Forecasting with Monetary Aggregates: Recent Evidence for the United States

We investigate the out-of-sample forecasting performance of various monetary aggregates in four-variate models of real output growth, inflation, changes in an interest rate, and nominal money growth from 1992 to 2004 using vector autoregressive (VAR) and regime-switching (RS) VAR models. We consider both Divisia and simple sum monetary aggregates for M1, M2M, M2, and M3 as well as sweep-adjusted M1 measures. We find little evidence that either aggregation method or level of aggregation has a big impact on the forecasting performance of our model with respect to inflation and real output growth. Our results indicate that VAR models with monetary aggregates appear to produce at best marginal improvements in RMSE over VAR models that omit money growth altogether. We also find that RS-VAR models usually provide better one quarter ahead forecasts than comparable VAR models, but often did worse when forecasting inflation four quarters ahead.