Abstract
Using administrative data, this article measures the ‘effective’ duration of unemployment benefit recipiency. Our results suggest that this duration varies considerably when we use spells instead of data on individuals. The exit hazard rate from unemployment using spells is overestimated when compared to the one obtained with data on individuals and, therefore, the expected duration of unemployment benefit recipiency is underestimated.
Acknowledgements
José M. Arranz acknowledges financial support from the Ramón Areces Foundation and Carlos García-Serrano from the Ministry of Science and Innovation (National Plan, ECO2010-19963).
Notes
1See, for instance, Katz and Meyer (Citation1990); Stancanelli (Citation1998); Alba (Citation1999); Rosholm and Svarer (Citation2001); Røed and Nordberg (Citation2003); Jenkins and García-Serrano (Citation2004); Arranz and Muro (Citation2004, Citation2007); Fallick and Ryu (Citation2007); Alba et al. (Citation2012); Arranz and García-Serrano (Citation2013, forthcoming).
2 This model is preferred to the rest of parametric models (Exponential, Weibull, Gompertz, Log-normal and Log-logistic) because it exhibits the largest log-likelihood value and the lowest AIC and BIC (results not shown but available upon request).
3 We have also estimated the models incorporating unobserved heterogeneity (using a gamma distribution). The likelihood ratio test of a model with unobserved heterogeneity against the one without it suggests that the unobserved heterogeneity is not significant. Moreover, no differences are detected in the variables’ coefficients between both models.
4 Using the estimates parameters of , we have computed the predicted duration of unemployment recipiency for the explanatory variables. The results (not shown) confirm that the SM underestimates the predicted mean and median duration of unemployment spells for all the categories when compared to the IM.