Short-run money demand functions: Estimated speeds of adjustment and serial correlation

Parameter estimates in short-run money demand functions begin to behave badly when the sample includes data before and after say 1973, when the "missing money" becomes important. For example, the estimated speed of adjustment becomes implausibly slow. In such cases, there are often multiple minima in the error sum of squares taken as a function of the autocorrelation parameter [rho]. Low [rho]'s give implausible estimates, larger [rho]'s more plausible. The ESS should always be searched for such sensitivity. First differencing is not generally a good practice, but is sometimes useful.