Research On Price Forecast Of Stock Index Futures Based On Support Vector Machine

Price prediction of financial products has been one of the research topics of interest to many researchers.However,due to various factors such as macro regulation and control,economic market operation,and investor expectations,data in financial markets are usually more complex,with nonlinear,time-varying,and uncertain characteristics.This is still a challenging task.Stock index futures are the futures of the stock price index.Compared with the stock market,the price prediction of stock index futures is more complicated.Financial product price prediction methods can generally be divided into two categories.One is the fundamental analysis method,which uses the law of value to analyze certain factors that affect the price of financial products in order to predict future price trends.The other is technical analysis,which treats price prediction as a pattern recognition problem.In this framework,first use historical prices and certain indicators to build a prediction model,so that for new data,prices can be predicted based on the model.In recent years,with the rise of machine learning and artificial intelligence,scholars have gradually tried to apply machine learning methods to the price prediction of financial products.Support vector machine is a two-class classification model proposed by Cortes and Vapnik.Its basic model is a linear classifier with the largest margin defined in the sample space.When introduced the kernel method,Support vector machine has also achieved good results in nonlinear classification problems,and is a widely used machine learning algorithm.This paper uses support vector machine to model the data of China's stock index futures market,and verifies the prediction accuracy of the model on the test set.The kernel function determines the final performance of the support vector machine and the kernel method,and becomes the largest variable of the support vector machine.Unfortunately,how to choose the appropriate kernel function for a specific problem is still an open question.In practical applications,kernel functions are often selected based on experience and domain knowledge,but the effectiveness of selected kernel function needs to be verified through experiments.This method has large limitations.Especially when the sample features contain heterogeneous information or the sample size is large,or the data is unevenly distributed in the high-dimensional feature space,it is unreasonable to process all samples by single-kernel mapping.Multi-kernel learning is a research direction to deal with this problem in recent years.This paper uses multiple kernel functions and learns to obtain their optimal combination as the final kernel function instead of a single kernel function to build a combined kernel support vector machine model.This article uses the market data of the CSI 300 stock index futures main contract,divided into two time periods.The first piece of data is from April 16,2010 to May 1,2018,as a training set for parameter optimization and model building,and the second piece of data is from May 1,2018 to November 4,2019,as a test set,used to test the accuracy of the model.The selection of the input features has a great influence on the performance of the support vector machine model.The selected feature index should be closely related to the prediction target.At the same time,it should be noted that more features will increase the complexity of the model,increase the calculation overhead,and may cause overfitting.This article selects the seven basic market indicators of opening price,highest price,lowest price,closing price,trading volume,trading amount,and holding volume as input features.This article considers the classification of support vector machines,so when the closing price rises or is flat,the output index is +1,otherwise it is-1.Parameter optimization is a research focus of this article.The learned model performance is often significantly different with different parameter configurations.If the parameters are not optimized properly,the performance of the combined kernel function may even be lower than that of a single kernel function,and lost its meaning.For the combined kernel support vector machine model,this paper uses particle swarm optimization to optimize the parameters.The combined kernel support vector machine model established with the optimal parameters obtained by the optimization achieves higher prediction accuracy on the test set than the traditional single-kernel support vector machine model.Therefore,this article has achieved the expected results and enriched the methods for stock index futures price prediction.