«Der Open-Access-Publikationsserver der ZBW – Leibniz-Informationszentrum Wirtschaft The Open Access Publication Server of the ZBW – Leibniz ...»
Der Open-Access-Publikationsserver der ZBW – Leibniz-Informationszentrum Wirtschaft
The Open Access Publication Server of the ZBW – Leibniz Information Centre for Economics
A Bayesian approach to determine
the impact of institutions on the
ZEW Discussion Papers, No. 10-058
Provided in cooperation with:
Zentrum für Europäische Wirtschaftsforschung (ZEW)
Suggested citation: Sachs, Andreas (2010) : A Bayesian approach to determine the impact of institutions on the unemployment rate, ZEW Discussion Papers, No. 10-058, http:// hdl.handle.net/10419/40159
erste Nutzung einverstanden erklärt.
zbw Leibniz-Informationszentrum Wirtschaft Leibniz Information Centre for Economics Discussion Paper No. 10-058 A Bayesian Approach to Determine the Impact of Institutions on the Unemployment Rate Andreas Sachs Discussion Paper No. 10-058 A Bayesian Approach to Determine the Impact of Institutions on the Unemployment Rate Andreas Sachs
Download this ZEW Discussion Paper from our ftp server:
ftp://ftp.zew.de/pub/zew-docs/dp/dp10058.pdf Die Discussion Papers dienen einer möglichst schnellen Verbreitung von neueren Forschungsarbeiten des ZEW. Die Beiträge liegen in alleiniger Verantwortung der Autoren und stellen nicht notwendigerweise die Meinung des ZEW dar.
Discussion Papers are intended to make results of ZEW research promptly available to other economists in order to encourage discussion and suggestions for revisions. The authors are solely responsible for the contents which do not necessarily represent the opinion of the ZEW.
Non-technical Summary Large diﬀerences in the unemployment rates of industrialized countries and the underlying causes of unemployment have been subject of recurring discussion for a long time.
Since the early 90’s, labor and product market institutions like employment protection legislation, the unemployment beneﬁt system or the labor tax system moved towards the center of attention. However, while theoretical papers provide clear predictions about the impact institutional factors should have on the labor market, empirical contributions disagree as to which factors are of empirical relevance.
One major problem is that institutional concepts used in theoretical work like the bargaining power of workers are unobservable in reality. Although a large number of indicators is available, the low number of observations prevents the inclusion of all of them. Empiricists have therefore to decide which indicators to use in order to capture the eﬀect of a speciﬁc institutional concept. However, this pre-selection could give rise to model mis-speciﬁcation and biased results. This paper oﬀers a solution to this problem by using a bayesian model averaging approach. The major advantage of this method is that a large set of institutional indicators can be tested for signiﬁcance without running into a degrees-of-freedom problem and without requiring to specify one particular model. Rather, information of a large number of models and, particularly, model uncertainty can be taken into account.
Labor and product market regulations aﬀect the unemployment rate of a country without doubt. Econometricians, however, have yet to establish an unequivocal signiﬁcance of this impact. Model mis-speciﬁcation, one of the main underlying problems, is overcome by adopting a Bayesian Model Averaging approach. I apply this method to a panel data set that covers 17 OECD countries for the time period from 1982 to 2005 and for up to 20 potential explanatory variables. 8 institutional indicators are identiﬁed as signiﬁcant determinants of unemployment. Endogeneity due to reverse causality is also considered by applying an instrumental variable estimation approach.
JEL classiﬁcation: C33, E02, E24 Keywords: Unemployment, Institutions, Labor and Product Markets, Model Averaging ∗ I am grateful to Wolfgang Franz, Jan Hogrefe, Marcus Kappler, Hans-J¨rg Schmerer, Andreas o Schrimpf, and Werner Smolny for very helpful comments and suggestions. I also thank the participants of the SMYE 2010, the ARGE Workshop 2010, and the ZEW Conference on Recent Developments in Macroeconomics 2010.
† Centre for European Economic Research (ZEW), P.O. Box 103443, D-68034 Mannheim, Germany, Phone: +49/621/1235-145, Fax: +49/621/1235-223, E-mail: firstname.lastname@example.org 1 Introduction Over the past two decades a number of theoretical and empirical studies sought to identify how labor and product market institutions aﬀect the labor market performance of a country. In this context, factors such as the employment protection legislation, unemployment beneﬁts and entry barriers for ﬁrms have been considered. Theory predicts a clear impact of institutions on labor market performance. Empirical evidence, however, is yet to conﬁrm the predictions of the theory. The empirical ﬁndings fail both to distinguish the crucial from the less important institutions but also to determine whether deregulation lowers or raises unemployment. While the inconclusive results can be partially explained by diﬀerences in the time period or the country selection, model mis-speciﬁcation also seems to be an important source of error.
