Fuzzy neural networks for modelling commodity markets

The paper describes a method for forecasting the price of crude oil. This market sector is important for both, forecasting the market for trading the commodity (crude oil or its derivatives) itself, and for its effect on other market segments. For the presented approach a certain effect of two observable market states is used, which allows for establishing a combination of two neural networks, or specialist models, each of which is specialized on a different market state. The model used is a fuzzy neural network which is trained to determine the state of the market and then uses the output of the respective specialist model for forecasting. The two states are called contango and backwardation respectively and can quite easy be determined by looking at the prices of the two futures contracts which are due next. Establishing a model based on the market states allows to increase the accuracy of prediction (in comparison to a classical model)