There has been much recent discussion ahout the ultimate sources of macroeco-nomic variability. A number ofauthors attribute most of this variability to only B few sowces, sometimes only one. Although there may be only B few important sources, this is fsr from obvious, since economies seem compliested. The purpose of this paper is to provide qusntitstive estimates of various sowces of vsriability using B U.S. econometric model. Stochastic simulation is used to estimate how much the overall variances of real GNP and the GNP deflator are reduced when various shocks BIG suppressed in the model. I. Ii-4TR0000~10~ There has been much recent discussion about the ultimate sources of macroeconomic variability. Shiller [1987] surveys this work, where he points out that a number of authors attribute most of output or unemployment variability to only a few sources, sometimes only one. The sources vary from technology shocks for Kydland and Prescott [1982], to unanticipated changes in the money stock for Barre [1977], to “unusual structural shifts, ” such as changes in the demand for produced goods relative to services, for Lilien [1982], to oil price shocks for Hamilton [1983], to changes in desired consumption for Hall [1986]. (See Shiller [1987] for more references.) Although it may be that there are only a few important sowces of macroeconomic variability, this is far from obvious. Economies seem complicated, and it may be that there are many important sources. The purpose of this paper is to estimate the quantitative importance of various sources of variability using a macroeconometric model. Macroeconometric models provide an obvious vehicle for esti-mating the sources of variability of endogenous variables. There are two types of shocks that one needs to consider: shocks to the stochastic equations and shocks to the exogenous variables. Shocks to the stochastic equations are easy to handle. They ax simply draws from the postulated distribution (usually normal) of the structural error terms, the distribution upon which the estimation *This paper grew out of discussions with Robert Shiller, to whom I am indebted for many helpful suggestions and comments. Some of the results in this paper are

[Was the Depression forecastable? After the Crash, how long should it have taken contemporary forecasters to realize how severe the downturn was going to be? Data assembled by the Harvard and Yale forecasters--together with modern historical data--are subjected to statistical analysis. Neither contemporary forecasters nor modern times-series analysts could have forecast the large declines in output following the Crash.]

This paper extends the classical test for structural change in linear regression models (see Chow (1960)) to a wide variety of nonlinear models, estimated by a variety of different procedures. Wald, Lagrange multiplier-like, and likelihood ratio-like test statistics are introduced. The results allow for heterogeneity and temporal dependence of the observations. In the process of developing the above tests, the paper also provides a compact presentation of general unifying results for estimation and testing in nonlinear parametric econometric models.

Was the Depression forecastable? After the crash, how long should it have taken contempo rary forecasters to realize how severe the downturn was going to be? These questions are addressed by studying the predictions of the Harv ard Economic Service and Yale's Irving Fisher during 1929 and the ear ly 1930s. The data assembled by the Harvard and Yale forecasters, tog ether with modern historical data, are subjected to statistical analy sis to learn whether their verbal pronouncements were consistent with the data. Both the Harvard and Yale forecasters were systematically too optimistic. Yet, nothing in the data suggests that the optimism w as unwarranted. Copyright 1988 by American Economic Association.