When it comes to the explanation of unemployment, specifying the model correctly is a challenging task. Researchers beneﬁt from a large pool of potentially signiﬁcant institutional factors, which also makes the decision on which to include or neglect more diﬃcult.
In general, institutional variables can be divided into ﬁve groups: tax system, employment protection legislation, workers’ bargaining power, product market regulation and unemployment compensation. Each group contains several indicators, measuring diﬀerent aspects. For instance, bargaining centralization, bargaining coordination, minimum wages, union density and union coverage serve as indicators for the bargaining power of workers.
In principle, one could estimate a single model containing all explanatory variables and let the data sort out the important factors. Availability of macroeconomic data is unfortunately limited to few periods and countries. This prevents me from conducting the described method, since the model’s reliability would suﬀer from the small number of degrees of freedom. Therefore, in order to exploit the complete available set of data, and to separate the important and the less important indicators, a consistent analytical framework is required.
While the outcomes of one single model may not be unreliable, pooling information from a large set of models can improve the validity of the ﬁndings. Sala-i-Martin et al.
(2004) introduce such an approach based on a Bayesian-type model averaging, called BACE (Bayesian Averaging of Classical Estimates). The central idea is to estimate a large set of models containing a varying number of explanatory variables taken from the pool of all variables. The quality of a model j serves as a weighting coeﬃcient for the variables kj included in model j. Thus, variables which are incorporated in models with better ﬁt receive higher weighting than variables in models that exhibit smaller explanatory power. The weights of a variable over all models are summed up and serve as a measure for evaluating the importance of the factor in explaining the dependent variable.
I expect to identify for a given sets of countries and periods indicators for the variables that have contributed robustly to the determination of the unemployment rate. Furthermore, the direction of inﬂuence can illuminate the ongoing debate on employment-friendly deregulation reforms and the beneﬁts - if any - of certain institutional rigidities. More speciﬁcally, I hope to clarify which institutional changes in product and labor markets facilitate and which hinder the eﬀorts to lower the unemployment rate. Finally, the cross-sectional nature of the data set can shed some light on the question why some countries have lower unemployment rates than others. The last question concerns especially the debate between the ”employee-supportive” European system - featuring, e.g., high employment protection and considerable unemployment beneﬁts - and the more ”marketfriendly” Anglo-American system.
The paper is organized as follows. In the ﬁrst part of section 2, I give an overview on the ﬁndings of empirical studies dealing with the identiﬁcation of the direction of inﬂuence of institutional variables on the labor market. The second part of this section recapitulates the main theoretical considerations about the impact of institutions on the labor market.
Section 3 focuses both on the data as well as on the description of the applied econometric method. Section 4 shows the main empirical results with a corresponding discussion, while section 5 concludes.
2 Labor and Product Market Institutions
2.1 Inconclusive Empirical Results and Speciﬁcation Problems Among studies that have attempted to estimate the eﬀects of institutional factors on unemployment, Nickell et al. (2005) or Amable et al. (2007) in a dynamic setting, and Bassanini and Duval (2006) or Baccaro and Rei (2007) for static models are good examples for analyzes in a cross-country context.1 However, the results are far from being consistent, showing diﬀerences not only in terms of magnitude but also in terms of direction of inﬂuence. Howell et al. (2007) mention three main causes of the inconclusive results: the selection of the time period, of the institutional indicators, and the speciﬁcation of the model. While the ﬁrst point is hard to tackle since data limitations restrict the ﬂexibility of choosing the time period, the latter two aspects can A more comprehensive overview on cross-country studies dealing with the identiﬁcation of the institutional impact can be found in Table 3 in the Appendix.
be reasonably dealt with.
Researchers studying institutions are confronted with the problem of ﬁnding indicators which are reasonable proxies for an institution. However, for some institutional categories there are several indicators available which all capture some aspects of an institution. For example, labor taxes consist of taxes payed by the employer, by the employee, and by the consumer. The replacement rate can be split up into the beneﬁt payments for diﬀerent states of unemployment, say, ﬁrst year of unemployment or ﬁfth year of unemployment.
The bargaining system can be displayed by, for instance, both the union coverage and the bargaining coordination. Furthermore, empirical investigations have generally focused on the impact of labor market institutions on the labor market. The role of product market regulation for the determination of unemployment has been considered by, for instance, Nicoletti and Scarpetta (2005) or Griffith et al. (2007). However, to my knowledge there are no contributions which particularly deal with the identiﬁcation of eﬀects on a less aggregate level since in most cases an overall measure of product market regulation is used.2 This approach assumes homogeneity of diﬀerent regulation measures in terms of labor market eﬀects. The selection of indicators is closely related to the question of how to specify the model. A priori neglecting indicators would probably lead to omitted variable bias. Considering all available indicators in one model is often not possible due to the small size of observations in cross-country panels. An incorrectly speciﬁed model probably provides misleading results, both concerning signiﬁcance and direction of inﬂuence.