The ARMA/GARCH Model Parameter Estimation Method Of Neural Network

With the progress of society,a large amount of data is produced every day.Therefore,how to find important information from a lot of data becomes very important.Time series is a series that the variables are arranged in a chronological order,and the observation sequences can be found anywhere in the daily life.Through the analysis of the observation sequence,we can find its law of development,and by extracting the information from it,we can predict the future tendency.What is said above is the time series analysis.The time series can be divided into stationary and non-stationary time sequences,and the suitable models can be selected for the fitting prediction according to different types of time series.For stationary time series,autoregressive moving average model is usually used for fitting,whilefor non-stationary time series,the autoregressive integrated moving average model,residual autoregressive model,autoregressive conditional heteroskedastic model or generalized autoregressive conditional heteroscedastic model are generally used in fitting prediction.Artificial neural network is a burgeoning binoic subject which imitates human brain neural network structure and function.Due to its functions of nonlinear mapping,recognition,associative memory,optimal computation,knowledge processing,etc.,it is widely used in various areas,such as information processing,pattern recognition,medical expert system,market price forecasting,risk assessment.Nowadays,the parameter estimation of time series analysis model generally uses the conventional parameter estimation methods,such as least square method,maximum likelihood estimation method,square estimation method.However,using the artificial neural network to estimate the parameters of time series analysis model is not common.In view of the advantages of artificial neural network,this paper establishes the BP neural network parameter estimation structure of the ARMA model and the GARCH model,and deduces the calculation procedure of parameter estimation.Through the analysis of examples,the corresponding time series model was established,and through the least square method and BP neural network method,the parameters of the model are respectively estimated.By comparing the BP neural network method with the least-square method,this paper draws conclusion that two methods have got similar accuracy of parameter estimation in time series model,but the BP neural network method needs a smaller amount of calculation and is more convenient than the least square method.