Research On Futures Price Trend Prediction And Quantitative Trading Strategy Based On Machine Learning

Quantitative trading is a kind of objective trading methods which combines mathematics and financial knowledge and realizes by computer programming.In recent years,various industries have been applying artificial intelligence methods,and quantitative trading is no exception,machine learning algorithms are the core of artificial intelligence.In quantitative trading,the targets are generally low-risk products such as stocks and bonds.Futures are rarely paid attention to by investors due to their high leverage,However,the functions of futures are very important to human production and life,So it is of great significance to predict the trend of futures price.This dissertation uses machine learning algorithms to build a price trend prediction model in futures quantitative trading,seek a better model and variety and proposes relevant quantitative trading timing strategies,and innovatively introduced the decision-making theory to evaluate the strategy from the perspective of investors.Firstly,according to the classification rules of futures market,selects CSI 300 IF stock index,Soybean,Shanghai copper,PVC and China crude oil as the research targets to observe the whole futures market.After obtaining relevant data,it uses technical indicators to preprocess the data;At the same time,label the data according to the rise,fall and shock;Then,seven mainstream machine learning algorithm models are constructed to train each futures variety,and it is found that the four layer Multi-Layer Perceptron on CSI300 IF stock index futures has the best effect,with an accuracy rate of 77.49%.Then,after obtaining the basic model above,introduces the K-line shape for feature engineering operation.The K-line shape has certain trend indication effect,most of them are used in the stock market,but not common in the futures market.Therefore,more than 60 kinds of K-line shape are introduced in the ternary way,and the data set is expanded to about 121 dimensions and standardized.The model is used to train the expanded data,and most models have improved prediction accuracy,which proves that the K-line shape has a certain indicative role and can provide certain reference for investors.Finally,a quantitative trading strategy is proposed.The four-layer Multi-Layer Perceptron model and the CSI300 IF stock index futures are used as the target to conduct simulated trading.The time is selected from January 2018 to November 2020,there are three ways to construct trading strategies,namely "long only","short only",and "long short",the final results are quite considerable,among which the "long short" method has the best effect,the annual return rate is 115.94%,Max Drawdown is 5.77%,and the benchmark return rate is 9%,which shows that the model is more robust.In addition,considering the different risk preferences of different investors,this dissertation sets up evaluation indicators,innovatively introduces decision-making theory,uses the Best Worst Method(BWM)to determine the weight of different investors' preferences for each indicator,and then gives comprehensive evaluation and suggestions.It is concluded that conservative investors should choose "short only" approach,while risk investors should choose "long short" approach.This dissertation studies quantitative trading with futures as the targets,compares the effects of different machine learning methods,and introduces the K-line shape to obtain the optimal model and futures.At the same time,it also innovatively introduces decision-making theory,and comprehensively evaluates trading strategies according to the different risk preferences of investors,which enriches the research ideas of quantitative trading.