Empirical Research On High Frequency Statistical Arbitrage Strategies Based On Futures Industry Chain

With the continuous maturity and development of Chinese capital market,the trading volume of futures is increasing year by year.With the support of short selling mechanisms,statistical arbitrage strategies have become a research hotspot for investors in the futures market.Statistical arbitrage is an investment method that uses quantitative analysis to determine arbitrage assets portfolios with strong industrial chain relationships or price correlations,establish long-term equilibrium relationships between asset portfolios,and design effective trading rules to make profits.This arbitrage method does not rely on the subjective judgment of investors,and is usually based on statistically constructed price difference sequences to achieve arbitrage.This paper uses the economic relationships between commodities in the upstream and downstream of the industrial chain to construct an arbitrage asset portfolio.Taking the soybean press profit industrial chain as an example,using three futures varieties of soybean II,soybean meal,and soybean oil from the Dalian Commodity Exchange as research objects,we construct an arbitrage model based on press profit,an arbitrage model based on cointegration relationship,and a Kalman filter arbitrage model with time-varying coefficients,respectively,At the same time,corresponding trading signal rules are set for the three trading models,namely,the nonparametric method and the random model parameter method.The grid search method and parameter analysis method are used to give different trading thresholds for building and closing positions for different models and different frequencies.At the same time,in order to better capture the diurnal changes in asset prices,this article uses high-frequency data of 5minutes,15 minutes,and 30 minutes to conduct empirical research on three different models,and combines the Kelly formula in asset management.Different investment ratios are given for the training performance of data within the sample.Backtesting is conducted on data outside the sample,and results are evaluated and analyzed from the model level and data frequency level,respectively.Empirical research shows that the squeeze profit model and cointegration model have achieved positive returns in both intra sample and out sample backtesting,indicating the feasibility of high-frequency statistical arbitrage strategies in Chinse futures trading.The Kalman filter model has a strong time-varying coefficient,resulting in excessive arbitrage times,frequent stop loss phenomena,and poor earnings performance.In the model for determining trading signals,the trading thresholds determined by the nonparametric method can achieve the goal of improving trading times and returns.However,the random model parameter method determines that the trading signal has too few arbitrage times,and does not increase the trading times as the data frequency increases.The sample back test has a low probability of winning,and the results have a certain degree of randomness,so it is not suitable for highfrequency statistical arbitrage trading signal determination.Combining the fund management method to conduct a back test on data outside the sample,a lower investment ratio can achieve a higher return on investment based on a principal of 1million,indicating that the fund management method is effective in the implementation of statistical arbitrage strategies and can improve returns while avoiding risks.From the comparison of evaluation indicators,the arbitrage model based on cointegration relationship performs best in the backtesting of profitability and risk indicators,with annualized returns and annualized Sharp ratios higher than the squeeze profit model.From the comparison of trading frequency indicators,the 5-minute arbitrage frequency and winning rate are both the highest,and they are the best in terms of profitability and stability,with strong profitability and risk resistance,which proves the research value of statistical arbitrage strategies based on the futures industry chain under highfrequency data.