The Development of Regional Employment in Germany: Results from Neural Network Experiments

Abstract

This paper offers an overview of experimental results, based on neural networks (NNs) used to forecast regional employment variations in Germany. NNs are statistical optimization tools inspired by the functioning of biological neural networks. Their main characteristics are their non-linear and multiple-unit simultaneous data processing, as well as their ability to find functional relationships within the data. After a brief introduction and a methodological review of NNs, we present the results – for a set of NN models – based on regional data concerning full-time employment in Germany. The database used in our experiments consist of two panels of 326 and 113 NUTS 3 districts, which represent West and East Germany, respectively. Since the data sets display different time spans, West and East German NN models – also embedding also shift-share analysis components – were developed separately, in order to forecast employment growth rates for the years 2004, 2005, and 2006. The NN models are evaluated by means of appropriate statistical indicators. The paper concludes with theoretical, methodological and empirical observations in the light of future research developments.

Keywords: Neural Networks, Forecasts, Regional Employment

JEL Classification: C23, E27, R12