Abstract
The economist is often forced, by availability, to use data which is aggregated over time. In this study the effects of using temporally aggregated data in estimating and predicting economic relations is assessed. As an example the Michigan Quarterly Econometric Model (MQEM) of the U.S. economy is used. Assuming MQEM is the "true" specification of the Model the effects of aggregating the Model and the data to an annual model in estimation and prediction is evaluated. The bias implied by the aggregation process tends to be most severe when lagged endogenous as well as exogenous values are used, and does not pose much of a problem when no lag values exist in a given equation. In fact the magnitude of the estimated coefficients change very little in this latter case. The true problem is a specification bias that enters when temporally aggregated data is used when our equations involve lag values. Although the existence of an aggregation bias poses problems in finding the "true" structural parameters; the results of this study indicate that temporal aggregation causes only slight deterioration of predictive performance.
MacDonald, Steven Scot (1983). The effects of temporal aggregation of data in forecasting macro-economic relations. Texas A&M University. Texas A&M University. Libraries. Available electronically from
https : / /hdl .handle .net /1969 .1 /DISSERTATIONS -542437.