The Effects of Employment Tax Credits: Evidence from Spain

Transcription

1 The Effects of Employment Tax Credits: Evidence from Spain Ferran Elias 1 Columbia University September 16th, 2014 JOB MARKET PAPER Abstract This paper investigates the impact of employment credits on employment and earnings for young and senior workers. Using administrative data and exploiting both age discontinuities and policy changes, I find that the effects of employment credits differ greatly at distinct moments of the life cycle. The youth employment credit raises young employment and earnings by 1% and.8%, respectively. I show that the gains do not come at the expense of non-subsidized workers, and can thus be interpreted as net job creation. Conversely, senior employment credits do not increase employment or earnings, involve substantial displacement by retiming the entry into permanent jobs, and act as a transfer to employers. I explore several channels to explain the results. Employment credits increase young employment because they double the number of transitions into stable jobs and reduce the time spent on a temporary contract before becoming permanent by 18.9%. On the other hand, senior workers face an inelastic labor demand. One main reason why firms might not want to hire them is because they know they are likely to retire early: seniors marginally attached to the labor force are 13.5%-16.5% more likely to retire by claiming unemployment insurance (UI) thanks to a policy threshold at the age of 52. Finally, I evaluate the cost-efficiency of the policy. Despite the presence of windfalls (46%-76% of young subsidized contracts; 94% of senior ones), the youth policy is cost efficient, mainly because it decreases UI claimants by 3.82%, and actually generates overall net revenue of 4,220,709 euros. 1 I want to thank Ethan Kaplan, Wojciech Kopczuk and Bentley MacLeod for their help and guidance at all stages of this paper. Chris Boone, Davide Crapis, François Gerard, David López-Rodríguez, David Munroe, Olivia Nicol, Pablo Ottonello, Nicolás de Roux, Bernard Salanié, Oriol Vallés, and seminar participants at Columbia University provided helpful comments. I also want to thank Almudena Durán and Kathleen Dreyer for help accessing data. All resulting mistakes are my own. fe2139@columbia.edu 1

2 1 Introduction How can we improve the performance of the labor market? The recent weak recovery after the Great Recession has raised the interest in the role that employment credits might play to contain the risk of rising structural unemployment by bringing job losers back to work as quickly as possible (OECD, 2012) out of 29 OECD countries have implemented employment credits since the onset of the Great Recession, with many of them targeting groups considered the most disadvantaged such as young and senior workers (OECD, 2009, 2013). 3 Despite the increasing importance of these policies, it has proven very challenging to compellingly establish that employment credits can increase overall employment, and that they do so in a cost-efficient way. There are two main reasons for that. The first difficulty arises from the fact that employment credits might generate substantial displacement effects in non-treated groups (Davidson and Woodbury, 1993). Noneligibility for an employment credit might decrease the probability of finding a job for workers in the control group. These effects are difficult to capture both in randomized trials and natural experiment studies, since their estimates are based on the difference between treatment and control groups (Crepon et al., 2013). Therefore, the estimates in the literature are difficult to interpret in terms of net employment creation (Woodbury and Spiegelman, 1987; Egebark and Kaunitz, 2013; Kugler et al., 2005). The second difficulty relies on establishing the extent to which employment credits generate windfalls, or subsidize employment that would have been created absent the policy (Katz, 1998; Becker, 2011). Windfalls thus can impose large deadweight losses, and are a crucial estimate for a convincing cost-benefit analysis. This paper confronts these difficulties by taking advantage of several employment credit policies in Spain. In May 1997, the Spanish government enacted a reform expanding several employment credits policies. Firms could benefit from payroll tax credits 2 See The Wall Street Journal, In Europe, Job Protections for Older Generation Are Barriers for Younger Workers, August 8th, for a discussion on how payroll taxes and employment protection make employers reluctant to hire young workers in Spain, Italy and Greece. 3 There are 30 OECD countries, but Iceland did not provide information regarding their changes in labor market interventions. 12 countries have provisions that target young or old workers. Belgium, France, Japan, Portugal, and the US have specific policies both for young and old workers. Canada, Greece, Italy, South Korea, Turkey and the UK have specific policies for young individuals. Poland has an employment credit for old workers. 2

