A combined neural network model for commodity price forecasting with SSA

Commodity price forecasting is challenging full of volatility, uncertainty and complexity. In this paper, a novel modeling framework is proposed to predict the market price of commodity futures. Three types of commodity are selected as representatives: corn from agricultural products, gold from industrial metal and crude oil from energy. We decomposed the original series into independent components at various scales using singular spectrum analysis (SSA). A SSA-causality test is introduced to investigate the mutual influence between commodity futures prices. Additionally, using the SSA-smoothing scheme, we construct combined neural network models including back propagation, radial basis function and wavelet neural network to predict the commodity price. The experimental results illustrate that neural network models with the SSA outperform the benchmarks in terms of distinct measures.