Forecasting regional labour markets in Germany: an evaluation of the performance of neural network analysis

The aim of the present paper is to forecast regional employment developments in the 327 West-German districts. Using a Neural Networks (NNs) methodology we try to identify the existence of underlying structural relationships between the input variables - data on regional and sectoral employment and wages - and the future development of employment at a district level. In order to offer reliable forecasts for the years 2000 and 2001, a variety of NN models has been developed and compared. The emerging results confirm the ability of NNs in capturing the complex data structures - in the training and test phases - and hence in 'extrapolating' useful information in a multi-regional context. Concerning the forecasting phases, our analysis highlights the necessity of carrying out further research experiments - by introducing additional economic background variables - in order to get more insight into the mechanism and structure of spatio-temporal employment data.