3 when hiring workers as new permanent employees. The program featured age discontinuities that made it easier to hire with a tax credit workers before their 30th birthday and after their 45th birthday because they did not need to have been registered as unemployed for more than a year. For young workers, the tax credit decreased employers payroll tax rates by 40% for two years. The reduction for new old permanent hires was more generous, with payroll tax rates being 50-60% lower for the whole duration of the contract. In March 2001, the youth employment credit was removed, providing an additional natural experiment. I use a very rich administrative dataset that contains the labor lifes of over 1 million of individuals. For each worker, I can reconstruct his entire labor life since he joined the workforce: I know the day when he begins and ends each employment or unemployment spell, his wage, his employer and the location of the job, the type of contract and whether it is subsidized or not, the reason of the separation, etc. The data also includes personal information of the worker: date of birth, sex, education, citizenship and place of birth, etc. The fine-grained level of detail of the dataset is not only important to provide compelling causal evidence of the effects of the policy on treated individuals, but also to identify the negative aspects of employment credits such as windfalls or displacement of non-treated workers, that have been elusive in previous empirical research. The analysis delivers four main sets of results regarding employment, earnings, the mechanisms through which the tax credit operates, and cost-efficiency. First, I exploit the introduction and removal of the policy to explore its short-term effects on employment through a difference-in-difference analysis. I find that youth employment raises by.77% relative to its pre-1997 level. About half of the effect comes through a reduction in the number of unemployment insurance (UI) recipients. In contrast, senior employment decreases by.1%. Since the difference-in-difference estimates might underestimate the treatment effects because of optimization frictions (Chetty et al., 2011; Chetty, 2012; Kleven and Waseem, 2013), I use the age discontinuities through a regression discontinuity design (RDD) and a regression kink design (RKD) to explore the long-run effects. The results show that young employment was rising by 1% in the long-run, an estimate that is 30% higher than the short-run one. Conversely, the long-run analysis for senior workers reveals a null effect on employment and substantial displacement: firms hoard workers aged in short-term positions, and delay their conversion to permanent until they cross their 45th birthday. Consequently, the senior employment credit acts as 3

4 a transfer to employers. The key threat to interpreting the positive young employment effect as job creation is that the estimates might be confounding positive and negative effects. In March 2001, the only change at 30 was the removal of the youth employment credit. This allows me to explore potential displacement effects in three ways: first, between the age distribution of hiring features a jump at the threshold. Once the policy is discontinued, we would expect that the jump disappears because hiring above 30 moves upwards, while hiring below 30 moves downwards. Yet, both visual and regression results show that the jump disappears only because hiring below 30 converges to the level of hiring above 30, and that the latter stays constant. Second, individuals born in March 1971 reach their 30th birthday once the youth employment credit is removed. Thus, they cannot suffer from displacement and their employment evolution after 30 should be more positive than that of cohorts that crossed their 30th birthday when the policy was in place (i.e. born in March 1970). However, their probability of employment after 30 is not significantly different than that of previous cohorts. Finally, cohorts that crossed their 30th birthday before March 2001 should experience a positive rebound in employment in March Again, I do not find any positive and significant estimate. Overall, the three strategies are consistent with the youth employment credit not creating displacement effects, and confirm the interpretation of the youth employment effect as net job creation. In the second main set of results, I show that there are no effects on the wages of new permanent young hires. However, youth earnings increased by.8% after 1997, consistent with their level of employment increasing. Results exploiting the 2001 reform indicate the same effect. Conversely, wages of new senior permanent hires decrease by a significant 2.14%, a result that suggests that the supply of old workers increased as a consequence of the policy, and that senior labor demand was inelastic. Overall senior earnings are also negatively affected though the coefficient is not significant. In the third set of results, I explore several mechanisms that could explain the differential effect of the policy at each age discontinuity. The youth employment credit is successful in increasing employment because it improves the transition process from unstable to stable jobs for young workers. The number of conversions from short-term to permanent doubles, and the length of short-term spells before a transition occurs decreases by 18.9%. 4 Moreover, subsidized permanent contracts have longer duration than 4 Compared to other OECD countries, youth transitions to a stable job take longer in Spain (Dolado 4

5 short-term contracts, and their rate of separation during the first 12 months is very similar to that of non-subsidized permanent contracts, confirming that youth employment credits increase job stability. For the case of senior workers, the evidence suggests that the interaction between the UI system and early retirement might explain the null employment effect. It has been shown that deferred wage compensation as in Lazear (1979) deters firms from hiring older workers (Hutchens, 1986, 1988; Hu, 2003) because they are too close to retirement. This is an important concern in Spain because workers over 52 can claim UI benefits very easily. I show that between the ages of 52 and 60, UI recipients increase from 3% to 11%. Moreover, workers who had been hired with a tax credit after their 45th birthday are 13.5%-16.5% more likely to claim UI between the ages of 52 and 60. Thus, firms might not be willing to increase senior employment because they do not consider worthwhile to invest on them given their higher chances of early retirement. 5 Finally, to evaluate the cost-efficiency of the policy I need to know how many of the subsidized contracts would have existed in any case. Since the data identifies both subsidized and not subsidized contracts, I can repeat the difference-in-difference analysis to measure windfalls. After 1997, 76% of the jobs would have existed otherwise. For senior workers, windfalls are even larger and account for 94.4% of the subsidized jobs. However, in 2001, windfalls of young workers had fallen to 46.22%, probably as a result of the enactment of new limitations in the use of employment credits. 6 Despite the large windfalls, the program for the youth is very cost-efficient because it decreases the number of UI recipients: I estimate that each subsidized job-month increased public revenues by euros. Overall, the young employment credit increased public revenues between 1997 and 2001 by 4,220,709 euros. Unsurprisingly given the lack of employment effects of the program for senior workers, each subsidized job-month cost et al., 2013) 5 I rule out two additional channels that could explain the null senior employment effect. First, there was no lack of supply of senior workers. The number of individuals older than 45 looking for jobs actually increased after 1997s policy change. This is consistent with a search-and-matching framework that predicts that search efforts raise when workers are more likely to receive job offers (Pissarides, 2000). Second, I discard that the tax cut was shifted into higher wages and thus not decreasing labor costs of senior workers. This latter channel is a theoretical possibility in a Nash bargaining model of wage-setting. 6 Specifically, the new regulations restricted firms ability to fire subsidized workers and then use such contractual arrangements again with new workers. 5

