Inter-commodity spread trading using neural network and genetic programming techniques

Abstract

We employ the methods of neural network (hereafter NN) and genetic programming (hereafter GP) in this paper to construct a spread trading system, respectively, to forecast the trend of the price spread between Taiwan Stock Exchange Electronic Index Futures (hereafter TE) and Taiwan Stock Exchange Finance Sector Index Futures (hereafter TF). To forecast the trend of the spread, we use a variety of technical indicators as the inputs to our two models. We tend to long one contract and short another if the next-day return of the former is predicted to be larger than the latter. If the spread trend is predicted to change its direction, we close off the position and open a new position completely contrary to the closed one. We compare the trading performances of this momentum strategy to the day trade strategy, i.e. closing off our positions before the market close ever day. We find that the momentum strategy tends to outperform the day trade strategy and that the Back-Propagation NN (hereafter BPNN) model is superior to the GP model under both strategies whilst both are profitable.