| Popis: |
Experience using twenty-one actual economic series suggests that using the Box-Cox transform does not consistently produce superior forecasts. The procedure used was to consider transformations x (λ) = (x λ −1) λ , where λ is chosen by maximum likelihood, a linear ARIMA model fitted to x (λ) and forecasts produced, and finally forecasts constructed for the original series. A main problem found was that no value of λ appeared to produce normally distributed data and so the maximum likelihood procedure was inappropriate. |
