Zusammenfassung:
The purpose of the paper is to revisit the demand for money specifiections by using U.S. quarterly data over the sample period 1951:1 - 1983:4. Utilizing the so-called threshold models suggested by Tong and Lim (1980) we first demonstrate the unsatisfactory performance of standard linear partial adjustment type specifications. Then we turn to compare error correetion type models; the generalized error correetion type demand for money model seems to outperform other specifications, but suffer from heteroscedasticity of residuals. Finally, an attempt is made with some success to account for this heteroscedasticity by augmented variables - variance of nominal interest rate and inflation and covariance between nominal interestrate and inflation - which attempt to measure changes in uncertainty over time. The resulting specification passes all standard diagnostic checks and shows also otherwise reasonable properties.