The framework of a general ARIMA model is agnostic and atheoretic in nature, thus neglecting
the inclusion of explanatory variables, like in econometric models, basing the forecasts only on past
values of the dependent variable y in combination with present and past error terms ε. The error part is
simply a different notation for a moving average representation. An interesting characteristic trait of
the ARIMA approach is that although it has no theoretical appeal to economics it frequently
outperforms its theory based econometric pendants in short-run forecasts (e.g. (Litterman 1986;
Stockton and Glassman 1987; Nadal-De Simone 2000)). For example, (Stockton and Glassman 1987)
conduct a comparative study on three different inflation processes (Rational expectations model,
monetarist model and the expectations-augmented Phillips curve) that are based on economic theory
and widely accepted among economists as an adequate representation of the underlying structural
relationships that explain and form inflation. These processes are compared to each other utilizing their
out-of-sample forecast performance on an eight-quarter horizon and in addition a simple ARIMA
model is used as a benchmark to substantiate the theoretical validity of the econometric models.
Surprisingly, their findings suggest that the ARIMA model outperforms both the rational expectations
model and the monetarist model and is found to perform just as good as the Phillips curve in all
specifications. Their concluding remarks depict tellingly what this means for econometricians: “…after
considerable theoretical effort has been expended to explain its causes [inflation], it seems somewhat
distressing that a simple ARIMA model of inflation should turn in such a respectable forecast
performance relative to the theoretically based specifications. Theory has yet to yield only small
dividends in terms of improving our ability to predict accurately the course of inflation” ((Stockton
and Glassman 1987), p. 117).