Research On Cross-variety Arbitrage Trading Strategy Of Oil Futures Based On Improved GMDH Neural Network

Futures is a kind of financial derivatives with high liquidity and standardized management.With the development of China’s commodity futures market,investors began to use all kinds of investment strategies to invest.The use of these strategies has created benefits for investors and is also conducive to promoting the development of the market.With machine learning,data mining and other algorithms gradually applied in stock,commodity futures and other markets,how to use data mining model to build a more accurate strategy is the concern of investors and investment institutions.In the research and application of arbitrage trading strategy,mean regression is a classical model,but the mean regression trading frequency is less and the return is lower.Neural network is also used in the study of arbitrage strategy.At present,BP neural network and elm neural network are mainly used.However,in practical application,there are many parameters of this kind of neural network,which is prone to over fitting,leading to unstable performance.GMDH self-organizing network has the characteristics of good prediction performance and strong stability,and has been widely used in price prediction and other fields.Therefore,based on the structure principle of GMDH network and particle swarm optimization algorithm,this paper further optimizes the structure and parameters of GMDH neural network,and uses the network model to construct cross species arbitrage strategy,and at the same time,it carries out empirical research on the combination of oil products.The research in this paper provides certain theoretical and practical basis for cross species arbitrage.This paper mainly focuses on palm oil,rapeseed oil and soybean oil futures,and designs arbitrage trading strategies for these three kinds of oil futures,and carries out empirical verification.First of all,this paper studies the theory of cross species arbitrage and various data mining models and methods.Secondly,through the research on the fundamentals of the above three kinds of futures,it is proved that there is an internal relationship between the three kinds of oil,so it has theinternal basis and logic to design arbitrage trading strategy.Thirdly,this paper makes a co integration analysis of these three kinds of oil varieties from the data layer,and further studies the long-term equilibrium relationship among these three kinds of oil varieties.Fourth,this paper uses the mean arbitrage model to test the three kinds of trading.From the test,we can see that the traditional mean regression model has poor performance,there are problems of trading frequency rate and low return.Fourthly,this paper improves the GMDH neural network model,optimizes GMDH from the aspects of sample division method,external criteria,intermediate model,and constructs the trading strategy based on the optimized GMDH neural network.Fifthly,this paper collects data,makes an empirical analysis on the trading strategy of the improved GMDH neural network model,and verifies the effectiveness of the improved GMDH model by comparing with the improved GMDH neural network model and BP neural network model.After research,the following conclusions are obtained:(1)from the results of fundamentals,correlation analysis and co integration analysis,there is a long-term co integration relationship between palm oil,vegetable oil and soybean oil futures,which has the basis of arbitrage trading;(2)from the prediction results of the improved GMDH neural network model on price difference,the optimized GMDH has a good fitting effect on the profit,and at the same time,By comparing the prediction results in and out of the sample,we can see that the prediction out of the sample keeps stable,which shows that the optimized GMDH model can avoid the occurrence of over fitting problem to a certain extent;(3)the trading strategy based on the neural network model to predict the price difference and build,increases the frequency of trading,and filters the trading signal by introducing the threshold value,The stability of the measurement is further improved.From the back test results of trading strategy based on neural network model,the optimized GMDH neural network model has certain advantages in the overall performance of the strategy.The research of this paper has a certain reference value for expanding the design idea of cross breed arbitrage trading and improving the performance of cross breed trading strategy.