Research On Future Market Trend Based On Machine Learning

With the continuous development of financial markets and quantitative trading,high-frequency trading has become the core of quantitative trading development.More and more investors have increased their investment in high-frequency trading research and development,and are gradually turning to obtain a stable profit with high-frequency trading.The core of the high-frequency trading is to analyze the high frequency transaction data.Machine learning technology is one of the most important development technologies in modern statistical learning theory.By modeling and analyzing sample data,we can describe the patterns behind the sample data.In the field of quantitative investment and high-frequency trading,machine learning also has a very wide range of applications.This paper mainly studies the application of the machine learning algorithm on the study of the price trend in the field of high frequency trading in the future market.This paper first studies the random forest algorithm and support vector machine(SVM)algorithm which is widely used in the machine learning algorithm.Considering the characteristics of data timeliness in financial market,considering the information content of proximal data is higher than that of the remote,a weighted support vector machine model is constructed,which improves the accuracy and timeliness of model fitting by weighting the sample data with time.According to the research of price trend,we need to transform it into classification problem,and define price trend into three categories: rise,fall and stability in a quantitative way.Then the high frequency transaction data index system is constructed as the feature set of the sample data.The domestic gold futures contract has a large trading volume and high liquidity in the market,and uses the gold futures data to make an empirical analysis.First,we study the distribution of intraday data of gold futures,and use random forest model to extract index parameters and train parameters.It is found that there are strong regularity and stratified structure in the index system.The support vector machine model is optimized by the resampling method.Finally,the weighted support vector machine(SVM)is trained in this paper.The study found that high-frequency trading price trends and market entry volume have a high correlation,and have little relevance with the market price index.By comparing the training results of three kinds of models,it is proved that the weighted support vector machine improves the accuracy of model fitting to a certain extent,and also reduces the training time of the model.The weighted support vector machine model integrates the advantages of random forest without sample resampling,the advantage of training time and the high accuracy of support vector machine prediction.It has higher application value in practical applications.