Semiparametric Bayesian time-space analysis of unemployment duration

Abstract

"In this paper, we analyse unemployment duration in Germany with filial data from. the German Federal. Employment Office for the years 1980-1995. These data, contain individual information about unemployment duration measured in months, covariates like age, nationality, gender, education and unemployment benefits, and in addition, calendar time of unemployment spells and the district where the individual works. Our goal is to. show how this detailed temporal and spatial information can be used to explore patterns of calendar time and small-scale regional effects on unemployment durations, and, simultaneously, to determine nonlinear functional for of duration dependence and age effects, and to estimate usual linear effects of other covariates. Application of conventional duration models becomes problematic, at least, because it is alpinist impossible to specify parametric functional forms or spatial patterns a priori. We therefore suggest and apply a semiparametric hierarchical Bayesian approach which ean cope with these tasks. Inference is fully Bayesian and uses recent Markov chain Monte Carlo techniques. Our results show that this approach arrows to develop realistically complex models; provides an appropriate tool for empirical analysis of spatio-temporal trends in labour markets and can be useful for generating economic hypotheses. In this paper, we analyse unemployment duration in Germany with filial data from the German Federal. Employment Office for the years 1980-1995. The... data, contain individual information about unemployment duration measured in months, covariates like age, nationality, gender, education and unemployment benefits, and in addition, calendar time of unemployment spells and the district where the individual works. Our goal is to. she-vi how this detailed temporal and spat information ean be used to explore patterns of calendar time and small-scale regional effects on unemployment durations, and, simultaneously, to determine nonlinear functional for= of duration dependence and age effects, and to estimate usual linear effects of other covariates. Application of conventional duration models becomes problematic, at least, because it is alinost impossible to specify parametric functional forms or spatial patterns a priori. %le therefore suggest and apply a semiparaxnetric hierarchical Bayesian approach which ean cope with these tasks. Inference is. fully Bayesian and uses recent Markov chain Monte Carlo techniques. Our results show that this approach arrows to develop realistically complex models; provides an appropriate tool for empirical analysis of spatio-temporal trends in labour markets and can be useful for generating economic hypotheses." (Author's abstract, IAB-Doku) ((en))

Cite article

Fahrmeir, L., Lang, S., Wolff, J. & Bender, S. (2003): Semiparametric Bayesian time-space analysis of unemployment duration. In: Allgemeines statistisches Archiv, Vol. 87, No. 3, p. 281-307.