Weighted Dynamic Forecasting System

Fit and forcasting are two most important appliance in regression analysis. However, up to now, many work in the field of statistics only is done to improve accuracy of fit, people can hardly find out any article especially for forecasting. In fact, forecasting is meaningful in regression analysis. In addition, even if some methods take it important to forecast, such as selecting factor, but it just puts emphasis on the factor's influence on forecasting. It often happens that there are many factors which have influence on forecasting. With the limitation of our learning level and objective case, the variables X1,X2,...,Xm being taken into account are only a part of all relevant variables. The other variables being ticked out may well be more important. So it is not plenty to think about the factor's influence on forecasting, the sample's information also play its role on forecasting. And moreover, each sample play a different part in forecasting. Of course, not only the information of the old samples X(1),X(2),...,X(n) is useful for forecasting, information of the new sample X(n+1) is more meaningful. It's not correct that many models are used directly for new forecasting because the models are always brought out only on the base of the old samples. The new sample is relevant to the old samples, we must take it into account.On the base of the above knowledge, one important idea of this article is put forward: Firstly, cut down the accuracy of fit with the earlier samples, improve the accuracy of fit with the later samples, and you will improve the accuracy of forecasting. Further more, with the knowledge of the relations between new and old samples, the article put forward the iteration algorithm of the weight in ordered samples' weighted least-square estimation, the model if good at forecasting. At last, as to disordered samples, this article proposed an method to order the samples, and then forecasting is done by the algorithm of the ordered samples' weighted estimation. Above of all, a weighted dynamic forecasting system is brought out.