Economic forecasting using neural networks
Bernd Freisleben, Klaus Ripper · 2002
In this paper, neural networks trained with the backpropagation algorithm are applied to predict the future values of three time-series relevant to assess the German economy: the gross national product, the unemployment rate, and the number of employees. The performance of the networks is evaluated by comparing them to appropriate linear regression techniques and ARIMA models. The comparison shows that the networks produce good results which are superior to those obtained by linear regression; the ARIMA models are better for predictions one time period ahead, but they are outperformed by the networks when predictions for several time periods ahead are made.