extended deadline. As concerns the analysis we have done so far, we see the following issues and options that need to be further explored: - explore if claiming the card typically follows the job offer and any signs of a stigma effect - explore if attrition is selective and handle it accordingly - test model specification, especially if log-odd ratios are truly a quadratic function of age - estimate separately for high ed and secondary (primary are excluded) - experiment adding further job history variables to logit - use other measures of employment outcomes e.g. job retention as LHS - use a broader measure of employment (including casual work) as outcome - examine wage outcomes (question 4) - use a narrower measure of employment (excluding public works as much as possible since it is not explicitely recorded as such) - test wider definitions of treatment group. We also plan to further explore how take-up can be best measured and the level of the best available indicator, using aggregate level administrative data and the currently available dataset (question 1). Once we have estimated employment and earnings effects, we can also calculate whether the extra subsidy is cost-effective (question 5). 15

16 1. References Betcherman, G., K. Olivas and A. Dar (2004): Impacts of Active Labor Market Programs: New Evidence from Evaluations with Particular Attention to Developing and Transition Countries, World Bank Social Protection Discussion Paper, No Card, David, Kluve, Jochen and Weber, Andrea (2009): Active Labor Market Policy Evaluations: A Metaanalysis. The Austrian Center for Labor Economics and the Analysis of the Welfare State Working Paper No. 0902, February Cseres-Gergely, Zs. and Scharle, Á. (2009): A START programok értékelésének lehetőségei adminisztratív adatok felhasználásával [Feasibility study of evaluating the START programmes using administrative data] research report, Budapest Institute the_start_wage_subsidy_programmes Csoba, Judit Nagy, Zita Éva Szabó, Fanni (2010): Aktív eszközök, munkaerő-piaci programok kontrollcsoportos, többváltozós értékelése [Evaluation of active labour market programmes with control groups] Elek, P. Osztotics, A. Scharle, Á. Szabó, B. Szabó, P. A. (2008): Az OEP, az OnyF és a MÁK kezelésében lévő adatbázisok [The administrative data of the Health and Pension insurance funds and the Treasury], MTA-KTI, Insitute of Economics Dubin, J.A. and Rivers, D. (1993): Experimental Estimates of the Impact of Wage Subsidies, Journal of Econometrics 53, Goos, Maarten and Konings, Jozef (2007): The Impact of Payroll Tax Reductions on Employment and Wages: A Natural Experiment Using Firm Level Data, LICOS Discussion Papers 17807, Centre for Institutions and Economic Performance, K.U.Leuven. Jenkins, S. P. (1995): Easy Estimation Methods for Discrete-Time Duration Models, Oxford Bulletin of Economics and Statistics 57(1): Huttunen, Kristiina, Pirttilä, Jukka and Uusitalo, Roope (2010): The Employment Effects of Low-Wage Subsidies, IZA Discussion Paper No. 4931, May 2010 Katz, Lawrence F. (1996): Wage Subsidies for the Disadvantaged, NBER Working Paper No. W5679, July 1996 Kluve, J. (2006), The Effectiveness of European Active Labour Market Policy, IZA Discussion Paper, No. 2018, Bonn. O Leary, Cristopher J. (1998): Evaluating the Effectiveness of Active Labor Programs in Hungary, Upjohn Institute Technical Reports Papers yet to be included in the literature review Bell, B., Blundell, R., Van Reenen, J. (1999): Getting the Unemployed Back to Work: The Role of Targeted Wage Subsidies, International Tax and Public Finance 6, Boockmann, B., Zwick, T., Ammermüller, A., Maier, M. (2007): Do Hiring Subsidies Reduce Unemployment Among the Elderly? Evidence from Two Natural Experiments, ZEW-Discussion Paper Brouilette, D., Lacroix, G. (2008): Heterogeneous Treatment and Self-Selection in a Wage Subsidy Experiment, IZA Discussion Paper