6 the government 138 euros. The results of this paper have several implications for labor market policy and research: first, I show that a properly designed youth employment credit can increase net employment, raise public revenues, and not create displacement effects. In contrast, a similar program for workers aged 45 or more fails to increase employment, generates substantial displacement, and is not cost efficient. Both results have important implications for policy. On one hand, the findings provide a more optimistic view regarding the usefulness of active labor market policies for young jobseekers. Recent surveys of US and European active labor market program evaluations (Heckman et al., 1999; Card et al., 2010) found that most policies were ineffective for young workers. One potential reason for the different results is that decreasing payroll taxes might be an effective policy tool in contexts in which tax rates are high and unemployment affects a broad range of the population. 7 The findings are also consistent with Pallais (Forthcoming) evidence that hiring inexperienced workers improves their labor market outcomes. 8 On the other hand, the lack of effects for senior workers suggests that employment policies will fail when early retirement through UI is an option. 9 As many countries are increasing the incentives to work at older ages and delaying retirement (Whiteford and Whitehouse, 2006), the results suggest that policies aiming to expand the work life might have small effects if they are not accompanied by reforms including both hiring incentives for older workers and restricted pre-retirement through the UI system. The second implication relates to a growing literature studying the design of labor market policy over the business cycle (Jung and Kuester, Forthcoming; Landais et al., 2010; Michaillat, 2012a,b; Shimer, 2012). More specifically, Jung and Kuester (Forthcoming) argue that employment credits should be increased during recessions. Michaillat (2012a,b) and Shimer (2012) reach similar conclusions in an economic context of rationed jobs and jobless recoveries, respectively. 10 The results in this paper show that, during 7 The combined payroll tax rate in Spain is 28.3%, whereas in the US is 15.3%. 8 Pallais (Forthcoming) hires randomly selected workers in an online marketplace and shows that their subsequent labor market experience improved. She argues that hiring inexperienced can increase market efficiency because it generates information about their ability. 9 Blundell et al. (2013) show how retirement and early-retirement schemes decrease the number of people working at older ages in France and UK relative to the US. Moreover, age-discrimination in hiring older workers has been previously documented (Adams, 2004; Chan and Stevens, 2001; Hu, 2003; Hutchens, 1986, 1988). 10 There is a long tradition of proponents of employment credits or wage subsidies as a policy tool, 6

7 a jobless recovery, employment credits might be a useful policy tool, but that its effectiveness depends crucially on the workers life cycle, an aspect not taken into account in the papers mentioned above. 11 Finally, I contribute to the empirical evaluation of labor market policies that might cause displacement and windfall effects. Recent research has detected important displacement effects of job-search assistance programs (Crepon et al., 2013), UI extensions (Lalive et al., 2013), and employment protection (Crepon et al., 2008). In contrast, evaluations of employment credits have not been able to provide convincing evidence of the importance of displacement effects (Perloff and Wachter, 1979; Huttunen et al., 2013; Kugler et al., 2005; Egebark and Kaunitz, 2013; Goos and Konings, 2007). Moreover, most papers do not provide estimates of windfalls, or the number of subsidized jobs that would have existed even without tax credits. I contribute to this literature by estimating both effects in a compelling way. The paper is organized as follows. Section 2 discusses the theoretical predictions. Section 3 explains the main features of the Spanish labor market, the administrative data I use, and the institutional details of payroll tax legislation in Spain. Section 4 presents the empirical results and section 5 evaluates the costs and benefits of the policy. Finally, section 6 concludes. 2 Conceptual Framework The theoretical predictions of employment credits have been previously analyzed by many authors (Hamermesh, 1993), (Johnson and Layard, 1986), (Mortensen and Pissarides, 2001), (Pissarides, 1998), and (Neumark, 2013). All analysis agree that an employer-side employment credit will shift out the labor demand curve as long as it not necessarily related to the state of the business cycle. See Kaldor (1936), Phelps (1997), and Snower (1994). 11 The empirical findings in Saez et al. (2012) are consistent with the results for senior workers in this paper, but not with those of younger workers. They study a cohort-based payroll tax reform in Greece that increased payroll tax rates for some groups. The average age of workers affected by the Greek reform was around 37. They find that increasing employer payroll taxes does not affect employment and wages. In light of the findings in this paper, their results might reflect a period in workers life cycle in which payroll taxes do not distort employment decisions. 7

