The Perverse Effects of Partial Employment Protection Reform: The Case of French Older Workers
Luc Behaghel*, Bruno Crépon** and Béatrice Sédillot***
French firms laying off workers aged 50 and above have to pay a tax to the unemployment insurance system, known as the Delalande tax. We evaluate the impact of this tax on layoffs as well as on hiring, taking advantage of several changes in the measure since its introduction in 1987. A legislative change in 1992 exempted firms from the tax for workers who were hired after age 50. Following this change, the transition rate from unemployment to employment increased significantly for workers over 50 compared to workers under 50. The difference is sizeable: between one third and one half of the initial transition rate. Evidence on the effect on layoffs is less clear cut. The impact is sizeable only for the most stringent tax schedule, after 1998.
Keywords: Employment protection; older workers; layoff; hiring. JEL classification: J23; J63; J65. * Corresponding author. Paris School of Economics (lea, INRA) and Crest-INSEE – 48, bvd Jourdan – 75014 Paris ;
[email protected] ** CREST-Insee ;
[email protected] *** DARES ;
[email protected]
We are particularly grateful to David Blau, Pierre Cahuc, Thierry Magnac, Thomas Piketty, and two anonymous reviewers for helpful comments. We also thank participants to the 2006 EEA-ESEM meetings, seminar participants at CREST, Lunch Seminar Jourdan and at Insee Research Seminar. The usual disclaimer applies.
1
1. Introduction Confronted with persistently high unemployment rates, most countries in Continental Europe have reformed their employment protection over the past two decades. These reforms have however generally been partial, as governments developed exceptions to the main employment contract while leaving the latter largely untouched. A well-studied example is the development of fixed-duration contracts and its coexistence with indefinite duration contracts. Such partial employment protection reforms have unintended perverse effects. In the case of France, for instance, Blanchard and Landier (2002) and Cahuc and Postel-Vinay (2002) show how fixed-duration contracts increased the segmentation of the labor market. In the case of fixed-duration contracts, segmentation occurs because firms and workers select one type of contract depending on their characteristics and their negotiation power; the law in itself does not make a distinction between workers based on such characteristics as age, gender, disability status or experience. In this paper, we study employment protection reform in a case where the law explicitly treats workers differently depending on their age. Specifically, we study the impact of the Delalande tax, a measure introduced in 1987 in France to discourage firms from laying off older workers.1 Firms have to pay a tax to the UI system each time they lay off a worker aged above 50.2 The amount of the tax is proportional to the worker’s gross wage at the time of layoff; it has varied over time but can be as high as twelve months of gross wages. Moreover, in 1992, a major reform occurs that exempts workers who are hired after the age of 50. Studying the impact of the Delalande tax is interesting for two reasons. First, the tax provides a good illustration of the potential benefits and drawbacks of targeting employment protection on specific groups of workers. On the one hand, the benefits of having the firms internalize the cost of laying off older workers are likely to be high. Indeed, the probability that an unemployed older worker finds a job is very low in France. Jointly with high layoff rates and the widespread use of early retirement schemes, this explains the low employment rate of older workers in France: for male workers aged 55 to 59, it has fallen dramatically in the early eighties and stabilized below 60% since then. The social cost of this low employment rate is probably high, but firms do not seem to internalize it; they rather tend to view the early exit of older workers as a painless way to adjust their workforce (Behaghel and Gautié, forthcoming). The question is therefore whether the tax has been effective in modifying firms’ and workers’ behavior. On the other hand, standard theory predicts that firms will refrain from hiring heavily protected workers, as they expect higher total labor costs. This negative effect of employment protection on hiring may be particularly acute in the case of the Delalande tax.
1 Temporary layoffs do not exist as such in France. Throughout the rest of this paper, we use the word “layoff” for permanent layoffs. 2 The age threshold was 55 in 1987 but was lowered to 50 after 1992.
