«Der Open-Access-Publikationsserver der ZBW – Leibniz-Informationszentrum Wirtschaft The Open Access Publication Server of the ZBW – Leibniz ...»
The model averaging framework I apply in this paper enables me to tackle the aforementioned problems. I can include a large set of institutional indicators without being forced to preselect an indicator subset. Furthermore, the uncertainty of not knowing the true econometric model is particularly taken into account. Hence, the main contribution of the paper to the literature is to focus the empirical discussion on factors which indeed are correlated to the unemployment rate. This can help future research to conﬁne attention to relevant variables, leaving out the dispensable ones.
2.2 Theoretical Mechanisms
As argued by Daveri and Tabellini (2000) labor taxes are assumed to aﬀect the unemployment rate mainly by increasing the cost of labor and, thus, lowering labor demand. Furthermore, the eﬀect of a labor tax increase depends crucially on the degree of the workers’ bargaining power. The stronger the workers, the more of the tax increase the One exception is the study of Fiori et al. (2007).
ﬁrms have to bear. Hence, higher labor taxes only aﬀect labor demand, if the workers’ bargaining power is high. An important point is the utilization of the tax income by the government. If part of the taxes serve as funding for, say, qualiﬁcation measures for unemployed workers to reduce the spell of unemployment, taxes can indirectly help to reduce unemployment.
The wage level is mainly determined by the bargaining system and bargaining power of workers and ﬁrms. Blanchard and Giavazzi (2003) show that powerful employees (represented by strong unions) can demand relatively high wages, resulting in lower labor demand and a considerable rise in the number of unemployed. However, Calmfors (1993) point to the positive labor market eﬀects if unions take into account the negative consequences of their bargaining power. More precisely, if unions keep an eye at the unemployed persons, they will refrain from excessive wage claims for the employed with beneﬁcial eﬀects for the whole economy.
Furthermore, Stigler (1946) argues that minimum wages reduce labor demand by setting wages above a level which is justiﬁed by the workers productivity. Thus, especially low-skilled workers with wage payments below a (hypothetical) minimum wage are blocked out from the labor market. Yet, Manning (1995) shows in a model of shirking that minimum wages can decrease unemployment by lifting wages to a level at which shirking is less likely due to the increased incentive to work.
Providing employment protection is usually conducted by imposing severance payments on the ﬁrm or to exacerbate layoﬀs by legal regulations. Concerning unemployment, the eﬀect of higher employment protection is assumed to be twofold (see, for instance, Ljungqvist 2002). On the one hand, it lowers the ﬂows from employment to unemployment since ﬁrms take the additional costs for layoﬀs in consideration when evaluating the productivity of a worker. On the other hand, labor is allocated less eﬃciently what comes along with a fall in productivity and, ﬁnally, decreased labor demand.
While the employment protection can be seen as an insurance against getting unemployed, the unemployment beneﬁt system aﬀects predominantly those who are already out of work. According to Holmlund (1998), an increase of unemployment beneﬁts on the one hand causes unemployed persons who are eligible to beneﬁt payments to raise their reservation wage. On the other hand, unemployed who are not eligible to beneﬁts will have a higher incentive to accept a job in order to get qualiﬁed for the beneﬁt payments in case of future unemployment.
Furthermore, the fear of losing job-speciﬁc human capital can convince the workers to attach no importance to high unemployment beneﬁts (see Arulampalam 2001). In this case, unemployment beneﬁts will not lower the unemployed workers’ incentive to search a job.
Additionally, Acemoglu and Shimer (1999, 2000) mention the beneﬁcial inﬂuence of unemployment insurance if workers are risk averse. Then, high beneﬁts serve as an insurance against unemployment, and workers are willing to take jobs associated to higher unemployment risk but also to higher wages, higher job quality, and, eventually, increased output and lower unemployment.
The degree of product market regulation aﬀects labor demand through adjusting the competitive environment in a market. Blanchard and Giavazzi (2003) show that increasing competition in a market means more competition for labor if entry barriers are suﬃciently low. Hence, the lower the governmental regulative intervention (e.g. barriers to entry or public ownership) the lower the unemployment rate.
Nevertheless, a certain degree of barriers to entry can also help to increase the ﬁrms’ productivity in a market, as explained by Melitz (2003). This, in turn, can lead to an increasing demand for labor. Furthermore, a change from public to private ownership boosts the performance of workers and managers since monitoring is much easier to implement (see Schiantarelli (2008) for a discussion).
Besides the traditional institutional variables, some other factors have been brought into the discussion. While data on family policies, migration policies, education and training, active labor market policies, retirement programs as well as the regulation of working hours are only scarcely available, one other variable deserves to be recognized. Dromel et al. (2010) recently argued that credit market imperfections can slow down job creation by restricting access to money for ﬁrms. The more access to credits is restricted the higher unemployment should be. I consider this variable as a control factor.