8 is not perfectly inelastic. 12 The new employment and wage equilibrium levels will be determined at the intersection of the new labor demand curve and the supply curve. The relative size of the effects on employment and wages will depend on labor demand and labor supply elasticities: the more elastic is labor supply, the larger will be the employment effect and the less will wages rise. On the other hand, the more inelastic is supply, the larger will be the positive effect on wages and the smaller the effect on employment. Figures 5a and 5b represent the extreme cases with perfect elastic supply and perfectly inelastic supply, respectively. Shimer (2012) analyzes the implications of wage rigidities and jobless recoveries through the lens of a neoclassical growth model, and concludes that payroll tax cuts might have positive employment effects in a depressed economy. Michaillat (2012a) and Michaillat (2012b) also conclude that payroll tax cuts can increase employment in a search-and-matching framework that allows for jobs to be rationed in equilibrium. However, the effects of employment credits in a new keynesian framework are more nuanced. Employment will only increase in response to hiring subsidies if it is accompanied by an expansion of aggregate demand (Gali, 2013). Overall, the theoretical literature supports the case of employment credits as a policy tool to increase employment. However, the case of age-targeted wage subsidies introduces more complexity to the analysis. Davidson and Woodbury (1993) develop a model to study whether a reemployment bonus displaces workers not offered the bonus. The model predicts important negative employment effects on the group of individuals who are not offered the subsidy. Since the employment credits feature age discontinuities at 30 and 45 years old, it is plausible that non-treated workers might be displaced. Such displacement effects should be stronger in the proximity of the threshold, since workers of similar age are expected to be better substitutes for each other. The negative effects of targeted policies on non-treated groups have been shown to be empirically relevant by Crepon et al. (2013) for targeted job search assistance, Lalive et al. (2013) for UI extensions, and Crepon et al. (2008) for age-targeted employment protection. The presence of displacement effects difficults the interpretation of difference-in-difference and regression discontinuity estimates. For the former, the key assumption is that both the treatment and control 12 That result is common across competitive labor market models and models departing from perfect competition by introducing union wage bargaining, search-and-matching, and efficiency wages(johnson and Layard, 1986; Pissarides, 1998). In those frameworks the supply function is replaced by a wage function. 8

9 group would have had the same trends absent the reform. However, the possibility of effects on the non-treated violates this assumption. Then, a difference-in-differences framework can only identify the differential effects of the policy across treatment and control groups, obscuring any interpretation of the estimates as net employment effects. Similarly, age-targeted employment credits should create a jump in hiring at the RDD threshold. However, such jump could be caused both by a movement upwards in hiring of the treated, and a movement downwards for the non-treated. Despite these difficulties, evidence on displacement effects can be shown under plausible assumptions. I will follow three different strategies to show whether the positive difference-in-difference and RDD estimates for youth employment reflects net employment creation. The first one is related to the discontinuation of the youth employment credit in March The cohort born in March 1971 benefited from the credits until their 30th birthday, and afterwards should not have suffered as much displacement from it than cohorts born earlier since the policy was no longer in place. Thus, in case displacement exists, their employment evolution after the age of 30 should be much more positive than that of cohorts born before. The second strategy uses the age distribution of new permanent hires, before and after each policy change. Figure 16 illustrates three different scenarios. The upper figure shows the case absent the policy. Hiring should have been smooth across the threshold. The middle figure shows the case when the policy generates both job creation and job destruction in the proximity of the threshold. The lower figure shows the case when the employment credit only induces job creation by shifting upwards the distribution of new hires below the age of 30. I will use figures of the raw hiring data to search for a pattern similar to those just described. Another potential concern with wage subsidies is that they might generate windfalls: workers who would have been hired in any case might be hired with subsidized contracts. Such windfalls might occur in several ways. The main one is that workers that would have been hired with a regular permanent contract without a subsidy might now be hired with a subsidized permanent contract. Such type of strategic behavior, though might not necessarily offset net employment increases, will decrease the cost-effectiveness of employment credits. The second type of windfall can occur by contract shifting. Tax credits might be used to hire workers who would have been otherwise employed as shortterm or self-employed. If firms respond in that way, the effects on net employment will be even lower. In that case, though, the impact on cost-effectiveness is more ambiguous 9

