Stochastic Neural Network Prediction Model And Financial Time Series Volatility Research

Forecasting the fluctuations of financial energy time series has become a focus of e-conomic and social research.In this paper,in an attempt to improve the prediction accu-racy of energy prices,a novel neural network is developed through combining wavelet neural network and random time effective function.The wavelet neural network is a predictive system with the ability to implement strong nonlinear approximation.The random time effective function is applied to formulate the varied impact of historical data on current market,which endows historical data with time-variant weights to make them affect differently on the training process of the model,and this makes the model more in line with the volatility of the real market.In addition,we introduce the recur-rent layer into the random wavelet neural network to improve the history data memory of the model,and combine it with the wavelet decomposition to construct the hybrid wavelet neural network prediction model.The function of wavelet decomposition is to decompose the original unstable time series into subseries with different frequencies and they are more predictable.Combining the recurrent layer,random time effective function and wavelet decomposition will further improve the accuracy of the model.In the empirical study,we use two crude oil price series and two oil financial in-dexes,and study the prediction effects of price series and yield series.We make a comparison with traditional neural network model,SVM model,deep learning LSTM model,etc.In addition to using some conventional error analysis methods,this paper also introduces a multi-scale composite complex synchronization method,which pro-vides a new idea for error analysis.The MCCS analysis combines the sample entropy and the complexity invariant distance to measure the synchronism between two time series.By analyzing the synchronization between the prediction result and the real se-ries,we can analyse the prediction results.The error analysis results show that the two neural network models we proposed have high accuracy in global energy price series prediction.