Comparison of Strategies for Multi-step-Ahead Prediction of Time Series Using Neural Network

Abstract

If the one-step forecasting of a time series is already a challenging task, performing multi-step ahead forecasting is more difficult. Several approaches that deal with this complex problem have been proposed in literature: recursive (or iterated) strategy, direct strategy, combination of both the recursive and direct strategies, called DirREC, the Multi-Input Multi-Output (MIMO) strategy, and the last strategy, called DirMO which aims to preserve the advantageous aspects of both the Direct and MIMO strategies. This paper aims to review existing strategies for multi-step ahead forecasting using neural networks and compare their performances empirically. To attain such an objective, we performed several experiments of these different strategies on three datasets: NN3 competition dataset, the Vietnam composite stock price index (VNINDEX) and the closing prices of the FPT stock. The most consistent findings are that the DirREC strategy is better than all the other strategies for multi-step ahead forecasting using neural network.