The Research Of Dynamic Bayesian Network On Stock Price Dynamics And The Application In Chinese Stock Market

The Price-earnings ratio(PE)is one of the most widely applied tool for the firm valuation in a security market.It is also a basic indicator in fundamental investment analysis.In practice,the basic price-earnings ratio is usually estimated by subjective opinions of experts.Based on the theory of dynamic Bayesian network(DBN),this paper studies the PE ratio dynamic model based on DBN and designs the timing strategy based on the model.The main work of this article is the application of the timing strategy,which based on the price-earnings ratio DBN model,in stocks and investment portfolios.The fundamental price-earnings ratio estimated by this model can be used as a reference indicator for investors to make investment decision,and it can also be combined with the actual price-earnings ratio to design the timing strategy.The price-earnings ratio(or stock price)valuation model based on Bayesian inference has the following advantages:1)It has financial theory support:The behavior factors of abnormal stock price fluctuation and Mean regression theory;2)Bayesian framework is same for all stocks;3)Expert knowledge can be integrated into the model;4)Trading-operations are simple and practical.The application process of the model is following:1)Set up a dynamic Bayesian network that describes the P/E dynamic,use the forward-backward algorithm to derive the filtering formula and the smoothing formula.2)Model parameter learning:adopt the expectation maximization algorithm(EM)based on Maximum a posteriori(MAP)to learn parameters.3)Inference problem:after obtaining the model parameters,the model's smoothing formula and filter formula are used to estimate the fundamental PE ratio of the stock.4)Online filtering:using the mobile window to filter medium noise status online,design and execute 2 timing strategies based on the inferenced fundamental PE ratio.This article selected 15 stocks from Shanghai Stock Exchange and Shenzhen Stock Exchange for empirical analysis.The simulation transaction process mainly includes three steps:model parameter estimation,filtering and trading.The empirical timing strategy applies not only to single stocks,but also to portfolios which constructed by equal weights and market-valued weighted methods.The paper will validate the performance of the timing strategy in the portfolio.The empirical results of this paper show that applying a timing strategy to individual stocks,the earnings performance is better than the benchmark buy&hold strategy.At the same time,the use of 2 timing strategies in the investment portfolio,compared with the market portfolio and benchmark portfolio,further improving the excess return while control systemic risk.Equal weight portfolio have higher levels of returns,and other market weighted portfolio have better risk levels.