آزمایش بر روی تحریم های بیکاری و رفتار جستجوی کار

This paper presents the results of an experimental study on unemployment benefit sanctions. The experimental set-up allows us to distinguish between the effect of benefit sanctions once they are imposed (the ex post effect) and the threat of getting a benefit sanction imposed (the ex ante effect). We find that both effects matter. Moreover, the ex ante effect turns out to be substantial and bigger than the ex post effect. Benefits sanctions stimulate the outflow from unemployment.

Despite the fact that over the past decades unemployment went down in many OECD countries it is still high in most of these countries. This is mainly caused by the long duration of unemployment. Governments face the problem of how to design policies that bring the unemployed back to work more quickly. It turns out that many of the active labor market policies that are used (training, subsidized jobs etc.) have insufficient effects on overall unemployment (Martin and Grubb, 2001). One of the problems is that in many countries for those who lose their jobs there are substantial unemployment benefits that are provided for a long period. We show that benefit sanctions1 can help to reduce unemployment even if the expected benefit (over time) remains constant. That is, the uncertainty of the level of the benefit is an effective instrument to induce unemployed to accept jobs. In particular, compare two benefit systems. First, a constant benefit system that always pays a benefit equal to 60. Second, a sanction system which pays a benefit equal to 70 in 60% of the periods and 45 in the other periods (where 0.6*70+0.4*45=600.6*70+0.4*45=60; we interpret the difference between 70 and 45 as a benefit sanction). In the sanction system, the job acceptance probability turns out to be substantially higher. To illustrate, at a wage equal to 125 (out of a uniform distribution on [90,200][90,200]; see below for details), the (simulated) acceptance probability under the constant benefit system equals 9.4%. Under the sanction system it equals (on average) 23.5%. Note that this increase in acceptance probability is achieved without making the unemployed financially worse off on average. Further, our design allows us to quantify the so-called ex ante and ex post effects of the sanction system compared to the constant benefit system. The ex post effect is the change in job acceptance probability once a benefit sanction is imposed. The ex ante effect refers to the increased outflow from unemployment even before a benefit sanction is imposed. The mere threat of receiving a benefit sanction affects job acceptance behavior. In the example, the ex ante effect of introducing the sanction system equals (23.5-9.4=)(23.5-9.4=) 14.1 percentage points while the ex post effect—the difference between acceptance probabilities at benefit levels 70 and 45—equals 10 percentage points in this case. This suggests that evaluating a sanction system on the ex post effect only leads to an underestimation of the effects of such a system. The current evidence on the effectiveness of a sanction system is partly based on empirical research in which micro data about unemployment durations and benefit sanctions are used. Abbring et al. (2005) and Van den Berg et al. (2004) provide empirical evidence for the Netherlands. Jensen et al. (1999) does the same for Denmark and Lalive et al. (2005) presents an analysis for Switzerland. These micro econometric studies typically compare unemployed with and unemployed without benefit sanctions, taking into account that the process by which sanctions are imposed may have been selective. Therefore these studies provide estimates of the ex post effect.2 In contrast to our paper, these studies do not provide information on the ex ante effect. This effect can be formalized in a number of different ways. In papers like Boone and Van Ours (2006) and Boone et al. (2007) the ex ante effect appears because by searching harder the unemployed can reduce the probability of being sanctioned. The idea is that government officials monitor search effort. When the effort is (perceived to be) high, they conclude that someone did not get a job, because none were available, not because he did not try to find a job. Using this set up Boone et al. (2007) shows that it is optimal from a welfare point of view to introduce a system of benefit sanctions. Boone and Van Ours (2006) shows that the strengths of the ex ante effect and the ex post effect depend on the monitoring intensity. If this intensity is low, the ex post effect is more important, but if the monitoring intensity is high, the ex ante effect is more important. In the model presented in this paper, the unemployed cannot affect the probability of being sanctioned, except by accepting the current wage offer. This leads to slightly different results as explained below. As mentioned, this paper reports a laboratory experiment to analyze the effects of unemployment benefit sanctions. Previous experimental investigations of search models are reported in Braunstein and Schotter, 1981 and Braunstein and Schotter, 1982, Hey, 1982 and Hey, 1987, Cox and Oaxaca, 1989 and Cox and Oaxaca, 1992, and Sonnemans (1998). In general, all authors conclude that observed behavior is close to, but not fully consistent with, optimal search behavior. Braunstein and Schotter, 1981 and Braunstein and Schotter, 1982 test the