Prediction of White Noise Time Series using Artificial Neural Networks and Asymmetric Cost Functions

Artificial neural networks in time series prediction generally minimise a symmetric statistical error, such as the sum of squared errors, to learn relationships from the presented data. However, applications in business elucidate that real forecastine rrroblems contain non-svmmetric errors. The costs by an experimental evaluation of neural networks trained with asymmetric cost functions in competition with expert sofhvare-systems for time series prediction in section 4. Conc~usions are given in section 5 . . 11. NEURAL NETWORKS FOR WHITE-NOISE arising from suboptimal business decisions based on overversus underprediction are dissimilar for errors of identical magnitude. To reflect this, a set of asymmetric cost functions is TIME SERIES PREDICTION used as objective functions for neural network training, deriving suoerior forecasts even for white noise time series, some Forecasting time series with non-recurrent artificial neural experimental results are computed using a multilayer networks (A") is generally based on modelling the network perceptron trained with various asymmetric cost functions, in analogy to an non-linear autoregressive AR(n) model [ 141. evaluating the performance in competition to conventional At a point in time I , a one-step ahead forecast j,,, is forecasting methods on a white noise time series extracted from computed using n observations y,, y,.,, ._ . ,y,.,,, from n the popular airline passenger data. preceding points in time I, I / , 1-2, ..., t -n+ / , with n denoting