Abstract
In short-range forecasting of demand by time seires data should be treated by the additive-type forecasting model after some suitable transformation. The authors have proposed an estimation method of the power transformation parameter by likelihood function in the multiple regression model, where the regression errors are assumed to be normally distributed. However, the assumption of normality is often unrealistic in the practical situations. This paper deals with an estimation method of the power transormation parameter by AIC derived from the postulation that the errors follow non-normal distributions, e.g, exponential power distribution and contaminated normal one.