theoretical implications of numerous variants of an infinite horizon sequential job search model. They observe that individuals react to variations of the environment as predicted by theory, even though the duration of search often falls short of the theoretically predicted duration. Additionally, they find reservation wages to drop over time, even though theory predicts constant reservation wages in the infinite horizon models that they use as benchmarks. Cox and Oaxaca, 1989 and Cox and Oaxaca, 1992 report job search experiments in a finite horizon model. They argue that the individuals in Braunstein and Schotter, 1981 and Braunstein and Schotter, 1982 experiments had not actually believed in the infinite horizon and had rather played a finite horizon game, in which optimal reservation wages fall over time. While they find search duration and income to be very close to the theoretical predictions with risk-neutral individuals (Cox and Oaxaca, 1989), they also find the directly elicited reservation wages to be lower than predicted (Cox and Oaxaca, 1992). They conclude that a model with risk aversion explains their observations best. Sonnemans (1998), however, shows that a fully rational model of risk-averse search is not consistent with the search strategies that participants choose in a finite search model experiment. Most individuals in his experiment use search heuristics that combine the rational marginal net benefit aspect with some satisfying rule that is applied to total income (i.e. they do not ignore the sunk cost of search). This observation is well in line with the results of Hey, 1982 and Hey, 1987 research on consumer search heuristics. Our paper contributes to the experimental literature on job search in three important ways. First, the model we use is especially suited for an experimental implementation, because it provides a constant reservation wage benchmark in a finite horizon setting.3 The models implemented so far had either the one or the other advantage, but not both.4 Second, we present the first experimental analysis of the effect of random benefit sanctions on job acceptance behavior.5 Finally, this study presents the first attempt to assess the relative importance of unemployment benefits in determining job acceptance as compared to the wage. We find that observed acceptance probabilities are lower than the theoretically optimal ones for a risk-neutral individual. This suggests that individuals are risk-seeking more than risk-neutral. Moreover, this effect is most pronounced in the treatment with constant unemployment benefit levels. Again this suggests that a system with constant unemployment benefits over time raises unemployment. While the combination of a positive ex ante effect with reservation wages that lie above the optimal risk-neutral reservation wages cannot be simply explained with uniform risk preferences in a classical decision-making model, more complex models such as Prospect Theory (Kahneman and Tversky, 1979) may organize the data better. Assuming that the current wage offer is the reference point, an individual who is risk-seeking in gains and losses, but also feels a strong disutility from benefit sanctions (loss aversion), theoretically may exhibit behavior that is consistent with the observed behavior. Finally, we find that individuals react as predicted to financial incentives in the form of wages offered or unemployment benefits provided. The set-up of the paper is as follows. In Section 2 we present the theoretical model underlying our experiments. Section 3 describes the experimental set-up. Section 4 presents the results of the experiments. Section 5 concludes. The appendices give the instructions used in the experiment and provide more details on prospect theory in our search context.

In this paper we analyze job search behavior in the presence of unemployment benefits and unemployment benefit sanctions. Our analysis is based on laboratory experiments in which the arrival of job offers is exogenous. We focus on wage offer acceptance behavior. In theory the job search behavior of an unemployed worker is influenced by a system of benefit sanctions in two distinct ways. First, if confronted with a reduction of the benefits, an unemployed worker will be more likely to accept a given job offer. This is the so called ex post effect. Second, due to the threat of receiving a benefit sanction a worker will be more likely to accept a given wage offer than he would be if his benefits would be constant and no system of benefit sanctions existed. This is the so called ex ante effect. Our experiments show that both ex post effect and ex ante effect are relevant. The magnitude of both effects depends on the structure of the system of benefit sanctions and on the specific wage offers. In most cases the ex ante effect (which is hard to estimate using econometric techniques on real world data) is larger than the ex post effect. In the experiment, we compare two unemployment benefit systems where one system has the same benefit every period and the other a lottery over a high and a low benefit. In both cases the expected unemployment benefit per period is the same but the outflow out of unemployment is far higher for the lottery scheme. The policy implication is that introducing the uncertainty associated with benefit sanctions can reduce unemployment even when the expected unemployment benefit level is unchanged.