A Study On Pairs Trading Based On Support Vector Machine And Convolution Neural Network

Pairs Trading is a trading strategy that makes use of temporary differences between long-term equilibrium financial asset prices.Pairs Trading is a market-neutral trading strategy in which arbitrage trading algorithms are used in two or more highly correlated financial products.Investors buy undervalued financial products while selling overvalued financial products.The market risk of financial products will be eliminated and the financial products that are bought and sold will return to a reasonable level of valuation to take profit.The emergence and development of pairs trading are both the result of innovation in financial theory and the inevitable outcome of the development of financial markets.Through the study of pairs trading strategies and methods,first,improve the efficiency of resource allocation in the financial market.Pairs trading stabilize price caused by speculation and other factors,correct distortions in financial market prices,improve the efficiency of financial markets.Second,improve the liquidity of financial products.Pairs Trading increased the trading volume of financial products and contributed to the successful formation of transactions.Pairs Trading have attracted investors to participate in the financial markets and have significantly increased the liquidity of financial products.Third,provide investors with new investment methods.Matching transactions have low-risk characteristics to avoid the risk of market downturns and to obtain a stable absolute return.This dissertation will be guided by the basic theory of financial markets,support vector machines and convolutional neural network theory,using literature analysis,inductive deduction,empirical analysis and other methods,based on the establishment of pairs trading strategies and management and management mechanisms,based on the analysis of cross-time,cross-product,cross-market financial markets match transaction application performance.This article will follow the logical line of "Theory Review-Trading Strategy-Risk Management-Empirical Research-Conclusion".The dissertation introduces the research background,research significance,research evaluation at home and abroad,and puts forward the research content and technical route of the dissertation.The basic theory of pairs trading,effective market theory and market vision are introduced,and the market basis and significance of pairs trading are proposed.The co-integration theory is a key method to find matching trading objects.Only when there are two or more financial products in the co-integration relationship can the pairs trading performed.Backpropagation neural network,support vector machine and convolutional neural network basic theory are mainly used to identify pairs trading signals.Finally,the basic model of matching transactions is proposed.The principle of the matching trading strategy and the design flow are described.The execution flow of the transaction and the algorithm pseudo code are proposed.The execution flow of pairs trading strategy mainly includes parameter design,trading signal identification,transaction execution and fund management.Risk control measures are proposed by analyzing the risk types and risk measurement methods of pairs trading.The price fluctuations in the financial market are frequent and violent.Risk management is the key technology for pairs trading,and it is also the core issue throughout the entire pairs trading.The risk control measures for pairs trading mainly include risk process management,fund management,decision risk prevention and control and internal control restriction.Under the condition of cross-time pairs trading,a co-integration and back propagation neural network model(C-BPNN model)is proposed to identify trading signals,and the intraday high-frequency cross-time matching trading strategy of the CSI 300 Index futures is established,and performance analysis is conducted.Under the condition of cross-product pairs trading,a co-integration and support vector machine model(C-SVM model)was proposed to identify trading signals,and cross-product matching trading strategies of corn and corn starch futures were established,and performance analysis was conducted.Under the condition of cross-market product pairs trading,a co-integration and convolutional neural network model(C-CNN model)was proposed to identify trading signals,and Weichai Power A shares and H-share intraday high-frequency cross-market pairs trading strategies were established.And perform performance analysis.The dissertation mainly focuses on signal recognition and empirical research on pairs trading.The conclusions include:First,the C-BPNN model is used to identify trading signals in the high-frequency cross-time pairs trading strategy of the CSI 300 Index Futures Day.The empirical test found that the price series of IF1612 and IF1702 training set and test set are non-stationary time series;the yield series of IF1612 and IF1702 training set and test set are stationary time series.The IF1612 and IF1702 yield series training sets and test sets have a co-integration relationship.In the training data,when the buying threshold(BT)is-0.7,the selling threshold(ST)is 0.7,the take profit line(TP)is 1.0,and the stop loss line(SL)is-1.0,the profit and risk control are optimal.In the test data,the C-BPNN model has a higher winning ratio,a higher total profit,and a larger retracement than the co-integration model(C model),and the risk value decreases.Second,in the cross-product pairs trading strategy of corn and corn starch futures,the C-SVM model is used to identify trading signals.The empirical test found that the price series of the training set and the test set of C1701 and CS1701 are non-stationary time series;the yield series of the training set and test set of C1701 and CS1701 are stationary time series.The training set and test set of the C1701 and CS1701 yield series have a co-integration relationship.In the training data,when the buying threshold(BT)is-1.5,the selling threshold(ST)is 1.5,the take profit line(TP)is 1.0,and the stop loss line(SL)is-1.0,the profit and risk control are optimal.In the test data,the C-SVM model has a higher winning ratio,a higher total profit,a maximum retreat,and a reduced risk value compared with the C model.Third,in the Weichai Power A-share and H-share intraday high-frequency cross-market matching trading strategies,the C-CNN model was used to identify trading signals.The empirical test finds that the price series of CH000338 and HK2338 training set and test set are non-stationary time series;the yield series of CH000338 and HK2338 training set and test set are stationary time series.CH000338 and HK2338 yield series training sets and test sets have a co-integration relationship.In the training data,when the buying threshold(BT)is-0.7,the selling threshold(ST)is 0.7,the take profit line(TP)is 7,and the stop loss line(SL)is-8,the profit and risk control are optimal.In the test data,the C-CNN model has the highest winning rate,the highest total profit,and the lowest risk value compared with the C model,the C-BPNN model,and the C-SVM model.