The Duration of Unemployment a Proportional Hazard Model with Non-stationary Inflow Rates

In this paper the effect of age, educational level, occupational group and regional labour market conditions on the probability of leaving the state of unemployment is estimated. We use a proportional hazard model based on aggregate data on incomplete spells of unemployment for males in the Netherlands. Effects of omitted variables and time dependence are studied in combination with a non-constant probability of becoming unemployed. The conclusion is that the subgroups identified with respect to age, education, occupation and region show very different patterns of the duration of unemployment. Especially age and education have a large effect on unemployment duration, but occupational group and regional labour market conditions do also play a significant role. Model specification tests lead to the conclusion that there is unobserved heterogeneity but no time dependence in our data set. The estimation results do not change if we incorporate the effect of a non-constant probability of becoming unemployed.