Research On Analysis And Forecasting Of Stock Price Time Series

The stock market is one of the most important parts of the financial market. The trend and fluctuation of stock price can reflect the political, economic and social situations of a country. At the same time, with the development of economy and the improvement of people’s life, more and more people are involved in the investment and the stock is one of the most common investment ways. Therefore, stock price forecasting is a very important financial topic that it has practical significance in both the financial supervision of the government and the prevention of investors’risk on stock market to get most income. However, stock price is affected by a lot of factors, such as politics, economy, company situation, investor psychology and so on. Therefore, stock price series is complex, nonlinear and dynamic that it’s difficult to predict it accurately.The ARMA model is one of the most popular and widely-used time series models that can predict linear problem, while BP neural network is commonly used to process nonlinear problem. This paper firstly introduces the basic principle of these two single models, and realizes them to fit and forecast the stock price of Petro China and Microsoft Corporation by MATLAB simulation. However, because of the complexity of stock price series, a single forecasting model often can’t get satisfied result. In this paper, we propose two combined forecasting models, ARMA-BP model and ARMA-BP-Markov model. ARMA and BP neural network are used to solve the linear and nonlinear component of the stock price series respectively and Markov model can modify the result and make it more accurate. The experimental results show that the combined models outperform single ARMA or BP neural network, while ARMA-BP-Markov model gets more accurate result than ARMA-BP model.