Does labor market history matter?

Flinn and Heckman (1983) argue for a differentiation between unemployment and out of the labor force.

This benefit level is from 2003, which is the last year observed in the data.

If a person moves from an education to unemployment, the compensation is up to 82 % and the maximum amount in 2003 was 617 DKK.

The use of a hazard function as a flexible estimator of the wage distribution originates from Donald et al (2000).

See Honoré (1993) and Horny and Picchio (2010) for identification in a single-spell framework relying on the MPH assumption and see Abbring and Van den Berg (2003), Horny and Picchio (2010) and Picchio (2012) for identification relying on time-varying exogenous variables or multiple observations per individual.

See Rosholm (2001), D’Addio and Rosholm (2002a) and D’Addio and Rosholm (2002b) for methods to account for the initial condition problem.

Alternative one might consider following the procedures of Gaure (2012).

In order to test this initial condition assumption, the empirical model is also estimated with another version of the initial condition. In this version, the sample is selected such that in each year all individuals at the age of 16 are included in the sample and followed until the end of the sampling period. The assumption in this version is that there exists no relevant labor market history before the age of 16. This assumption is stronger than the one used in the paper. However, the data sample in this version will, by construction, include a lot of students, which will make the information that can be drawn from transitions in and out of the state outside the labor market harder to interpret. The sample also includes a lot of employment spells ex ante primary education that are not necessarily a relevant part of the individual employment histories.

However, the primary empirical results based on the two different data samples are the same. Hence, the initial condition used in the paper does not, in this sense, appear problematic.

Including the baseline hazards and unobserved effects, this gives a parameter space of 428 parameters where the coefficients are to be estimated.

The empirical model was also estimated with dummy variables for each of the five regions in Denmark, using the metropolitan region as reference. Since the four province regions seemed to react very similarly, the parameter space was simplified to include only one variable describing the meteropolitan area around Copenhagen.

See the Fig. 10 in Appendix 1 for an illustration.

Given the functional form of the hazard function, the effect in percent can be found by \((\)exp\((\beta )-1)*100\), where \(\beta \) is the estimate from Tables 3 or 4.

These include state dependence effects between the non-employment states.

The model was also estimated on a regional level and for different educational groups with similar results.

Except for the case of unemployed women, where employment experience has a negative effect on the transition rate into employment.