Econometric Modeling as Information Aggregation
NBER Working Paper No. 2233 (Also Reprint No. r1468)
A forecast produced by an econometric model is a weighted aggregate of predetermined variables in the model. In many models the number of predetermined variables used is very large, often exceeding the number of observations. A method is proposed in this paper for testing an econometric model as an aggregator of the information in these predetermined variables relative to a specified subset of them. The test, called the "information aggregation" (IA) test, tests whether the model makes effective use of the information in the predetermined variables or whether a smaller information set carries as much information. The method can also be used to test one model against another. The method is used to test the Fair model as an information aggregator. The Fair model is also tested against two relatively non theoretical models: a VAR model and an "autoregressive components" (AC) model. The AC model, which is new in this paper, estimates an autoregressive equation for each component of real GNP, with real GNP being identically determined as the sum of the components. The results show that the AC model dominates the VAR model, although both models are dominated by the Fair model. The results also show that the Fair model seems to be a good information aggregator.
Published: "Comparing Information in Forecasts from Econometric Models." From The American Economic Review, Vol. 80, No. 3, pp. 375-389, (June 1990).