Study On An Non-steady Ecnomic Series Forecast Model

First described in the preamble of the time series model development and related research results. on the economic and financial forecast sequence facing a number of issues, the neural network methods in the economic application and optimization was induced.Secondly, the time sequence of the study, neural networks as a forecasting tool, genetic algorithm optimization tool, three of the basic theory and principle of the integration .. And the use of time-series modeling theory, Box-Jenkins methodology proposed to establish a time-series forecasting model as the order neural network input nodes is greatly shorten the learning efficiency.Then, in view of the economic and financial non-stationary sequence (including the exchange rate, stock index) information in a number of randomness, There is a strong volatility and heteroscedasticity, and different periods of different factors, it is difficult to grasp and respond to the whole sequence of these features. Despite the traditional methodology of the establishment of the Box-Jenkins ARIMA model ARIMA model is difficult to completely response and explanation from the residual relevance and heteroscedasticity. Information on the incomplete extraction implied that the ARIMA model limited. And artificial neural networks for strong nonlinear dynamic systems with strong self-learning, self-organizing and adaptive capacity, can be effective in numerical approximation to the time series with quantitative description of the mutual relations and good flexibility. This paper will be a combination of both organic : the use of economic ARIMA model sequences of the linear prediction Extraction of uncertainty, with the remaining residual characteristics of the nonlinear ANN model used information extraction, Through the selection of the model parameters and testing repeatedly tried to establish a ARIMA-ANN hybrid model, the Japanese yen against the dollar exchange rate forecast, and with the exclusive use of ARIMA model and ANN model compared. At the same time has come to adapt to fluctuations and the nonlinear long memory financial sequence GARCH model for the exchange rate forecasts. An empirical analysis and prediction, modeling simple mixing, extracting information ability, better reflected the characteristics of exchange rate fluctuations, explained residual informationFinal consideration of genetic algorithms and neural network in the economic forecast of the current report also relatively small, in the stock index forecast of virtually no domestic. Focus on specific economic and financial issues involved in genetic algorithm neural network optimization many difficulties. In this paper, using the Matlab toolbox legacy algorithms and neural networks, the combination of Based on a genetic algorithm for neural network time series prediction model A right on the card value refers weeks Kline empirical analysis and ARIMA model and the general ANN model, Based on the results of the model predictions were verified by comparing the genetic algorithm for neural network forecasting ability is enhanced to a certain degree, have been better robustness and generalization of the model prediction.