2
Indeed, firms may easily substitute younger (less protected) workers to older (more protected) ones. Moreover, as the Delalande tax is paid to the UI system, its negative effect cannot be undone by private transfer, as opposed to severance payments in Lazear’s (1990) argument. Overall, the Delalande tax appears as a case where the desired as well as the undesired effects of employment protection may both be particularly strong: it therefore appears as an interesting measure to study. The second interest of the Delalande tax is that it has undergone several major legislative changes that provide useful sources of identification. Building on these changes, we are able to separately analyze the effects of the tax on the probability that a worker is hired or laid off. The approach and the key results can be summarized as follows. We start by studying the adverse impact of the tax on hiring. We take advantage of a change that occurred in 1992 and exempts firms from the tax for workers hired after the age of 50, if they are laid off later on. If there was any adverse impact of the tax on unemployed workers aged above 50, it disappeared in 1992; it should therefore be evidenced ex post by an increase in the hiring rates for this particular age group. Indeed, we find that the chances of finding a job increased for unemployed workers aged just above 50 relative to those just below 50. The difference-indifference estimate is sizeable (around one third or one half of the initial transition rates), pointing toward a large adverse effect of the tax on hiring. Then, we study the desired effect of the tax on layoffs. We use three changes in the tax schedule (in 1987, 1992 and 1998) to infer consequences on layoff rates, using workers in the unprotected age group (below 50) as a control group. Our results show that the tax reduced layoffs for older men; however, the estimates are imprecise and rather small, except for the most stringent tax schedule in 1998. Results for women are sensitive to the specification. We view the negative impact of the tax on the hiring rates of workers older than 50 as an example of ‘perverse effect’ of a partial reform: by targeting employment protection on a limited age group, the legislator has induced substitution from these workers toward younger ones. It is unclear whether the reduction in layoffs is sufficient to compensate for this unfavorable effect. Our paper relates to a handful of recent papers that use variation in labor legislation across groups of workers in a given country as a way to provide micro evidence on the impact of employment protection. Acemoglu and Angrist (2001) look at the effect of the American with Disabilities Act: they find that it reduced the employment rate of the protected group, the disabled workers, most likely through lower hiring rates – but they do not find significant effects on layoffs. Autor, Donohue and Schwab (2002) find that employment of protected groups is ultimately reduced by the adoption of wrongful-discharge laws by various States in the U.S. In Colombia, Kugler (2004) finds that a reduction in dismissal costs increased separations and accessions for formal workers relative to informal workers, and more so in larger firms that were most likely to be affected by the reform. Kugler and Pica (2005) study the impact of a reform increasing unjust dismissal costs for firms below 15 employees in
3
Italy. They find that both accession and separation probabilities decreased in small relative to large firms. In Italy as well, Schivardi and Torrini (2004) study the impact of employment protection on the distribution of firm size. They find that more stringent employment protection for large firms has slightly reduced the propensity of firms to grow. Hernanz, Jimeno and Kugler (2003) study the age-specific employment protection in Spain. They analyze the 1997 reform that reduced dismissal costs and payroll taxes for workers under 30 and over 45. They find that, for men, permanent employment probabilities increased for the two age groups, as did transition rates between unemployment and employment. Effects on women are weaker. Last, Hakola and Uusitalo (2005) study the impact of the experience rating of unemployment pension benefits in Finland, using a reform that affected firms differently depending on their size. They find that layoffs are sensitive to the degree of experience rating; but they do not assess the adverse effect on hiring. Overall, we believe that our results add to the micro evidence on the two effects of employment protection on hiring and layoff, especially by providing unusually clear evidence on the effect on hiring, thanks to a unique natural experiment. The paper proceeds as follows. Section 2 details the legislation of the Delalande tax and the sources of identification provided by the various legislative changes. Section 3 presents the data. Section 4 deals with the effect on hiring, and section 5 with the effect on layoffs. Section 6 concludes.
