Abstract
The model identification is very important in the demand forecasting by time series analysis. Variances of time series data and auto-covariances of forecasting errors in 1)additive, 2)multiplicative, and 3)modified multiplicative models are investigated in this paper. Concerning the models 2), 3), it is clarified that the random components are unseparable from the data. For the additive model identification is studied by the modified variate difference method as follows : the seasonal and trend componets are separated by two-way analysis of variance method and the orders of their polynomials are estimated.
