The Nonlinear Method Research Of Forecasting Of Stock Price

Stock is the product of the market economy and influences the hearts of tens of millions of investors from the day of its birth on. High-risk bond that offers a high return is the stock investment characteristics, personal investment and institutional investors trying to forecast the development trend of the stock through always concerned about the market of stock and analyze the financial data, but the stock market is a highly complex nonlinear dynamic systems, the changes have their own internal laws, and also by market, economic, social and other factors, a few statistics based on the traditional method of quantitative prediction on the stock market prediction of the effect is not very significant, therefore, to find a suitable price prediction method it is extremely necessary. Neural networks is a self-organizing adaptive, non-linear characteristics and can be automatically extracted the financial activities of the inherent laws from historical data, but neural networks also difficult to determine the presence of network structure and network to the right to set the initial value is very sensitive and it is very easy to fall into the local network training solutions, Therefore based the stock market of nonlinear dynamic characteristics and the predictability of stock for the foundation, the papers tried to use recurrent neural network to the nonlinear function approximation capability, its own dynamic characteristics and the genetic algorithm optimization of the overall ability, and build Based on genetic algorithm and improved recurrent neural network (GA-Elman) nonlinear price forecasting model. The main work of the c includes:First, The use of the gradient algorithm in an improved Elman recurrent neural network value the right to amend the method is derived specific research and experimental studies. Second, the genetic algorithm using real number to encode, using a choice of roulette wheel, through cross-hybridization arithmetic, the use of non-uniformity variation method variability, and the process of evolution to retain the best individual strategies to global optimization design the improved Elman neural network and network structure of the initial value.Finally, about the choice of factors about the input of model, this paper chooses a group of effective input combinations through analyzing the impacts of the historical stock price data, technical indicators and other factors.A GA-Elman nonlinear neural network model of stock price forecast is established through the above-mentioned study. In order to investigate the paper stock prediction model predictive ability, This paper uses GA-Elman stock price forecasting models on the Shanghai Stock Exchange Ashare China Petrochemical (August 1,2007 to January 18,2008), Lotus Group (January 16, 2007 to August 7,2007) and the Shenzhen Stock Exchange Ashare of the three spent shares (January 4, 2007 to July 13,2007), Vanco (August 7, 2008 to January 8,2008) four of the 110 stocks trading days, Using the next two trading days of highest price for the forecast target, the 70 trading days as modeling samples to establish a stock price forecast model about in the next trading day, and uses this model to forecast modeling test 30 independent samples. The test results of China Petrochemical Ashares on the Shanghai Stock Exchange and Sanhua Ashare on Shenzhen Stock Exchange as follows:(1) About the China Petrochemical modeling test, the average absolute error of GA-Elman model is 0.18525 yuan (training times to 2,500 times) in which the smallest absolute error is 0.00015 yuan and the largest absolute error is 1.30959 yuan. Further, the number of absolute error between 0 yuan to 0.15 yuan is 21, and between 0.15 yuan to 0.3 yuan is 4, and the numbers of absolute error between 0.3 yuan to 0.45 yuan, between 0.45 yuan to 0.6 yuan and over than 0.6 respectively are 0, 3 and 2. (2) About the Sanhua modeling test, the average absolute error of GA-Elman model is 0.10242 yuan (training times to 2,500 times) in which the smallest absolute error is 0.00039 yuan and the largest absolute error is 0.39480 yuan. Further, the number of absolute error between 0 yuan to 0.15 yuan is 16, and between 0.15 yuan to 0.3 yuan is 9, and the numbers of absolute error between 0.3 yuan to 0.45 yuan, between 0.45 yuan to 0.6 yuan and over than 0.6 yuan respectively are 4, 1 and 0.To verify the validity of the GA-Elman model in further. The paper also built the improve Elman neural network model to predict the same four stock shares and the same test data of the four stock shares in 30 separate test samples modeling, The test results of China Petrochemical Ashares on the Shanghai Stock Exchange and Sanhua Ashare on Shenzhen Stock Exchange as follows:(1)About the China Petrochemical modeling test, the average absolute error of Elman model is 0.34986 yuan (training times to 3,000 times) in which the smallest absolute error is 0.01529 yuan and the largest absolute error is 1.69272 yuan. Further, the number of absolute error between 0 yuan to 0.15 yuan is 9, and between 0.15 yuan to 0.3 yuan is 7, and the numbers of absolute error between 0.3 yuan to 0.45 yuan, between 0.45 yuan to 0.6 yuan and over than 0.6 yuan respectively are 7, 4, and 3.(2) About the Sanhua modeling test, the average absolute error of Elman model is 0.28174 yuan (training times to 3,000 times) in which the smallest absolute error is 0.03096 yuan and the largest absolute error is 0.93834 yuan. Further, the number of absolute error between 0 to 0.15 yuan is 10, and between 0.15 to 0.3 yuan is 11, and the numbers of absolute error between 0.3 to 0.45 yuan, between 0.45 to 0.6 yuan and over than 0.6 respectively are 3, 3 and 3.By comparing the predicted results of the two nonlinear price forecast model, we have that the four stock price'predicted precisions of the GA-Elman nonlinear forecasting method is high, and its properties in all respects obviously are better than the improved Elman nonlinear forecasting method. The above results of the statistical analysis show that the research results of the paper can provide new ideas and suggestions for the stock market of individual investors and institutional investors, investment activities and investment decision-making.