Comparative Performance of U.S. Econometric Models

Comparative Performance of U.S. Econometric Models

Comparative Performance of U.S. Econometric Models

Comparative Performance of U.S. Econometric Models

Synopsis

This volume compares strategic properties of the leading macroeconometric models of the United States. It summarizes the work of an ongoing seminar supported by the National Science Foundation and chaired by Lawrence R. Klein of the University of Pennsylvania. The Seminar meets three times annually. Comparisons are made across models for such characteristics as conventional multipliers (fiscal, monetary, and supply side shocks), J-curve response to dollar depreciation, and forecast performance under consistent assumptions. There are in-depth comparisons of some models and investigation of use of high frequency information to improve forecasts. There are also analyses of the sources of forecast error. The core structures of models, especially their ISLM cores, are compared. The volume contains one chapter on comparison across models of different developing countries. In addition to the contributions by participating model builders who meet regularly, the book contains critical appraisals by outsiders. The contributors include many distinguished economists in model use and analysis. Many are operators of the countries best known modelling facilities. The introduction was written by Lawrence R. Klein, winner of the Nobel Prize in Economic Science in 1980, for his work in construction and use of econometric models.

Excerpt

Roger E. Brinner and Albert A. Hirsch

Chapter 2 presented comparative multipliers for 10 models calculated over the decade 1975-1984 using apparently common sets of changes in exogenous inputs. As in previous exercises, the spans of outcomes among the models for such aggregates as real gnp and the gnp implicit price deflator are distressingly large, both in terms of the amplitudes of the multipliers and their time profiles. Although there is some tendency toward clustering, the persistence of such sharp differences is disturbing, particularly insofar as models share a common theoretical basis. Such an outcome tends to breed exaggerated cynicism among economists and noneconomists alike: if the methods of economic science cannot extract common empirical findings in an apparently controlled experiment, then how much science is really involved? the proper answer is that considerable science is employed; indeed, the differences are analogous to the ambiguity of empirical results from natural sciences where the opportunities for experimental controls are limited.

Two main categories of causes underlying the intermodel differences in outcomes can be identified. First, there is the familiar fact that single-equation specifications for the same endogenous variable differ across models, and the associated unpleasant truth that available statistical methodology all too often fails to offer strong tests for selecting among competing behavioral hypotheses. Even subtle variations in the specification of, say, a household automobile demand equation, not to mention variations in the estimation period, can yield substantial differences in response characteristics that may be magnified over long simulation spans. This problem is often exacerbated in economists' time-series regressions by multicollinearity, that is, high correlations among cyclical explanatory phenomena such as income, wealth, inflation, and credit conditions. Such difficulties are, in fact, common to most empirical sciences. For example, medical researchers are apt to be frustrated in cross-section regressions by high correlations among dietary choice, exercise, location, and age--the principal non- pharmaceutical explanatory factors in the analysis of the incidence of particular diseases. Second, some models treated important macroeconomic variables as exogenous for the multiplier experiments; others treated them as fully or partly endogenous. Examples include the dollar exchange rate, the macrovariables of the trading partner economies (production, prices, and interest rates), the rigidity of the Federal budget in real or nominal terms, and state-local government expenditures and tax rates. This source of differences did turn out to be quite important. There is once again a parallel . . .

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.