17 Brown, A.J.G., Merkl, C., Snower, D.J. (2006): Comparing the Effectiveness of Employment Subsidies, Kiel Working Paper Burtless, G. (1985): Are Targeted Wage Subsidies Harmful? Evidence from a Wage Voucher Experiment, Industrial and Labor Relations Review 39, Calmfors, Lars. (1994): Active Labour Market Policy and Unemployment - A Framework for the Analysis of Crucial Design Features, OECD Economic Studies 22 pp Eichler, M.and M. Lechner. (2002): An evaluation of public employment programmes in the East German State of Sachsen-Anhalt" Labour Economics 9 (2002): Gerfin, M., M. Lechner and H. Steiger. (2006): Does subsidised temporary employment get the unemployed back to work? An econometric analysis of two different schemes. Labour Economics 12 (6) (2006): Göbel, C. (2006): The Effect of Temporary Employment Subsidies on Employment Duration, Départment des Sciences Économiques de l Université catolique de Louvain Discussion Paper Heckman, James J., Robert J. Lalonde and Jeffrey A. Smith (1999): The Economics and Econometrics of Active Labor Market Programs. In Orley Ashenfelter and David Card, editors, Handbook of Labor Economics, Volume 3A. Amsterdam and New York: Elsevier, 1999: Imbens, Guido and Jeffrey M. Wooldridge (2008): Recent Developments in the Econometrics of Program Evaluation. IZA Discussion Paper #3640. Bonn: Institute for the Study of Labor Jaenichen, U., Stephan, G. (2009): The Effectiveness of Targeted Wage Subsidies for Hard-to-Place Workers, Applied Economics. Jahn, E.J., Wagner, T. (2008): Do targeted hiring subsidies and profiling techniques reduce unemployment?, IZA-Discussion Paper Jespersen, Svend T., Jakob R. Munch and Lars Skipper (2008): Costs and Benefits of Danish Active Labour Market Programmes. Labour Economics 15 (2008): Dubin, J.A., Rivers, D. (1993): Experimental Estimates of the Impact of Wage Subsidies, Journal of Econometrics 53, Kluve J., H. Lehmann and C.M. Schmidt. (1999): Active Labor Market Policies in Poland: Human Capital Enhancement, Stigmatization, or Benefit Churning? Journal of Comparative Economics 27 (1): Kluve, J. (2006), The Effectiveness of European Active Labour Market Policy, IZA Discussion Paper, No. 2018, Bonn. Kluve, J., D. Card, M. Fertig, M. Góra, L. Jacobi, P. Jensen, R. Leetmaa, L. Nima, E. Patacchini, S. Schaffner, C.M. Schmidt, B. van der Klaauw and A. Weber (2005), Study on the effectiveness of ALMPs, report prepared for the European Commission, DG Employment, Social Affairs and Equal Opportunities, Essen. Lubyova, M. and J.C. van Ours. (1999): Effects of active labor market programs on the transition rate from unemployment into regular jobs in the Slovak Republic. Journal of Comparative Economics 27(1999): Mueser, P.R, K.R. Troske and A. Gorislavsky (2007): Using State Administrative Data to Measure Program Performance. Review of Economics and Statistics 49(4) pp Rodriguez-Planas, Nuria and Jacob Benus (2010): Evaluating Active Labor Market Programs in Romania, Empirical Economics, 38(1) pp Stephan, Gesine (2009) Employer wage subsidies and wages in Germany. IAB Discussion Paper 9/2009. Terrell, K. (1999): Labor Market Policies and Unemployment in the Czech Republic. Journal of Comparative Economics 27, pp

18 Appendix Table A1. Summary statistics for the treatment and control group for June 2007 Treatment Control Variable Obs Mean Std. Dev. Obs Mean Std. Dev. sex age (years) lower secondary education upper secondary education tertiary education student (months) July06-June training before July07 (dummy) over 50% disabled (dummy) sick leave (ever b July07) labour income and transfers reservation wage (HUF/month) current income (HUF/month) smallholder (ever before) disab benefit months July06-June UB months July06-June UA months July06-June disab benefit months b July UB months b July UA months b July Labour market participation months worked July06-June casual work July06-June months worked b July casual work b July student (months) b July Notes: Retrospective data cover the period starting in January 2002, except for transfers, where we have data from Jan b=before 18

19 Table A2. Coefficients and marginal effects in various model specifications of exit to job probabilities (logit) for men Coef. dy/dx P>z Coef. dy/dx P>z Coef. dy/dx P>z Coef. dy/dx P>z treat age age lowsec upsec upsec upsec tert ter training month month month month month month month month month worked J06-J unemp J06-J worked b J constant Log likelihood Pseudo R