The principles and practice of time series forecasting and business modelling using neural nets

Abstract

This paper is intended as a ‘hands-on’ practical discussion of how and why neural networks are used in forecasting and business modelling. The need for forecasting is briefly examined. The theory of the multilayer perceptron neural network is then covered both qualitatively and in mathematical detail, including the methods of back-propagation of error and independent validation. The advantages of the neural net approach to forecasting, namely nonlinear modelling capability, plausible interpolations and extrapolations, robustness to noise, ill-conditioning and insufficient data, and ease of use, are discussed. Finally, some working notes are offered for the practical implementation of neural nets in forecasting, and four real-life examples are given from the pursuits of econometrics, sales forecasting, market modelling, and risk evaluation.