Research On Artificial Intelligence Algorithm For Stock Price Volatility Rules Prediction

With the emergence and development of the stock market, the stock trading of high risk and high return is closely watched by lots of investors and scholars. In order to increase the yield rate and decrease the risk rate, they desperately want to find the intrinsic regulation of the stock price volatility.However, there are numerous factors, which can affect the rules and characteristics of the stock price volatility, to make the stock price volatility prediction to be a difficult task. The properties, such as high turbulence,nonlinear, high noise and redundancy, and so on, of the stock market can decide the level of the complexity and difficulty of the stock price volatility prediction.The traditional prediction methods hardly train an accurate prediction model.Luckily, with the development of Artificial Intelligence, there are lots of existing algorithms and models, such as Random Forest, Support Vector Machine, BP Neural Network, RBF Neural Network, etc., can solve the stock price volatility prediction which contains many disadvantageous conditions,such as high turbulence, nonlinear, high noise and redundancy, and so on.In this paper, by carefully investigate the disadvantages of the existing stock price prediction models, we propose a new stock price volatility predictor,which can solve these disadvantages of the stock price volatility prediction.Firstly, we transform the tradition regression prediction problem, which is predicting the growth of stock price of each trading day, into the fresh two-class-based classification prediction problem, which is forecasting whether there is a trading day that the growth between the closing prices of it and the reference transaction is lager than the investor expectation value in several days or not. After the transforming of prediction problem, some disadvantages,including noise, redundancy and turbulence, may be decreased. Secondly, forfurther reducing the noise which embeds the original technical feature vector and improving the prediction accuracy, we employ RBM-structure-based deep learning method to extract the inherent characteristic information. Thirdly,Kernel-based Support Vector Machine and BP Neural Network algorithms are used to learn the knowledge of the stock price volatility from two different views. Support Vector Machine is a classical binary classifier which has good performance of the binary classification. Using kernel technique, Support Vector Machine can learn the nonlinear learning problem. BP Neural Network can directly learn a nonlinear network model which can find the inherent knowledge from the stock market. Lastly, Ada Boost ensemble algorithm is utilized to union Support Vector Machine and BP Neural Network. The last ensemble model can nicely enhance the stock price volatility prediction performance by combining the advantages of Support Vector Machine and BP Neural Network algorithms. 10-fold and Leave-One-Out cross-validation test results demonstrate the proposed stock price volatility predictor has a good performance. The proposed predictor has a certain reference value for practical stock price volatility prediction.