| As the first distributed virtual currency,bitcoin's price has reached new highs recently.The attention and recognition of bitcoin in the capital market has also been rising.A large number of funds are constantly pouring into bitcoin market.Compared with other financial products,bitcoin has some advantages that are suitable for portfolio construction and statistical arbitrage.This paper empirically analyzes the effectiveness of statistical arbitrage strategy in the bitcoin market by comparing the bitcoin market with various machine learning models and parameter optimization algorithms.The main contents of this paper are as follows:Firstly,this paper expounds the research background and significance of statistical arbitrage strategy of high-frequency data in bitcoin market,then summarizes the research status of scholars on bitcoin and statistical arbitrage strategy,and finally puts forward the main research contents and main innovation points according to the above contents.Secondly,it introduces the principle and characteristics of bitcoin,the classification of bitcoin futures and the meaning of various bitcoin futures contracts;Secondly,it introduces the idea and steps of statistical arbitrage strategy;Finally,this paper introduces the basic theory of random forest,decision tree,KNN learning,logistic regression and SVM,and the basic theory of grid search algorithm for parameter optimization.Thirdly,this paper constructs a bitcoin statistical arbitrage strategy based on machine learning model.As the input variable of the machine learning model,the technical analysis index takes the past and future contract price differences into account when labeling on the training set,and then sets the opening and closing operation.Finally,the classical performance index is used to analyze the arbitrage results.In the part of empirical comparative analysis,bitcoin weekly contract and bitcoin quarterly contract are selected as intertemporal arbitrage targets,and 1-hour high frequency trading data from April 15,2020 to July 10,2020 is selected as sample data.Firstly,the descriptive statistical analysis of the two contract data is carried out;Secondly,the sample data is divided into training set and test set,and the co integration test is carried out on the sample data.The results show that there is a co integration relationship between the prices of arbitrage portfolio;Then,five machine learning models are used to carry out intertemporal arbitrage.Each machine learning model includes two cases of considering and not considering the handling charges,so as to examine the sensitivity of the model to the handling charges.The results show that the stochastic forest model and decision tree model have the best comprehensive performance,and the yield and risk-taking ability are better than the other three models,In this paper,the random forest model is selected for further research and analysis.Fourth,the statistical arbitrage strategy of bitcoin based on grid search algorithm is constructed.The grid search algorithm is used to optimize the parameters of the random model,and the overall performance of the optimized parameters is improved.Then we use the optimized random forest model for empirical analysis of out of sample data,and get a good profit considering the handling charge.Finally,this paper draws the conclusion that the statistical arbitrage strategy based on machine learning model is effective against the high-frequency data of the special currency market. |
PDF Full Text Request