10 since workers might benefit from longer tenure in permanent jobs. While the relevance of windfalls for evaluation of wage subsidies has been recognized, it has been rather difficult to estimate. However, since my dataset includes information on each type of contract, including both subsidized and not subsidized permanent contracts, I can shed light on the extent to which windfalls are ocurring. 13. Short- and long-run responses to the policy might differ. Previous research has shown that agents in the labor market face important optimization friction that make shortterm responses smaller than long-term ones (Chetty et al., 2011; Chetty, 2012; Kleven and Waseem, 2013). Thus, employment estimates of the effects of the expansion of wage subsidies in 1997 might underestimate its long-term effects. However, as discussed above, such policies are likely to generate displacement and windfall effects. Once agents are able to optimize their behavior according to the new policy, such type of undesired consequences might also increase, questioning the desirability of wage subsidies in the long-term. 3 Institutional Context and Data 3.1 The Spanish Labor Market In this section I describe the two main characteristics of the Spanish labor market: first, Spanish unemployment is high relative to other OECD countries even in good times, and it is highly countercyclical; second, there is a high share of short-term contracts. First, as can be seen in figure 4, the unemployment rate in Spain is highly countercyclical. When the economy slows down, the unemployment rate jumps sharply, reaching levels around or above 25% both for the early 90s recession and the Great Recession. Also, the unemployment rate minimum for the last 30 years was around 8%, still a high number compared to other OECD countries. That minimum was achieved during the housing boom of the 2000s. Youth unemployment is much higher than that of senior workers. (Dolado et al., 2013) show evidence of some of the characteristics of the youth labor market in Spain. First of all, the transition from education to a first stable job takes longer in Spain com- 13 The only paper I have found providing an estimate of windfalls is Bishop and Montgomery (1993) They study the Targeted Jobs Tax Credit (TJTC) in the US for years 1981 and 1982 and conclude that 7 out of 10 TJTC jobs would have existed absent the subsidy. 10

11 pared to other European countries. Second, the high-incidence of temporary employment for young workers is the main determinant of the volatility of youth employment. Third, skills mismatch is a widespread phenomenon, exacerbated by low mobility and a small rental market. Finally, the ratio between the unemployment rates of youth and adults has stabilized since the early 1990s at a value of around 2.5, regardless of the business cycle phase. The second main characteristic of the Spanish labor market is the high fraction of short-term workers. Since the introduction of the temporary contract (1984), the fraction of short-term workers rose steadily until stabilizing around 1/3 of the workforce, a much higher level than in other European countries (Dolado et al., 2002). The short-term contract has been the main change in regulation to flexibilize the labor market, since temporary workers are not entitled to severance payments at the end of the job relationship. However, as can be seen in figure 1, there does not seem to be any relationship between unemployment and the % of short-term workers. 14 Temporary contracts are in general worse jobs, and less-skilled employees take them. (Jimeno and Toharia, 1993) show that temporary contracts are related with a negative wage differential of about 10%. However, (Davia and Hernanz, 2004) show that this wage differential is caused by different worker characteristics. Low-skilled workers are hired under temporary contracts, while the high-skilled take permanent contracts. (Arranz and Garcia-Serrano, 2007) show that job stability has declined in Spain since the introduction of temporary contracts. Regarding the effects on training of short-term contracts, (Albert and Hernanz, 2005) find that workers holding temporary contracts are less likely to be employed in firms providing training. More importantly, temporary workers employed in firms providing training are less likely to be chosen to participate in training programs. 3.2 Payroll Tax Legislation Payroll tax legislation sets different payroll tax rates depending on the regime to which the worker is affiliated. The main group, called Régimen General, includes most private and public employees (13,419,951 workers or 77% of total). 15 The following groups 14 Besides the flexibilization argument in favor of short-term contracts, there are also demand reasons to explain the rise of this type of work arrangement: two of the main sectors in Spain, tourism and construction, are highly seasonal. Also, the demand repercussions of these two sectors increases the seasonality of other sectors. 15 Data is for Source is Ministerio de Empleo y Seguridad Social (MESS),

12 are self-employed workers (2,951,021 workers or 17% of total) and farmers (685,960 or 4% of total). There are other small schemes for coal workers, sea workers, and housekeepers. The employment credits that are the focus of this paper apply to all new permanent jobs, except for the sector of self-employed individuals. I will thus focus on workers affected by the policy, but will also discuss the employment effect for self-employed workers, since it is a common practise by firms to declare some work as carried out by self-employed individuals to avoid paying payroll taxes and severance payments. Payroll taxes in Spain are paid both by the employer and the employee. They are a function of the wage of the employee and two tax rates: one that applies to employers and one that applies to employees. There is a maximum and a minimum base for the wage depending on the occupational category of the worker. Table 1 below shows, for employees in Régimen General, the minimum and maximum basis for each category of worker for 1997, and table 2 shows the general tax rates for both employers and employees. 16 The tax revenue collected is used to pay unemployment, workers accident, and health insurance; and retirement, widow and orphan pensions. The money is also allocated to pay for training courses and to protect the workers in case of firm s default. In order to stimulate hiring and increase employment, the Spanish government has implemented policies to reduce labor costs. 17 It has done so by establishing programs that reduce employers payroll taxes for specific groups of workers when hired as new permanent workers. Table 3 summarizes the programs for the period between 1984 and I will focus the estimations on the age-discontinuities at 30 and 45 years, since they provide both time variation and treatment and control groups. The main policy change occurred in May 1997, in the midst of a weak recovery. The government expanded both the number of workers that could be hired claiming a tax credit and the generosity of the tax cut. For instance, after 1997 s reform, workers below 30 and above 45 did not need to prove a minimum amount of time spent as unemployed or limited work experience to be hired with a tax credit. For young workers, the employment credit was a temporary reduction of the payroll tax rate for the first 2 years of the job relationship, whereas for old workers the tax credit lasted for the whole duration of the contract. The design of the employment credits aims to help specifically certain age groups believed to be the most disadvantaged in the labor market, but also to help low productivity 16 The payroll tax rate of the main group of workers has been very stable. Last reform took effect in 1995 and decreased firms payroll tax rate by 3.3% and employees payroll tax rate by 4.1%. 17 See appendix A for a full list of laws regarding payroll tax policy. 12

