It’s time to move from “what” to “why”

For example, it has long been recognized that single exponential forecasting (SES) is equivalent to an ARIMA(0,1,1) model (e.g., Harvey, 1989) The additional flexibility of ARIMA models may be thought to lead to more accurate empirical forecasts. However, Table 13 of Makridakis and Hibon shows that there is virtually no improvement in forecasting accuracy using ARIMA models (labeled B-J automatic). This is interesting, but has been widely known since at least the time of the first M-competition. It is comforting that the previous findings continue to hold, but our understanding of forecasting has not been advanced much by its confirmation. Instead, I would like to see more research effort in explaining such empirical phenomena.