A Study On Futures Arbitrage Strategy With Neural Network Model

With the development of China’s futures market, futures plays an important role inserving the real economy. Good strategy in futures market can help enterprises lock thecost of production. Hedging of the upstream and downstream goods can help enterprisesavoid the fluctuations of the commodity price. Chinese soybean enterprises pay greatattention to international soybean price because the amount of the imported soybean canbe almost4/5of the total soybean consumed. The violent fluctuation of soybean priceexposes Chinese soybean enterprises to huge operational risk. As China’s futuresmarkets of soybean, soybean meal and soybean oil are becoming more and more mature,soybean enterprises can use the futures market to avoid the price risk. Comparing withstudies on advanced arbitrage and hedging strategies in foreign countries, China’sscholars are still focusing on models theoretical study, rather than the empirical study.Summarizing the early researches of arbitrage and the artificial neural network andbeing based on co-integration theory, this thesis studies arbitrage opportunities ofsoybean, soybean meal and soybean oil futures. This thesis uses fair valueco-integration model and artificial neural network model to develop a suitable arbitragestrategy as well as explain it. The structure of the paper is as follows. The first part isintroduction, which explains the background and the aim of the paper, then points outthe innovation and the method. The second part of the paper is literature review，whichreviews the development of arbitrage strategies in the foreign market and the using ofartificial neural network in the finance markets. The third chapter introduces theco-integration theory, which is the foundation of building an arbitrage strategy.Moreover, the co-integration theory plays an important role in guiding the risk controlstrategy. Later the chapter discusses the basic structure of the artificial neural networkmode and introduces Elman neural network. In chapter IV, the stationary of soybeanmarket is tested. The results indicates that these three price series were found to be I(1)，based on which, the paper analyzes the co-integration relationship of them. Meanwhile,the result of error correction model is used to control the trading time of arbitragestrategies. On the basis of the foregoing chapters, a fair value co-integration model andartificial neural network model are created to trade the soybean crush spread in the fifthchapter. Data were analyzed both within and outside the sample. The sixth part of thepaper is conclusion，based on which we make some advice for the further spreadtrading. The result of empirical study shows that long term co-integration exists amongsoybean, soybean oil and soybean meal, while the regression is quiet slow. By setting aseries of opening and closing rules, arbitrage strategies based on the co-integrationrelationship and Elman neural network can both gain positive revenue within andoutside the sample.