Artificial Neural Networks for Time Series Prediction - A Novel Approach to Inventory Management Using Asymmetric Cost Functions

Artificial neural networks in time series prediction generally minimize a symmetric statistical error, such as the sum of squared errors, to model least squares predictors. However, applications in business elucidate that real forecasting problems contain non-symmetric errors. In inventory management the costs arising from overversus underprediction are dissimilar for errors of identical magnitude, requiring an ex-post correction of the predictor through safety stocks. To reflect this, an asymmetric cost function is developed and employed as the objective function for neural network training, deriving superior forecasts and a cost efficient stock-level directly from the network output. Some experimental results are computed using a multilayer perceptron trained with different objective functions, evaluating the performance in competition to statistical forecasting methods on a white noise time series.