Author(s)

Xu Jiale, Zhong Weihua, Sun Yuan

Corresponding Author

Xu Jiale

ABSTRACT

Due to the influence of many factors, such as economic environment, political policy, market news and so on, it becomes very challenging to predict stock dynamics. The traditional methods of stock volume forecasting usually decompose stock trading volume, and then select appropriate models for different parts to model and forecast, but this method is difficult to grasp the intra-day periodic structure of stock trading volume. Five commonly used prediction methods for predicting stock price changes are studied, and the prediction analysis is carried out by gradually increasing the input dimension of the model. First of all, five optimized prediction models are established-autoregressive average model based on time series (ARMA), grey prediction model (GM (1), BP neural network model (BPNN), support vector regression (SVR) model based on improved grid optimization algorithm, long-term memory neural network model (LSTM), based on Tensorflow to study the model input of single dimension, that is, The closing price of each stock is used as the input of these five models. The principal component method is used to extract the features of multiple indexes of corn index, and then five kinds of neural network models are established by using the extracted principal components, and the opening price is predicted, and finally compared with the ARIMA model. The results show that the PCA-RNN model has achieved better results, is more suitable for the short-term prediction of stock prices, and can provide some reference for decision makers. It is found that the effect of LSTM-based machine learning algorithm is obviously better than other traditional machine learning algorithms. Then, the input dimension of the model is added, that is, 13 indexes that affect the stock price are used as the inputs of the LSTM model to predict the stock price. The mean square error of the model in the training set is 0.1438, which is compared with that of the BP network. The results show that the accuracy and stability of the LSTM network in predicting the intraday trading volume distribution are better than the BP network.

KEYWORDS

Stock forecast, Long and short time memory neural network, Regression analysis, Principal component analysis, The time series