Abstract
Natural gas is a major global energy commodity. Gas prices around the world face substantial volatility, inducing major downside market risks. Forecasting accuracy is thus a major concern for the consumers. Traditional econometrics models do not perform well due to inherent nonlinear and nonstationary gas price data. We thus propose an Autoregressive Neural Network (ARNN) model for forecasting daily spot gas prices. The model is benchmarked against the traditional Autoregressive Integrated Moving Average (ARIMA) model. Using a cross validation study, the ARNN model showed an improvement of around 33% over ARIMA in terms of mean squared error. This improvement is significant when price forecasts are used in gas purchase decisions.