Stock Market Trend Prediction Research Based On The Analysis Of Deviated Features And Risk Preference

The stock market is an important means of corporate finance and investors’ investment, the research on stock prediction has great theoretical and practical significance to the investors, enterprises and formulating policies of the government. In the shanghai stock market, the stock price and technical indicators often appears the phenomenon of inconsistent tendency, which bring to the traditional stock price trend forecasting model is lack of interpretability and prediction effect is poor. Meanwhile, the index deviation function can guide investors to predict risk and look for buying opportunities, among them risk appetite can be understood as the degree of tolerance to departure from the technical indicators, the thesis combine stock market departure characteristics and investor’s risk appetite with the relationship between them to forecast the market trend, Specific content is as follows:Firstly, extracting the MACD deviated characteristics and calculating the degree of it, then according to the degree and closing price to carry out stock price trend forecast by using BP network. While the market risk appetite is high, the correlation between the deviated characteristics and stock price is weak, thus through the bayesian network to learn the relationship among the risk preference, deviated characteristics and stock price. According to the change of the relationship between risk preference and deviated features, the paper proposes a Risk Preference based on the Deviated Characteristics prediction Algorithm (RPDCA).Secondly, owing to the RPDCA algorithm introduces the risk preference on the basis of existing deviated features, and it is assumed that the type of investor’s risk preference remain unchanged in the process of deviating. In order to solve the above problems, the thesis introduce the Multiple Deviated Characteristics based on Varying Risk Preference Prediction Algorithm (MDC-VRPA). First, learning the relationship between a variety of indicators and stock price trend by bayesian network, selecting the most closely related nodes as the factors of risk preference through markov blanket. Then, fixing the sliding window within the time period of the stock market to forecast, extracting the streaming data and Streaming deviated characteristics of markov blanket node within the window, putting the streaming data into the risk preference measurement model to obtain the current risk preference. Then, taking the multiple streaming deviated features for homologous accumulation processing, So as to establish the final prediction model. Last, trough moving sliding window back to track and extract features, realizing the continuity of stock market forecast. The empirical results show that the MDC-VRPA algorithm has higher prediction accuracy and wider application range.