| Presently the main technologies of macroeconomic data forecasting is some traditional methods such as time series model and multivariate regression model, but the actual result is not so good and satisfying when they are used to deal with the nonlinear and complicated system. On the other hand, the attributive variables' value has not only a connection with the forepassed time series data but also a closed connection with the simultaneous affecting factors which have complicated correlation themselves. All these have made forecast more hard-dealed.Artificial neutral network is a new kind of technology which appears these years. It is especially good at dealing with the nonlinear and complicated system. There have been many reports and researches about its using in forecasting and the satisfying results gained in many aspects.Taking all these into account, this paper presents an artificial neutral network-based forcasting model which takes both times series data and simultaneous affecting factors which will be processed by multivariate statistical analysis treatment into account. It not only takes the advantage of neutral network's good ability to deal with the nonlinear and complicated system, but also makes the forecasting more reasonably because it reflects both the cause and effect relation and time series' dynamically developing trend.Besides these, this paper also makes a further research on the structure of the network and put forward some methods to determine the number of the input nodes of the network.Using a set of real macroeconomic data, three different yearly GDP absolute value forcasting models have been set up according to the new method and the other two respectively. The comparative result indicates that this model is better than the other two models in many respects and proves the new, model's validity, feasibility and superiority to a certain extent. It has applied value in the field of macroeconomic data forecasting.
|
PDF Full Text Request