ABSTRACT
The recent addition of ethanol futures to the CBOT exchange has provided market participants with further hedging and speculative opportunities. In particular, this spread provides farmers, commodity processors and grain elevators with a hedging tool to directly manage price risk exposure. This chapter investigates the ‘crush’ spread trade between corn and ethanol futures commonly known as the ‘corn crush’ spread. A spread trading system based on daily closing prices for each of the commodities over a five-year horizon is constructed using various neural network architectures. Multilayer perceptron (MLP) neural network (NN), recurrent neural network (RNN) and higher order neural network (HONN) architectures are all applied to the task of forecasting next-day spread returns. Results produced by each of these NN models are compared to linear trading models such naive and MACD trading strategies, as well as an ARMA model in order to measure effectiveness.
From the analysis the HONN outperforms all of the other forecasting methods in terms of both trading performance and statistical accuracy. The risk metrics volatility market timing filter also enhances annualized returns while reducing volatility and maximum drawdowns. Furthermore, the corn crush spread is found to display similar characteristics to the soybean crush spread as observed by Dunis et al. and as stated by the CBOT.
