Three-class Classification Models Predict The Movement Trend Of Stock Prices

As an important part of the financial market,the stock market is closely related to the wealth appreciation of stock investors,the financing cost of listed companies and the stable development of the country’s macro economy.Therefore,violent fluctuations in the stock market will not only damage the interests of stock investors,but also cause an imbalance in the industrial structure,and even interfere with the development of the national economy on a macro level.In addition,affected by various factors such as the company appears,industry development levels,and economic policies,the stock market has both non-stationary and nonlinear characteristics.This makes the prediction of stock price trend become a hot topic in academic circles.Among them,the prediction of the three movement trends of stock price rise,sideways and decline is more helpful for stock investors to make choices in all decision-making behaviors,namely buying,holding and selling stocks.In view of this,it is of great significance to establish an effective prediction model for the prediction of the three movements of stock prices.This paper attempts to use three kinds of three-class classification models(support vector machine,random forest and penalized trinomial logit models with LASSO)combined with technical analysis method to predict the stock price movement trend,in order to select the most effective stock price movement trend prediction model.First of all,based on Murphy’s technical analysis method,the prediction variables composed of multiple technical indicators are selected,and the three categories of stock price fluctuation are divided into three categories by using the quantile as the response variables.Secondly,under the condition that the selected individual shares are listed earlier and belong to different industries,the three individual shares of Ping An Insurance,Grand Industrial Holding and Shanying International Holdings randomly selected from the Fortune 500 in 2021 are taken as the research objects,and the data of the last 10% of each individual share from January 4,2010 to September 30,2021 will be used as the test set,and eight-ninths of the first 90% of the stock data will be randomly selected as the training set,the remaining one-ninth of the stock data is used as the verification set.Thirdly,in order to improve the prediction accuracy of the model,the parameters of three kinds of three-class classification models are optimized by using the verification set of three stocks.In particular,the radial basis kernel function with the strongest prediction ability is selected from the four kernel functions as the best kernel function of the support vector machine.Finally,under the optimal parameters of each model,the test set of three individual stocks is used to test the prediction effect of the model,and the corresponding time consumption is obtained according to the modeling and prediction process of the model.Therefore,the prediction effect and time consumption of the model are further used to comprehensively evaluate the prediction performance of the model.The evaluation results show that compared with support vector machine and random forest,penalized trinomial logit models with LASSO has the best prediction performance in the prediction of stock price movement trend of the three stocks,with the average and standard deviation coefficient of its accuracy are65.38% and 1.94% respectively;the mean and standard deviation coefficient of Kappa coefficient are 0.4781 and 3.76% respectively;the average total cost time is 5.951 seconds.Therefore,penalized trinomial logit models with LASSO may be effectively applied to the prediction of stock price movement trend.