Construction And Application Of AP-FIG And Dkipso-SVR Model For Stock Prices Forecast

With the rapid extension development of economic integration, the stock market in the financial sector is developing with remarkable speed. However, due to stock samples structure is incomplete and high redundancy and linear correlation between variables. Which impact directly on the forecast of opening price in recent stock market, so the reference providing to investors are not reliable whether to buy stock or sell stock, thus lead to the result that the opening price of stock market ups and downs. Therefore, the study of the stock market opening price prediction method has wide application prospects.Taking into account the traditional methods of generalization ability and nonlinear mapping is low and other shortcomings. Support vector regression (SVR) not only has strong global optimization, but also can translate nonlinear problem into a linear problem through kernel function. What’s more, considering that the particle swarm optimization (PSO) has global search capability, rapid convergence, and optimize SVR not only select the optimal parameters and kernel function by PSO, but also improve the generalization ability of the model. Accordingly, the stock forecast of in single-index and multi-index, this subject first proposed asymmetric parabolic (AP) function fuzzy information granulation (FIG) stock samples, and through principal component analysis (PCA), AP-FIG model, grid search (GS) optimization SVR (GS-SVR) and dynamic weight factor updating PSO particle velocity (DIPSO) optimization of SVR, and first introduced in the dynamic reduction factor update DIPSO particle position (DKIPSO) optimized SVR. Take the relevant combination forecasting trends and stock price fluctuation range. View of the above research situation and technical problems, complete the following five tasks:1. For three kinds of kernel functions of SVR, forecasting the stock market opening through GS-SVR in Shanghai and Shenzhen, and IBM index. Simulation results show that perceptron kernel function (Sigmoid) have the best effect in fitting.2. Taking into account the forecasting methods have been only choice for domestic or foreign data, and opening price or closing price as input variables lacks contrast and authenticity. Therefore, this article selects opening price and closing price by T-FIG (triangular, T-FIG) and GS-SVR to forecasting opening price, and selects the optimal single index. The results show that the closing price is stronger than opening price.3. For multi-index samples, by the PC A dimension reduction, and then through the AP-FIG and DKIPSO-SVR prediction recent opening. By its principal component load and sort derived:single indicator, the closing price are the best choices. Multiple indicators, the Shanghai and Shenzhen index down to 2-dimensional, IBM index falling to 1-dimensional (closing price).4. Considering that in the triangle, trapezoid function graining effect is not smooth, and opening price history is not symmetrical, this paper proposed AP function, and through AP-FIG and DKIPSO-SVR prediction opening price. At the same time compare the prediction results with triangle, trapezoid function. The results showed that the minimum interval AP changes, the highest accuracy, followed by a ladder.5. For setting the weighting factor update PSO particle speed (IPSO) optimization SVR inefficiencies, reduced value between feed updates IPSO particle velocity (RIPSO) each factor mutually redundant. This paper first proposed DKIPSO model, and through AP-FIG, DIPSO-SVR and DKIPSO-SVR combination forecast recent opening price. Results showed that DKIPSO-SVR change interval and generalization ability is superior to DIPSO-SVR.