Stock Price Forecasting Method Of Time Series Combination Model

With the rapid development of national economy, the stock market has becomean important part of our market economy. The most primitive and direct purpose ofthe almost investors is to get profit. To make profits as much as possible, Investmentneed reasonable actions of the investment decision-making.It requests us has a fulland effective analysis of historical stock data before a reasonable forecasts for theprice. The stock market is a complex nonlinear system, and its data is huge andcontinually change.Thus, it’s of important reference value and guidance to set up amoderate calculation speed and high precision automatic stock prediction model.Stock price series is a special type of time series, and it can be forecast withsome improved time series predicting methods. There is three method that commonlyused of time series: statistical model method, the machine learning method andcombination method. Simple statistical model method application in the complexbackground, which performance is often disappointing. At present more popular asthe machine learning method and combination method. Because combination methodis the organic integration of a variety of methods, matching the share prices of manyingredients consistent characteristics, this paper argues that combination method canmore effectively predict the stock price than the single method.Based on the well known three division of the stock prices, this paper try to builda framework of stock prices combination forecast method in which the finalforecasting result can be selected by certain rules. The final result is obtained from thekinds of result formed of kinds of methods used in different part, and it is morereliable than those obtained by the single prediction methods.On the basis of thisframework, this article also did the exploration and research to the The formationmechanism of the Some parts, and established the corresponding mathematical model.In this way, further improved the precision of decision system strong wave part of theprediction accuracy of forecasting module.Also introduces a kind of stochasticprediction correction method, hoping to further improve the reliability of the model.Some stocks or index daily closing price data are selected to be the object of aseries of comparative experiments from Shanghai A shares (group A total of946) andShenzhen A shares (group A total of1135). First of all, this paper make compare andanalysis the trend that different filter obtain; In the random-like forecast part, the new models proposed in this paper are analyzed and compared with the traditional ARmodel; And also this paper verify the effect of random prediction correction method;Finally the assembly experiment has been carried on in that the sub-results wereintegrated, also verified the validity of the selection principles in this paper which wasused to the selection of a combination of the sub-results. Through a series ofexperiments, The experimental data strongly verifies the rationality and validity of theassumption and methods of this article, and at the same time some problems exposeconcrete method.