Forecasting economic turning points with neural nets
Richard Hoptroff, M.J. Bramson, Trevor J. Hall · 2002
The authors describe an approach to a difficult forecasting problem: predicting the turning points of the economic cycle. Neural nets are applied to nonlinear multivariate forecasting. The neural network architecture used was essentially a fully connected multilayer perceptron, or feedforward network. Specifically, gross domestic product (GDP) in the UK economy was forecast one year ahead. The approach is compared to the conventional approach of forecasting using leading indicators. Concurrent descent, a cross-validation approach to backpropagation, allows the network to be trained on very small, noisy data sets. An economic forecast is given for June 1991.>