Developing models for accurate natural gas spot price forecasting is critical because these forecasts are useful in determining a whole range of regulatory decisions covering both supply and demand of natural gas or for market participants. A price forecasting modeler needs to use trial and error to build mathematical models (such as ANN) for different input combinations. This is very time consuming since the modeler needs to calibrate and test different model structures with all the likely input combinations. In addition, there is no guidance about how many data points should be used in the calibration and what accuracy the best model is able to achieve. In this study, the Gamma test has been used for the first time as a mathematically nonparametric nonlinear smooth modeling tool to choose the best input combination before calibrating and testing models. Then, several nonlinear models have been developed efficiently with the aid of the Gamma test, including regression models; Local Linear Regression (LLR), Dynamic Local Linear Regression (DLLR) and Artificial Neural Networks (ANN) models. We used daily, weekly and monthly spot prices in Henry Hub from Jan 7, 1997 to Mar 20, 2012 for modeling and forecasting. Comparison of the results of regression models show that DLLR model yields higher correlation coefficient and lower MSError than LLR and will make steadily better predictions. The calibrated ANN models show the shorter the period of forecasting, the more accurate results will be. Therefore, the forecasting model of daily spot prices with ANN can provide an accurate view. Moreover, the ANN models have superior performance compared with LLR and DLLR. Although ANN models present a close up view and a high accuracy of natural gas spot price trend forecasting in different timescales, their ability in forecasting price shocks of the market is not notable.

Keywords