This paper investigates the relative forecasting performance of some p
artially adaptive estimators of ARIMA models based on flexible paramet
ric error distributions which allow for kurtosis and skewness. These m
ethods are used to obtain forecasts of some well-known data sets.