Abstract
Transfer scaling method utilises both a larger data from older time point and a smaller data from more recent time point. A use of only the smaller data from more recent time point, however, reportedly produced better forecasts than the transfer scaling. This study analyses usefulness of the transfer scaling focusing on data collection time points and the numbers of observations. In any combinations of time points and the numbers of observations, the transfer scaling never produced statistically significantly better forecasts than the model using data from only the more recent time point. The transfer scaling method has advantages in cases where the number of observations from the more recent time point is substantially small; it produced better forecasts on average with smaller variance, and it is less likely to produce poor estimates and forecasts.
