Research On High Frequency CTA Strategy Based On Machine Learning

In the field of quantitative investment with many varieties,CTA strategy is one of the mainstream quantitative investment methods applied in futures market.With the rapid development of the capital market,electronic trading system has been widely used by major exchanges,which makes the proportion of high-frequency CTA strategy trading increase.High-frequency CTA strategy is to obtain yields from bid-ask spread.Because of frequent trading in the futures market in a short period of time,the liquidity of the market is stronger.Under the background of the rapid development of computer technology,the original financial market analysis methods are gradually replaced by machine learning methods,and people expect that machine learning can play its unique role in investment decision-making and bring considerable returns.The research object of this paper is the high-frequency data in stock index futures.Firstly,the tick data in the limit order book is observed and analyzed to understand the data structure and distribution at the micro level.Secondly,in order to better capture the price trend in the high-frequency data,this paper constructs the price label based on the bid-ask spread crossing movement and the mid-price movement,and uses the some feature metrics that can capture the price momentum of the limit order book well,and on this basis,we can filter the feature metrics,which has achieved the purpose of simplifying the feature set.Finally,resampling the data to alleviate the impact of the imbalance of data categories.The machine learning model selected in this paper is XGBoost model,which has attracted much attention in recent years.It uses the advantages of its ensembles algorithm to model high-frequency data.In the experimental results of the model,it can be found that the methods of constructing labels,selecting feature metrics and data resampling can improve the model effect to a certain extent.After obtaining the prediction results of the model,the historical data is used for back testing of strategies,and the trade results achieve a positive return without calculating the handling charge.In order to make full use of the information in the historical data,the dynamic high-frequency trading strategy is also constructed.Finally,compared with SVM and Random Forest,XGBoost model has some advantages in both precision and efficiency,and this advantage is also reflected in the results of back testing of strategies.