13 workers who do not fall in these age groups. As can be seen in table 3, between there are other programs that are not age related but target the long-term unemployed. However, what is important for the estimation is that the tax cuts available for hiring young and senior workers are much easier to claim since they are age-related. The main program for workers between 30 and 45 years old can only be claimed for workers who have been at least a year registered as unemployed ( ) or 6 months ( ). The non-age related tax cuts should minimize displacement effects created by the agetargeted policies, and thus provide legal support for the absence of negative effects of the youth employment credit on non-treated individuals that I show later in the empirical analysis. The second policy experiment happens in March 2001, when the policy for young workers was discontinued. In contrast, employment credits for senior workers remained very stable between 1997 and In 2006 there is another policy change: the tax credits were changed from a reduction in the tax rate to a lump-sum transfer to employers, making payroll taxation more progressive. Moreover, employment credits for young workers were reintroduced. Since the focus of the paper is the impact of employment credits during jobless recoveries, and in 2006 the Spanish economy had been expanding for several years, I will not use this policy change in the current paper. Besides using the policy changes in 1997 and 2001, I will also exploit the age discontinuities through a RDD and a RKD. This approach proves useful to understand the behavioral responses that happen in the proximity of the thresholds. Moreover, the age threshold can be exploited to explore both the short- and the long-run effects of the policies. The dose of the treatment is sizable. Figure 2 shows the temporary and permanent tax rate by age. For example, using the basis and tax rates for 1997 and assuming an employee with wage equal to 2,000 euros, the monthly tax credit after 1997 s reform, for hiring a young worker, is = Thus, the subsidy represents a 7.6% saving in total labor costs during the first two years of the job relationship. In contrast, the monthly tax credit for workers older than 45 years old when hired as permanent workers would have been = euros each month during 2 years, and = 236 euros for the rest of the contract duration. The saving represents 11.5% and 9.5% of the total labor costs for the first two years and the rest of the contract, respectively. Thus, the employment credits represent an important 13

14 reduction in total labor costs. Though the tax credits are not granted for firms that hire workers above some number, its administrative design makes them similar to a marginal employment subsidy (Johnson and Layard, 1986). There are several limitations that limit the scope of the employment credits. Most importantly, they can only be received for workers who have not been working in a permanent contract during the last 3 months. Thus, the programs target individuals with low job stability. Guell and Petrongolo (2007) estimate that 86% of new entries in Spain are under short-term contracts, and that only 5.7% of them are converted into permanent jobs. Moreover, given the divide of high-and low-productivity workers among permanent and short-term jobs, the program targets low-skilled individuals (Albert and Hernanz, 2005; Arranz and Garcia-Serrano, 2007; Davia and Hernanz, 2004; Jimeno and Toharia, 1993). Other administrative details of the employment credits are important to limit the possibility of strategic behaviors by firms, like excessive churning. Two of the limitations are specially important. The first one was introduced in 1999, and makes firms who wrongfully dismissed workers with a tax cut ineligible to use the contracts with tax credit either for a year since the dismissal happened or for as many workers as wrongful dismissals happened. The second limitation is that a employment credit contract cannot be signed with workers who hold a permanent contract with the same firm group during the previous 24 months. 18 Claiming a tax credit was an easy task. Figure 3 shows the back-page of the labor contract. The employer has to fill in one of the options available in the sixth clause of the contract. The first one corresponds to the tax credits available when hiring a worker who is younger than 30 years old. The writing in option a) specifies the tax credit that was available between January 2000 and March Option c) specifies the tax credit when the employer hires a worker over 45 years old. Finally, options b), c) and e) describe the tax cuts available if the firm hires employees in other situations not only related to their age: workers registered as unemployed for at least a year, women 18 Other limitations are: the tax-credited contracts cannot be used to hire relatives of the owner or of the management chief; there are people who cannot benefit from the contract too: firms managers, home service, people in jail, professional sportsmen, artists, and dockers working for public societies; the employers need to be current with tax payments and must not have been excluded from the program because of any infraction they could have committed. Finally, the tax-credited contract, combined with other programs, cannot suppose a tax credit of more than 60% of the annual wage. 14

