Re-employment probabilities of unemployment benefit recipients

Abstract

This article studies transitions out of unemployment for benefit recipients in Spain. We analyse the duration of unemployment, distinguishing between spells that end in recall (workers returning to the previous employer) and spells that end in exit to a new job. This distinction allows us to find that the recall hazard rate increases around the time of exhaustion of benefits. However, this happens only for workers receiving Unemployment Insurance (UI). Because we are unable to replicate this result for workers receiving Unemployment Assistance (UA), we believe the finding lends support to the hypothesis that in Spain firms and workers make a strategic use of UI.

Keywords:

• unemployment duration
• recall
• unemployment insurance
• unemployment assistance
• competing risks model

JEL Classification::

• C41
• J64
• J65

Notes

¹ Recent work has found recalls to be prevalent in Europe (Fischer and Pichelmann, 1991; Jansson, 2002; Jensen and Svarer, 2003; Mavromaras and Orme, 2004). See, in addition, Robertson (1989) for Canada and Katz and Meyer (1990) for the United States.

² See, for instance, Katz and Meyer (1990). For analyses of unemployment duration when benefits are near exhaustion – although the following studies do not distinguish between exits from unemployment through recall or through a new job – see, e.g. Fallick (1991) and Narendranathan and Stewart (1993), who show that the effect of unemployment insurance benefits on the hazard out of unemployment decreases over time, or Micklewright and Nagy (1998), Stancanelli (1999), Bratberg and Vaage (2000) and Puhani (2000), who do not find any rise in the hazard near benefit exhaustion.

³ Although Alba et al. (2007) also distinguish between recall and new job exits rates, our analysis differs from theirs in several dimensions. Apart from dividing the sample according to the type of benefits received by the unemployed (either UI or UA), we have information on the level of these two types of benefits and on the individual's unemployment duration beyond the exhaustion point. Therefore, we are able to know when the unemployed enter employment, either before or after exhausting the unemployment benefits.

⁴ Studies that analyse the effect of UA on exits from unemployment are scarce. See, for example, Earle and Pauna (1998), Erbenova et al. (1998), Stancanelli (1998), Micklewright and Nagy (1999) or Arranz and Muro (2007).

⁵ It should be stressed that recall is always just a possibility and never a certainty. Unfortunately, we only refer to temporary lay-offs in an ex post sense – i.e. job separations ending in recall. We have no information on ex ante temporary lay-offs – i.e. those that begin with a person expecting to be recalled. In any case, this ex post concept gives the proportion of unemployment from spells involving no job change (Feldstein, 1975; Clark and Summers, 1979), and it is not ambiguous in the sense that it is not based on whether individuals decide on what is a new employer and what is not (Alba et al., 2007).

⁶ As in most European countries; see Van Ours and Vodopivec (2006) for Slovenia; Winter-Ebmer (1998) or Lalive et al. (2006) for Austria or Fitzenberger and Wilke (2004) for Germany.

⁷ In any case, one may think that UA recipients in some senses may compete with UI recipients to be recalled. One way to test this possibility would be to include in the estimation for UI (UA) recipients a variable collecting the proportion of UA (UI) recipients that each company recalls in each period of time. However, this variable would be biased because our data set does not collect every unemployed being recalled by each company in the period of analysis.

⁸ We cannot distinguish between these two reasons for job termination. Nevertheless, we have information on the type of contract held by the individual in his previous job. Workers with permanent contracts are expensive to dismiss, so when they are laid off they are unlikely to come back or to be called back. In addition, a great majority of workers in Spain become unemployed because of the end of their contracts. We understand that the end of a contract is not exactly the same thing as being laid off. By contracting with fixed-term contracts, firms have no need for temporary lay-offs. Workers who quit their jobs (i.e. end their employment for voluntary reasons) are not considered in this study because they can access neither UI nor UA (Section II).

⁹ There is a small number of individuals with durations greater than 18 months. In order to avoid noise in the results, we artificially right-censor these observations at 18 months. In addition, although a spell of unemployment can end with an exit from the labour force, the data set does not include an identifier for this possibility. Anyway, observations beyond 18 months are censored, so we regard this inconvenience of the data set as being of minor relevance.

¹⁰ Although the replacement ratio is available, it has not been included because it is problematic to include both the time to benefit exhaustion and the replacement ratio as covariates. The reason is that the replacement ratio drops to zero after the benefits expire (so those variables are probably highly collinear and the estimated effects become imprecise).

¹¹ The oldest unemployed normally accumulate more labour experience, which generates a higher reservation wage (Folmer and van Dijk, 1988).

¹² It may be the case, however, that a worker with higher education is far below the category that would correspond to his formal education. For instance, an individual working in the lowest category, ‘labourers’, may well be in possession of an academic degree.

¹³ Tenure in the previous job is not included in our estimations because of the high correlation between this variable and the benefit entitlement variable (Section II).

¹⁴ We perform a test (available from the authors upon request) for the assumption of ‘Independence of Irrelevant Alternatives’ (IIA) through the Hausman test (Hausman and McFadden, 1984, HM) and Small–Hsiao test (Small and Hsiao, SH, 1985). In both tests, the null hypothesis of IIA is not rejected; therefore, the multinomial logit specification shows no indication that it is inappropriate for each arrival state (new job or recall). In addition, a Wald test and an Likelihood Ratio (LR) test are performed (also available from the authors upon request) in order to examine the null hypothesis that the coefficients of each category do not differ significantly from each other, for all the possible combinations. The rejection of the null hypothesis means that it is adequate to distinguish between exits into a new job and exits into a recall job; therefore, the competing risk specification seems to be appropriate, since none of the categories should be combined.

¹⁵ We use an identification theorem of Han and Hausman (1990), which gives conditions under which the competing risk model is identified even if the covariates for each risk are identical. The identification condition basically requires the presence of at least two continuous variables among the covariates. Our estimation includes several continuous variables: age, age squared and the regional unemployment rate.

¹⁶ A simple LR test of a model with unobserved heterogeneity against another without unobserved heterogeneity confirms the conclusion that unobserved heterogeneity is significant (these tests are available from the authors upon request).

¹⁷ Similar results are also detected by Katz (1986) and by Fallick and Ryu (2007), who in their analysis of unemployed persons with and without UI benefits find that the new job hazard rate more or less increases over the course of the unemployment spell and the estimated recall hazard rate exhibits negative duration dependence.