Forecasting GNP Using Monthly M1

In this paper, we present an application of mu1tivariate time series forecasting in which the data consist of a mixture of quarterly and monthly series. In particular, we use monthly series of M1 to forecast quarterly values of the nominal gross national product (GNP). Results from estimating models over the period 1959:IQ through 1979:IVQ indicate that models involving only movements in monthly MI series provide approximately the same explanatory power as one using quarterly MI. When these models are used t o forecast GNP over the time period 1980:IQ through 1984:IIIQ, the results are mixed. For one-quarter-ahead change, four-quarter-ahead change, and one-year change forecasts, the Root Mean Square Error (RMSE) for all the models ( including a univariate model of GNP ) have approximately the same RMSE ( for a given forecast horizon ) for the entire period. However, when we examine the period 1983:IIIQ through 1984:IIIQ, the models using MI provide better forecasts than the univariate model, in terms of RMSE, for four- quarter and one-year change forecasts. Also, the models using monthly M1 data, perform at least approximately equal to the model using quarterly M1 data, and in some cases substantially better. All of the multivariated models used in this study indicate that the growth in GNP was smaller than expected relative to changes in MI over the entire period. GNP growth had a larger variance from 1980:IVQ to 1983:IIQ than was expected based on all models used in this study.

Comparisons of forecast errors among different studies is often difficult because of the different time periods involved and because of the different amount of data available when the forecasts are actually made. However, comparisons of the forecasts errors for these models to results from other studies using St. Louis type equations indicate that the models presented in this study appear to perform slightly better than the St. Louis models for one-quarter forecasts in terms of RMSE. Also, results for one-year change forecasts are apparently better than the median of five early- quarter forecasts by the ASAINBER survey, Chase, Data Resources, Inc. (DRI), Wharton, and BEA.