15 hired in sectors in which they are underrepresented, and unemployed people perceiving unemployment assistance. Finally, the employment credits were accompanied by lower severance payments. However, Elias (2014) explores the effects of lower severance payments for young workers during the period , when no employment credits were available for that group. There are no effects on hiring, employment or wages of reduced dismissal costs. He argues that the main reason why this policy was not effective is that only firms that did not dismiss a worker in the last 6 months could hire another worker with lower severance payments. Therefore, the main effect of the policy changes in 1997 must have been related to the employment credits. The details of severance payments regulation are explained in web appendix Data The Continuous Sample of Work Lifes (Muestra Continua de Vidas Laborales, MCVL) is a joint dataset of administrative data from different sources: the social security administration, the census, and the tax administration in Spain. They have detailed information of the start and end of each employment and unemployment spell, monthly earnings (bottom- and top-coded), the reason why the job relationship ended, the type of contract (very importantly, whether the contract benefited from a tax credit or not), the size of the firm, the sector, whether the job was part-time and the number of hours, the location of the job, etc. The data also contains information about the individual: sex, education, date of birth, province of birth, citizenship, as well as the date of birth and sex of the members of their household. The sample was constructed in the following way: in 2004, over 1 million of workers, or 4% of all individuals who had some relationship with social security, were selected. 19 Sampling was random, without any kind of stratification. For these people, the dataset contains not only their labor and unemployment information for 2004, but since they had their first relationship with the Spanish social security system. The same individuals selected in 2004 were sampled for each edition of the dataset between 2005 and Thus, I can reconstruct the working life of the individuals since they started working up to In case a worker selected in 2004 leaves the sample in any of the future years, 19 Individuals who had some relationship with social security were either formally employed, receiving some kind of unemployment insurance, or were perceiving a contributory pension 15

16 because it stops having a relationship with social security (i.e. he is out of employment and does not collect UI; he dies), he is replaced by another randomly selected worker that had some relationship that year with social security. Similarly, the whole labor life of that new worker is included in the dataset. Finally, if any of the workers is not sampled during one of the editions of the dataset because he did not have any relationship with social security for a year or more, but he becomes employed again, he will reappear in the dataset the year in which he had restarted his relationship with the social security system. Table 4 reports summary statistics for the year 1997, when the tax credit policy was enacted. I classify the workers in 5 year age groups and report the descriptive statistics for the main groups of the empirical analysis: 25-30, 30-35, 40-45, and Workers are more likely to be men for all age groups. Most of them have achieved at most secondary education. The % that at most completed primary education is increasing in age. Most of them are Spanish citizens, but the importance of the immigrant labor force is bigger for the younger cohorts (almost 20%) than for older cohorts (around 5%). Their real daily wage is of 37 euros for young workers, rising to for the most senior group of workers. The % of workers in what is considered good jobs (permanent and public workers) is increasing in age, while the % of employees in short-term contracts is decreasing. Incidence of part-time work is higher for younger workers. The mean size of the firm is around 8-10 workers and most people work in the services sector. Table 5 displays summary statistics for workers in each type of contract in As can be seen in the table, permanent workers with a tax cut, compared to those without a tax cut, are younger, more likely to work in the services or construction sector, are less likely to have completed university education, have accumulated less experience during the last 12 months, have a lower wage, and are working in smaller firms. The main differences between short-term workers and permanent workers with a tax cut is that the former are younger, are more likely to have completed at most primary education, have accumulated less experience during the last 12 months, and have a lower wage. 20 I choose the year 2000 instead of 1997 as above to make sure that there is a wide enough sample of workers under contracts with a tax cut. 16

