Prediction of Short-term Natural Gas Prices Using Econometric and Neural Network Models

Abstract

Abstract The development of the natural gas industry in the last decade was shaped by deregulation from federal oversight. This created an environment with fierce competition but also highly volatile spot prices, which exposes market participants to a considerable price risk. Risk management becomes vital in such a business environment. An integral part of risk management besides understanding the market’s fundamentals is to develop quantitative models which help predicting gas prices or price changes based on market information. Current price prediction focuses on long-term equilibrium price forecasts, but does not give any information about the day-to-day volatile behavior of spot prices. For long-term delivery contracts this might be sufficient, but for trading on a speculative basis more accurate information about price movements is sought. This paper shows that a solution for day-to-day prediction of spot prices is feasible. The most appropriate models in achieving this task were selected to be an econometric model with lagged variables and a neural network model. As volatility of the spot price is usually higher in the winter and seasonality in the consumption pattern is given, the model was developed based on data for the winter. The performance of both models was tested in a simulated trading scenario and compared to a scenario where perfect prediction quality was assumed. The results showed that both models were profitable during the time span the test was conducted. The neural network showed better results than the econometric model. In comparison to the best case scenario the results were very pleasing. These models can provide a valuable, supportive tool for trading of natural spot gas in a speculative environment.