Forecasting the Behavior of Multivariate Time Series using Neural Networks

This paper presents a neural network approach to multivariate time-series analysis. Real world observations of flour prices in three cities have been used as a benchmark in our experiments. Feedforward connectionist networks have been designed to model flour prices over the period from August 1972 to November 1980 for the cities of Buffalo, Minneapolis, and Kansas City. Remarkable success has been achieved in training the networks to learn the price curve for each of these cities, and thereby to make accurate price predictions. Our results show that the neural network approach leads to better predictions than the autoregressive moving average(ARMA) model of Tiao and Tsay [TiTs 89]. Our method is not problem-specific, and can be applied to other problems in the fields of dynamical system modeling, recognition, prediction and control