Modelling and trading the gasoline crack spread: A non-linear story

Abstract

This paper investigates the gasoline crack spread time series, using the non-linear cointegration method developed by Enders and Granger (1998, ‘Unit-root Tests and Asymmetric Adjustment with an Example Using The Term Structure of Interest Rates’, Journal of Business and Economic Statistics, Vol. 19, pp. 166–176). The spread can be viewed as the profit margin gained by cracking crude oil, and therefore any non-linearity can be interpreted in the context of the effect on market participants. Further, a number of non-linear neural networks are used to forecast the gasoline crack spread. The architectures used are multilayer perceptron, recurrent neural networks and higher order neural networks, these are benchmarked against a fair value non-linear cointegration model. The final models are judged in terms of out-of-sample annualised return and drawdown, with and without a number of trading filters. The results show, first, that the spread does indeed exhibit asymmetric adjustment, with movements away from fair value being nearly three times larger on the downside than on the upside. Secondly, the best trading model of the spread is the higher order neural network with the threshold filter, owing to a superior out-of-sample risk/return profile.