Research On Stock Forecasting Based On Wavelet Analysis And Neural Network

The movement of stock price is a complex nonlinear dynamic system,while the features of high-yield and high-risk of stocks have attracted a large number of domestic and foreign professionals and scholars to devote to researching the movement of stock price.Wavelet analysis is a time-frequency domain analysis method,which takes into account the information of the signal in the time domain and frequency domain,and the artificial neural network can be used to process the prediction of nonlinear systems.There are sufficient theories which combine wavelet analysis and neural network for stock price prediction at present,for example,the common model of wavelet neural network is a neural network model based on the BP neural network topology,using the wavelet basis as the activation function of the hidden layer nodes,and transmitting the signal forward while transferring the error backward.There exists difference between the stock forecasting model discussed in this study and the common model of wavelet neural network: in the modeling process,the original signal is first subjected to be wavelet decomposed to separate the low-frequency signal that can represent the main trend of the original signal and the high-frequency signal that can represent the high-frequency change.Then some low-frequency signals and high-frequency signals are selected as the inputs of BP neural network for prediction.In the common model of wavelet neural network,the wavelet bases play the role of activation functions,while in the prediction model discussed in this study,wavelet decomposition plays the role of sorter and selector of the original signal.From the simulation results in Sections 3.2,3.3 and 3.4 of this article,we can see that after wavelet decomposition of the original signal(stock price),several high-frequency signals and low-frequency signals are obtained,and the high-frequency signals and low-frequency signals are input into the prediction model.Later,under the same prediction accuracy and prediction stability requirements,the number of nodes and training sample capacity of the hidden layer of the BP neuralnetwork can be effectively reduced,and the training times of the BP neural network can be effectively reduced.That is,under the same prediction accuracy,the prediction model consumes less computing power and system overhead after wavelet decomposition of the original signal.In addition,after the original signal is decomposed by wavelet,the original time series data is converted into cross-section data,and the time series prediction problem can also be converted into a regression problem(see section 3.6 of this article).There are few related theories or applications of existing wavelet neural networks that discuss the difference between the prediction model for stock index and the prediction model for individual stock.But in this article,this issue is discussed,and the optimal parameters of the model for stock index and for individual stock predictions are calculated respectively(see sections 3.3 and 3.4 of this article).The simulation results show that after substituting the optimal parameters,the model's prediction error of the stock index's closing price in the most recent trading day can be controlled within 0.5%.The algorithm in this article is based on Python and MATLAB.The calculation results and the source program are attached.Readers can verify the results by running the attached program and viewing the calculation results in the attached table.