Benchmarking Regions: Estimating the Counterfactual Distribution of Labor Market Outcomes

Local labor markets show strong regional differences in labor outcomes and economic conditions. Regional differences affect the welfare of local populations as well as the effectiveness of local policy interventions. In regional context, when assessing the labor market outcomes and the e ectiveness of policies, it is important to benchmark regional outcomes by the outcomes of comparable regions and by doing so to account for di erences in labor market characteristics. Put differently, when looking at the outcome of one region it is necessary to know the expected outcome distribution for a region with the same labor market characteristics based on the outcomes of comparable regions.

This paper develops a new non-parametric approach to assess the local labor market performance and to benchmark labor market regions based on kernel matching. We suggest to estimate the conditional quantile position of a region as a relative performance measure based on the counterfactual conditional distribution of outcomes for a group of comparable regions. To evaluate the possible absolute scope for improvement and deterioration in the outcome variable, an absolute performance measure is estimated as the difference between the observed outcome in one region and a reference point - we take the median - from the counterfactual outcome distribution. We implement different similarity measures capturing the spatial heterogeneity in labor market characteristics and propose two different ways to model spatial proximity - geographical distance and relative commuting flows. Accounting for spillover e ects from neighborhood is important since, in a regional context, the labor market situation in neighboring regions is likely to a ect the performance of the analyzed employment ofice area.

The new benchmarking approach is applied to 176 labor market regions in Germany during the time period of 2006 to 2008. The outcome variable of interest is the rate of hirings into regular employment among the unemployed (integration rate). A twodimensional leave-one-out cross-validation procedure regarding the prediction of the outcome variable is implemented in order to find an optimal weighting scheme between similarity in regional characteristics and spatial proximity. The results show that both observed labor market characteristics and spatial proximity are quite important to successfully match regions. Speciffcally, the modiffed Zhao (2004) distance measure and the geographic distance in logs work best in our applications. It turns out that the implied benchmark group of similar regions and the estimated relative and absolute performance remain quite stable over time. In addition, an examination of the matching quality shows only a negligible number of mismatch cases in some of the labor market characteristics considered.

Fitzenberger, Bernd and Marina Furdas (2012), Benchmarking Regions: Estimating the Counterfactual Distribution of Labor Market Outcomes, ZEW Discussion Paper No. 12-023, Mannheim.