A Brief Parable of Overdifferencing
This is a short note, showing how money demand estimation works very well in levels or long (4 year) differences, but not when you first-difference the data. It shows why we often want to run OLS with corrected standard errors rather than GLS or ML, and it cautions against the massive differencing, fixed effects and controls used in micro data. It's from a PhD class, but I thought the reminder worth a little standalone note.