2. Overview of the Delalande tax: legislative changes and natural experiments Since its introduction in 1987, the Delalande tax has been modified several times. However, its principle has remained unchanged. Firms laying off workers above a certain age threshold have to pay a tax (“la contribution Delalande”) to the UI system (Unédic). This tax is proportional to the gross wages of the worker. It is only due if the worker is employed under an indefinite term contract (CDI). Only the private sector is concerned (civil servants may not be laid off). From 1987 to 1992, the tax amounted to three months of gross wages for all workers aged 55 and above. From 1992 on, the amount of the tax depends on the age of the laid-off worker (starting at the age of 50) and, for some periods, on the size of the firm (table 1). Exemption rules also vary. These changes are sources of variation that help identify the effects of the tax. However, other major changes affecting the employment prospects of older workers occurred during this period. This section therefore reviews the main legislative changes in the Delalande tax to assess them as sources of identification. The introduction of the tax (1987) It is interesting to notice that the introduction of the Delalande tax occurred almost by surprise: it was voted as an amendment to a law on long-term unemployment, and the
4
government does not seem to have been involved in its elaboration – a member of Parliament, Delalande, proposed the amendment during the night when the vote took place. The amendment actually had two purposes: restore the financial balance of the UI system and counteract a recent and sharp increase in the number of layoffs of older workers. This increase was explained by a weakening of employment protection in 1986, with the suppression of the administrative authorization that used to be required for layoffs. The government used this administrative authorization as a way to induce firms to use early retirement schemes when dismissing older workers. Firms had to pay part of the cost of these schemes. After 1986, they rapidly turned to layoffs, which were less costly to them and still provided generous unemployment benefits to older workers. Hence, in part, the introduction of the Delalande tax is driven by a general change in the strictness of employment protection. As this change concerned all workers and was not due to changes in the employability of older workers around 1987, it is arguably exogenous with regard to the relative demand for older workers and can thus be considered as a good source of variation to study the impact of adjustment costs on that relative labor demand. One caveat applies, though: the surge in layoffs of older workers that preceded the introduction of the tax may have consequences on the relative evolution of layoff rates of older workers, compared with younger ones. Specifically, this surge may partly reveal an increasing trend in layoff rates of older workers; if that trend is neglected, the effect on layoffs of the Delalande tax may be underestimated. Conversely, if the surge in layoffs before 1987 was strictly accidental, perhaps only the most productive older workers remained in employment in 1987, thus leading to lower layoff rates after 1987: if this is not taken into account, the estimated effect of the Delalande tax would be biased upward. To summarize, the introduction of the Delalande tax in 1987 presents some of the features of a natural experiment. It is driven by an economy-wide shock (a global change in employment protection) that is not directly related to the outcome of interest (the employment of older workers). However, two issues arise: trends in the relative demand for older workers as well as changes in unobserved heterogeneity may bias inference based on a comparison of the evolution of layoff rates across ages.
The 1992 changes Two major changes occurred in 1992. First, the schedule of the tax was modified. Workers above 50 (instead of 55) are now liable for the tax, but the amount due in case of a layoff varies with age, from one to six months of gross wages. Overall, this is a substantial increase: the amount of the tax is multiplied by two for workers above 56. Figure 1 compares the cost of the Delalande tax to the mandatory severance payments that are also due (to the worker and
5
not to the UI system) in case of layoff.3 After the age of 55, the amount of the Delalande tax is clearly larger than the average mandatory severance payment. Second, the tax is now restricted to workers who were hired before the age of 50 or who had been unemployed for less than three months if they were hired after the age of 50. The increase in the tax schedule in 1992 constitutes a complex source of variation. The political context had changed: the government implemented this increase as part of an effort to raise the employment rate of older workers4 and of a global move toward more stringent employment protection, reversing changes made by the 1986-1988 government. Second, the economic context was characterized by a slowdown of growth that was part of France’s most severe recession since World War II, in 1993. This particular context may have had consequences on the relative demand for older workers. Third, the variation in the tax schedule with age renders the incentives more complex. On the one hand, a higher tax level at a given age makes it more costly to lay the worker off. But, on the other hand, the fact that the tax increases with age gives firms the incentive to lay off a worker earlier than otherwise, so as to avoid liability for a higher layoff cost in the future. Intuitively, the effect of the 1992 schedule can therefore be decomposed into two parts: the higher the level of the tax, the lower the layoff rate; but the higher the age slope of the tax, the higher the layoff rate.5 This second effect – the effect of the slope of the schedule – also existed for workers just below 55 in the 1987 schedule: it is now smoothed and applies to protected workers as well as to workers entering the protected age. The second change exempted firms from the tax for workers who were older than 50 at the time of hiring and had been unemployed for at least three months. This provides a unique setting to study the effect of employment