Monetary anticipations and the demand for money in the U.S. further results

Forecasting equations which attempt to capture the behaviour of rational agents should include any variables which cause the dependent variable in Granger's sense. Moreover regressions containing regressors generated from other estimations may require the covariance matrix to be estimated in a special way. Evidence is presented here on the shock absorber hypothesis for the US with the covariance matrix estimated consistently (a prerequisite for valid inference) using three separate anticipations generating equations - a univariate ARIMA process, a pure autoregression and a stationary "weekly rational' equation. Secondly, and more importantly, the rational expectations-shock absorber hypothesis using the methodology proposed originally by Mishkin is tested. That is, the forecasting equation and the shock absorber money demand equation is estimated jointly. Apart from obviating the need for a correction to the estimated covariance matrix, the Mishkin methodology has two further advantages: it allows more efficient parameter estimates; and it allows cross-equation rationality restrictions (implicitly imposed in the two-step method) to be tested jointly with the neutrality hypothesis (that is, only unanticipated money has an effect on real money demand). The joint hypothesis of rationality and neutrality is decisively rejected. -from Authors