The purpose of this paper is to estimate an appropriate broad-money de
mand function for the United States and to examine its stability after
1987, when the Federal Reserve System began using M2 as a policy guid
e. Special attention is paid to the model specification, its dynamic s
tructure, and to its cointegration properties. The results from variou
s dynamic error-correction models suggest that: (i) the money demand r
elationship is stable; (ii) most previously estimated models have undo
ubtedly misspecified the interest rate variable; (iii) interest-rate v
ariability is another important determinant of real money demand; and
(iv) in contrast to previous studies, the long-run scale variable is r
eal GDP, whereas real consumer spending is the short-run scale variabl
e. The sample period examined is 1953:1 to 1991:4.