Frequency-division Combination Forecasting Of Stock Market Based On Wavelet Multi-resolution Analysis

The financial stock market is an extremely complex evolutionary system.The stock price trend is often affected by many factors at the macro and micro levels.Accurately revealing its internal laws is often very difficult,but its fluctuation not only indicates the economic development situation and the degree of market prosperity,but also It affects the immediate interests of investors.In the early years,the industry mainly tried to use a single method to grasp the stock price volatility,which is commonly used ARIMA,GARCH,SVM,neural networks,etc.,a single model can describe the fluctuation of stock prices to some extent,but the performance and prediction accuracy Because of the lack,so some scholars gradually focus on the combination of prediction models,hoping to explore more accurate prediction algorithms.This article also tries to make efforts in the direction of combination forecasting.The article first elaborates on the predictability of the securities market.The market can only have value if the quantitative model predictive analysis is used under non-efficiency conditions,and some research work has an effect on the effectiveness of China's securities market.The verification was conducted to conclude that the market did not reach the weak and effective consensus conclusion;then the difficulties in stock price forecasting were introduced,including: uncertainty,high noise interference,miscellaneous information,etc.;Analysis,ARIMA model,BP neural network combination solution,wavelet analysis Mallat algorithm can project stock price data to each scale space,and then achieve the decomposition of the signal components of each frequency,ARIMA model is the classic choice in timing analysis,and BP neural network mold Self-learning features are also widely used in various fields;In the use of the Mallat algorithm,two reconstruction schemes are designed by changing the wavelet coefficients,and accordingly,two distinct sequences with fluctuating characteristics are generated: the low-frequency trend sequence is smoother close to the basic development trend of stock prices.The high-frequency fluctuation sequence changes frequently and fluctuates greatly,and the nonlinear features are strong.Further innovations are made to construct a combined forecasting model(M-ARIMA-BP).The new algorithm can be used to return the data to two pairs no matter how many layers of the sample sequence are selected by the waveletdecomposition.The analysis of the sequence can not only take advantage of the wavelet “digital microscope”,improve the prediction accuracy of the model,but also reduce the subsequent modeling workload,making the combination of wavelet analysis and some complex model algorithms,such as deep neural network,more convenient.In the empirical part of the simulation,the daily closing price series of the Shanghai Composite Index and Shencheng Index are selected as the original data,and the combined model(M-ARIMA-BP)is compared with the standard wavelet prediction model and the prediction results of the corresponding single model.The results show that the prediction accuracy of M-ARIMA-BP algorithm is better than that of the comparison model and M-ARIMA-BP algorithm also has more robust performance in terms of accuracy of ups and downs.