protection on hiring. Indeed, it introduces a sharp discontinuity, at the age of 50: firms hiring unemployed workers above 50 will not be liable for the tax if they lay off these workers later on; but they will if they hire workers below 50 and fire them after they turn 50. In other words, after 1992, the adverse effect of employment protection applies to unemployed workers just below 50, but not to those just above 50. This discontinuity provides a simple source of identification: the effect of the tax on hiring can be estimated from differences in hiring rates above and below the 50-year threshold. To get a sense of the advantage of this natural experiment, it is helpful to compare it to the introduction of the tax in 1987. When the tax was introduced, the consequence was that all unemployed workers became more costly to hire, as the firm was facing the possibility of paying the In France, mandatory severance payments depend on the tenure of the worker in his firm. This is taken into account in the computations of figure 1: we use the distribution of workers’ tenure to compute the average amount of severance payments. However, additional severance payments may be due depending on the industry; these are not taken into account here. For details on the cost of employment protection in France, see Abowd and Kramarz (2003). 4 Access to early-retirement schemes is slightly restricted on the same period. 5 This decomposition can be derived in a formal model of labor demand with adjustment costs (see Behaghel, Crépon and Sédillot, 2004). A key assumption is that there is no uncertainty, so that the schedule of the tax years ahead does not intervene. 3
6
Delalande tax if it were to lay the worker off after the age of 55. Only the degree to which the firm was likely to incur that cost varied according to the age of hiring (once discounted, the Delalande cost was probably negligible for workers hired below the age of 30, for instance). However, disentangling this (continuous) effect of age on the expected Delalande cost from the (continuous) impact of age on productive characteristics that determine hiring rates would be problematic. After 1992, a comparison of hiring rates just above and below the age of 50 is in principle enough to identify the effect of the tax, under the assumption that other age effects on hiring rates are continuous and thus negligible on very short age windows. As discussed in more details in section 4, this discontinuity approach identifies the relative effect of the tax on the two age groups rather than the effect on one of the groups. Indeed, the two groups are affected by the reform. One should not think of unemployed workers over 50 as a “control group” unaffected by the change. They are affected for two reasons: first, they used to be liable for the tax and they are not anymore after 1992 (their own labor cost decreases); second, they are indirectly impacted by the fact that the labor cost of workers under 50 increases, as it can induce substitution effects in their favor. The important point here, however, is that the two age groups are affected in opposite directions based on a discontinuous criterion: this is what we use for identification. The 1998 change In 1998, the major change was an increase in the schedule of the tax for firms with more than 50 workers. This rise in the tax was implemented in a context of rapidly growing employment that benefited all categories of workers, except older unemployed workers.
Other policies affecting older workers’ employment The pension reform that was progressively implemented after 1993 introduced incentives for longer careers in the private sector. Pensions were calculated on the best 25 years salaries instead of the 10 best years, and the number of years of contributions required for a full pension was raised from 37.5 to 40. This reform affected all workers and in particular introduced no discontinuity between workers aged under and over 50. Over the 1990s, other public policies were more specifically targeted on older workers and may have influenced their transition rate from unemployment to employment. From 1990 to 1995, the ‘Contrat de retour à l’emploi’ (CRE) provided financial incentives (monthly subsidy and exemption of employers social contributions) to recruit long-term unemployed in the private sector. The policy had an age-specific component: in 1992, the CRE was enlarged to almost all jobseekers aged over 50. The firms hiring long-term unemployed over 50 benefited from a permanent exemption of employers’ social contributions. This enlargement was almost concomitant to the change in the Delalande tax but did not restrict to indefinite duration contracts. In 1995, the ‘Contrat initiative emploi’ (CIE) replaced the CRE. From 1996, the
7
CIE also provided more financial incentives to firms recruiting long-term unemployed over 50 (exemption of employer social contribution up to retirement, higher level of monthly subsidy). Given the specific provisions of CRE and CIE for jobseekers over 50, it is particularly important for our empirical analysis to disentangle the impact of the CRE-CIE policy from the impact of the Delalande tax. This is rendered possible by the fact that, although the two policies were reinforced in 1992, they were modified at different dates (1995 vs. 1998), and that the CRE and CIE targeted broader groups than the Delalande tax (all long term jobseekers were affected, not only those over 50). Their effects can thus be identified separately. Additionally, the CRE and CIE policies apply to workers under all types of contracts, whereas the Delalande tax only applies to workers under indefinite duration contracts. To summarize, legislative changes in 1987, 1992 and 1998 provide various sources of identification to study the impact of the Delalande tax, by comparing the probability of layoff and hiring of different groups of workers at specific periods of time. However, the most convincing sources of variation to study the impact on layoffs are not as convincing to study the impact on hiring, and vice versa. In the next sections, we therefore choose to separately estimate these two effects, at the expense of providing a global assessment of the impact of the tax on employment. We start by briefly reviewing the data we use.
