1979 Volume 30 Issue 1 Pages 48-53
An adaptive forecasting is proposed to improve the adaptability to the non-stationary variation of the time series data. The distribution of weighted sum of square error approximated by Gamma distribution gives the criterion to obtain the confidence intervals of the variance of the forecasting error. Thus the new testing method which utilizes the above confidence intervals to change the parameter values of the model is established in this paper. Both theoretical and numerical analysis by simulation data and sales data show the superiority of the new adaptive method to the prevailing ones.