17 4 Empirical Strategy and Results 4.1 Sorting Into Contracts with a Tax Cut The descriptive statistics discussed in the previous section have just shown that workers hired as permanent with a employment credit are worse in several observable characteristics than permanent workers hired without a tax credit. I begin the empirical analysis of the effects of the policy by further exploring the type of workers who are hired with tax credits. I restrict the sample to workers hired as permanent workers and run a regression with a dummy equal to 1 if the worker was hired with a tax credit, and 0 if not, on several predictor and control variables. Besides learning about the characteristics of each type of worker, the exercise serves as a first test of whether the employment credits are actually acting as a marginal permanent employment credit. The administrative design of the policy is meant to target workers with a lower attachment to the permanent workforce. Thus, if the policy limitations are effective, workers hired with a tax cut should be worse in terms of observables such as education, experience or wages. The specification is as follows: Y ipct = α + δ p + φ c + γ t + βx ipct + ɛ ipct (1) Y ipct is a dummy that indicates whether the individual was hired with a tax cut or not. δ p, φ c, and γ t are province, cohort, and year fixed effects. X ipct is a vector of characteristics of the worker and the job: education, sex, citizenship, disability, experience, wage, industry sector, part-time, and firm s size. Table 6 reports regression estimates of sorting into a permanent contract with a tax cut relative to permanent contracts without a tax cut. Column 1 shows the results for workers under 30 years during the period between May 1997 and March 2001 (when the tax credit for young employees was available). Column 2 reports the estimates for senior workers years old between May 1997 and June The results are consistent with the policy limitations: young workers hired with a tax cut are 4% less likely to have completed university education, and 3.3% less likely to have attained secondary education. The marginal effect of the fraction of months worked during the last 12 months is associated with a.11% lower probability of being hired with a tax cut. A 1% increase in the wage is associated with a 4.7% lower probability of being hired with a tax cut for young workers, and a 1% increase in the size of the firm implies a 2.6% lower probability of receiving the tax cut. Young workers are less likely to work in the 17

18 agriculture and industry sector, compared to the omitted services category. Employment in the construction sector is not significantly different than that of the services sector. Selection into tax cut contracts has similar implications for senior workers: they have lower probability of having attained university education, secondary education (though not significant), lower experience, lower wage, and are hired by smaller firms too. The size of the coefficients is larger in absolute value for senior workers with respect to young workers. They are also less likely to work in the agriculture sector, though they are more likely to work in construction than in the services sector. Thus, workers hired with tax credits are worse in terms of several observables after controlling for province, cohort, and year fixed effects. If the use of employment credits among firms was concentrated on individuals who already had permanent contracts, we would not expect such negative sorting into subsidized contracts to occur. However, it might still be the case that the policy induces shifting between short-term and permanent contracts, without net effects on employment. The next section will show further evidence on that. 4.2 Transitions and Employment Short-Run Effects As a first step in testing whether employment credits had an effect on young and senior employment, I begin by implementing a difference-in-difference analysis comparing workers above and below the age discontinuities, before and after the policy change. I select a window of time of a year and a half before and after the reforms, and construct a quarterly panel of workers aged 25-30, 30-35, and Individuals aged and are the control group, and if they are displaced by the policy, the estimates have to be interpreted as the differential effect of the policy on the treated versus the control group. The outcome variables are employment, hires, lay-offs and quits. The latter two variables are important as well since they will indicate whether employment credits generate excessive churning from employers willing to game the regulation by separating from their workers, and later rehiring employees with a tax credit. Identification in a difference-in-difference analysis relies on parallel trends for both treatment and control groups. If transitions and employment for each group were following different trends, a difference-in-difference estimate might just capture these different patterns. Thus, I start by running a specification including interactions between the 18

19 treatment group and 10 quarter time interactions: Y iq = α + δ q + φ a + β q T reatment i + γx iq + ɛ iq (2) where Y iq indicates that individual i was hired, laid-off, employed, or quitted in quarter q. φ a are age specific dummies. β q are the quarter by quarter difference-in-difference estimates. X iq is a vector of control variables: sex, education, disability, immigrant, dummies for industry sector, part-time job, firm s workforce size, province fixed effects. Since permanent and short-term contracts are the most prevalent ones, I will focus most of the discussion on them. Figure 6 displays the coefficients of the treatment dummies interacted with quarter period dummies for young workers. The omitted period is the quarter just before the policy was enacted. The figures on the left show flows in and out of permanent employment, whereas the ones on the right report the effects for flows in and out of short-term employment. For the case of new permanent hires, the pretreatment coefficients oscillate around zero and are not significant, providing evidence of no differential pre-treatment trends. In the quarter when the policy is enacted, the estimate jumps upwards and becomes significant. The estimate remains significantly different than zero for the post-treatment quarters, slightly increasing over time. It reaches its maximum level 4 quarters after the policy is approved indicating a.4% increase in permanent hires. There is no significant effect on permanent workers laid-off unless for two periods after the policy change, when the estimate shows a.2% less lay-offs, but goes to oscillate around zero immediately after. The number of permanent workers quitting their jobs increases a few periods after the policy is passed. The estimate reflects a.1% higher number of quits 5 periods after the policy is enacted. The figures for short-term contracts suggest that both transitions in and out increase after the policy change. A potential explanation is that the policy stimulated short-term hiring too even if that was not subsidized. Firms might hire young workers under this type of contract to screen them, and then later convert to permanent employees those who performed better. I repeat the same strategy but with a dependent variable capturing transitions from short-term to permanent within the same firm. Figure 9 displays the evidence: the employment credit for permanent workers increased the number of transitions from short-term contracts within the same firm. Note that the coefficients are only slightly smaller than those for overall transitions into permanent employment. Thus, most of the new entries into permanent work are happening for workers who were previously working as short-term within the same firm. 19