3. The data We use the 1982 to 2002 waves of the French Labor Survey (Enquête Emploi). This is a rotating panel, with yearly interviews: members of a household are interviewed three years in a row, in March, and then leave the panel. We use the data at different time frequencies for studying hiring and layoffs. Concerning hiring, it is important to be able to precisely date the transition so as to use the age discontinuity at 50 and to be able to set apart workers who have been unemployed for less than 3 months, as they are not exempted from the tax. We therefore infer monthly transitions from the retrospective information on the date at which a new job started (if any). Concerning layoffs, we use yearly transition rates. The advantage of using the Enquête Emploi is that it provides detailed information on individual characteristics that affect hiring and layoff. Information on the firm (firm size, industry) is also available. Particular attention is paid to the definition of the employment status. Indeed, only workers under indefinite duration contracts in the private sector are liable for the tax. Due to a change in the design of the Enquête Emploi, this status has to be inferred from several questions (see Behaghel et al., 2005).
8
We end up with two samples. The first sample of unemployed workers is used to study the effect of the Delalande tax on hiring; it includes all workers observed at two successive interviews and who were unemployed and aged between 45 and 54 at the first interview. Descriptive statistics are given in table A1. They show the low levels of education among older unemployed workers, and the large share of long-term unemployed. The second sample is used to study the impact of the tax on layoffs; it includes all workers surveyed at two successive dates and who were employed under an indefinite-term contract and aged between 40 and 58 at the first interview. Table A2 displays descriptive statistics on that sample.
4. The effect on hiring Empirical strategy The identification of the effect on hiring uses the fact that the 1992 reform increases the tax schedule while exempting workers hired when they are more than 50 years old. As a consequence, the expected labor cost of a worker under increases, while the expected cost of a worker over 50 decreases.6 As the relative cost of workers under 50 increases compared to those over 50, standard labor demand theory predicts that the firm should substitute workers over 50 for workers under 50. To test that prediction, the strict application of the regression discontinuity approach would imply estimating:
Δ(ε ) post = (E(h (t,50 − ε )) − E(h (t,50 + ε ))) t ≥1992 , where h is a binary variable that denotes the occurrence of a hiring at date t, for a worker aged just below (or just above) 50, with ε being as close as possible to 0. However, two considerations lead us to increase the age window ( ε). First, the effects of the 1992 reform also vary with age within the under-50 and over-50 age categories. Consider the tax exemption: it concerns workers aged over 50, but among them, those aged 54-55 are the most impacted as they were the closest to being liable for the tax (remember that, before 1992, the tax started at age 55). Taking into account this discounting effect, we thus expect the highest increase in transition to employment for unemployed workers around 54-55 and the lowest for workers aged just above 50. The second part of the 1992 change is the increase in the tax. This concerns only unemployed workers below 50. Among them, due to the same
Note that, by contrast with severance payments whose effects can be undone by the side payments between the worker and the firm (Lazear, 1990), the tax is due to the unemployment insurance is therefore at least partly paid by the firm.
6
9
discounting reasoning, we expect the decrease in transition rates to be highest for those closest to the protected age, i.e. those just below 50. To clarify the two effects a bit more formally, take a very simple model with exogenous7 job destructions. All jobs generate a surplus p each year which is assumed to be independent of age. Jobs are destroyed at rate d, and all workers retire at age A. Time is discounted at rate ρ. If the job is destroyed before A, a tax f(a), varying with age a, has to be paid to the UI: this is the Delalande tax. In that setting, the expected surplus of a job can be written recursively as
S (a) = p + ρ ((1 − d ) S (a + 1) − d × f (a + 1)) if a
