Econometric models by design attempt to capture or represent simultaneously the data of interest and the ideas about how the modelled economy functions as reflected in that data. Economists have been building such models for several decades. Their primary task is one of identifying a set of parameters or constants that can be seen to characterize the economy being modelled. Such models are thought to represent a cause-and-effect relationship between two types of observable quantities. The effects are the so-called endogenous variables being explained and the causes are the so-called exogenous or autonomous variables whose values are determined by nature. Exogenous variables are determined by events beyond anyone's control or by artificial constructs that by design are completely within the control of governmental policy makers (tax rates, subsidies, etc.). It is the fixed parameters that determine the effects of any changes in the causes. The hope has always been that econometric model builders could succeed in developing a model that would simulate accurately the workings of the economy. Such a model, if the values of all its fixed parameters could be measured, would provide an excellent and reliable tool for forming predictions and forecasts of the future state of the economy or of the effects of changes in government policies.

Forecasting researchers have for the most part been disappointed with the performance of econometrics-based forecasting models. The reason for the unsatisfactory performance has always been considered a puzzle, and few critics have ever thought that the puzzle could not be solved. Nevertheless, as long as econometrics is the basis for building forecasting models, it is very unlikely that the puzzle will be solved.

The obstacle that stands in the way of solving this puzzle is to be found at the very foundation of econometric methodology. It is the fundamental view that sees the task of econometric theory as one of developing techniques to measure the values of fixed parameters. While physics may be based on the notion that nature provides fixed parameters such as the gravitational constant, it is questionable whether society can truly be seen to be governed by a set of nature-given fixed parameters. To think that fixed parameters can be the basis for an explanation of all of society ultimately leads to the view that all individuals' actions are pre-determined by the nature-given fixed parameters.

This obstacle leads forecasters to a forced choice. On the one hand, if one recognizes that parameters can be fixed (if at all) only for short periods of time then only short-term forecasts are warranted and only when based on recently collected data. On the other hand, if one builds only short-term forecasting models, then the forecasts will be plagued by the noise inherent in short-term data such as daily records of prices. The noise is the result of unexpected, unusual events that can temporarily distort prices and other data from their usual seasonal or trend-related values. Many forecasting researchers seem to be resigned to the dilemma of either rejecting the possibility of making model-based forecasts or accepting the necessary level of inaccuracy that is inherent in econometrics-based forecasting.

Lawrence A. Boland