The main purpose of the paper is to analyze the impact of labor and product market institutions under model uncertainty. In order to ensure comparability to earlier studies, I rely on established data sources on institutional characteristics.3 The existing data have been updated, resulting in a comprehensive and balanced data set from 1982 to 2005.
The econometric approach I use requires the application of a panel data without any Oswald (1997) argues that house ownership which is an indicator for the workers’ mobility can contribute to the explanation of unemployment. Nickell et al. (2005) tests this hypothesis and did not ﬁnd a signiﬁcant relationship. Additionally, data is only scarcely available and heavily interpolated.
Therefore, I do not consider this variable in my estimations.
gaps which is why only 17 OECD countries are included for the investigation. Data on the unemployment rates is taken from the OECD. I use the harmonized unemployment rates which are comparable over countries.4 The institutional indicators under inspection are brieﬂy described in the following. Further information on the construction and composition of the data set is given in the Appendix.
- The labor tax system is represented by the payroll tax, the income tax, the consumption tax, and the tax wedge which is the sum of the ﬁrst three taxes.
- Bargaining coordination and centralization, union density and coverage as well as the minimum wage all cover a part of the bargaining system and the workers’ bargaining power.
- The OECD provides an indicator for the strictness of employment protection.
- I construct indicators for the unemployment beneﬁt system according to Nickell and Nunziata (2001). Thus, I have an indicator for the replacement rate for the ﬁrst year, for the second and third year, and for the fourth and ﬁfth year of unemployment. Additionally, the OECD provides an overall indicator for the replacement rates which is the average of the three aforementioned partial replacement rates, and an indicator for the duration of payment which consists of weighted shares of the ﬁrst year and the fourth and ﬁfth year beneﬁts. Furthermore, I use a measure for the coverage of the unemployment beneﬁt system, i.e. how many unemployed are entitled to receive transfer payments.
- I use indicators for barriers to entry and for public ownership, as well as an overall indicator for the degree of product market regulation. The overall indicator is the average of diﬀerent partial indicators of product market regulation. Note, that the barriers to entry and the public ownership indicator are included in the overall indicator. The other parts of the overall indicator cannot be considered since data is missing for some countries or periods.
- According to Dromel et al. (2010), I include a measure for credit volume delivered to the private sector over GDP. The higher the value the lower are the constraints to credits.
I also run the estimations with data from the Labor Force Surveys collected by the ILO. The results concerning the signiﬁcance of indicators remain exactly the same.
3.2 General Empirical Strategy The annual unemployment rate is the dependent variable and will be regressed on several institutional factors which explain the long-run evolution of the unemployment rate, and on shocks capturing the short-run ﬂuctuations of the unemployment rate. Furthermore, I use the Within transformation to get rid of time- and country-speciﬁc eﬀects. The
equation can be expressed as follows:
U R = X1 β + X2 γ + ε, (1)
where U R is the unemployment rate N T x1, X1 is a N T xK1 matrix including all institutional factors which inﬂuence the unemployment rate in the long run, and X2 is a N T xK2 matrix containing macroeconomic shocks to capture short-run ﬂuctuations. β and γ are the corresponding coeﬃcient vectors of size K1 x1 and K2 x1, respectively. N is the number of countries and T the number of years.5 I follow Nickell et al. (2005) in considering four shock variables. More speciﬁcally, I include productivity shocks, labor demand shocks, real import price shocks and the real interest rate.6 Unfortunately, it was impossible to construct a money supply shock variable due to data constraints for the time frame required in this paper. However, the results in Nickell et al. indicate at most only slight importance of that shock.
3.3 Bayesian Model Averaging
Model mis-speciﬁcation can lead to severely biased results, mainly due to omitted variable bias, especially if theory does not provide a clear guide on which variables to include. For instance, the workers’ bargaining power is an important driver of the unemployment rate, but it is impossible to quantify. Measures like the union coverage or an indicator for the degree of bargaining coordination are often used to proxy the bargaining power. However, there are a lot of potential factors of inﬂuence and including all of them is risky due to limitations in terms of degrees of freedom. One possible solution to this problem is to avoid specifying a particular model. Rather, this model uncertainty is particularly taken into account by exploiting information of a large number of models. A particular model It is sometimes argued that institutions do not have the same impact in each country, i.e. the pooling assumption is invalid. The test result for poolability according to Baltagi (2003) mainly depends on whether I assume an F - or a χ2 -distribution. Furthermore, the test results might change when the set of explanatory variables is altered. The gain of lower variance due to pooling comes at the price of a potentially incorrect poolability assumption. This should be kept in mind when interpreting the results.
See the Appendix for further information on the construction.
consists of the ﬁxed regressors plus a random number of varying regressors. In Bayesian terms, the expected coeﬃcient value and the variance of variable can be calculated as