# Is the effective federal funds rate stationary?

From a statistical perspective, some authors have found R^{data}_t (as used in the observation equation guide paper) to be non-stationary statistically.

Nevertheless, R^{data}_t is used in some reduced-form models (including Bernanke’s FAVAR model) with no transformation, implying R^{data}_t is stationary in levels. I guess the support for stationarity, in this case, would be the following statements from the “A guide to specifying observations equations…” paper:

“For nominal interest rates the evidence is not as clear-cut, but at least for most developed economies where the (Generalized) Taylor principle should have been satisfied, they should also be stationary”

Thus, stationarity could be supported by a statistical test, the Taylor principle (for developed economies), and perhaps by observation? By observation, I mean can we say a non-trending series is stationary and can be used in a model although it fails a statistical test - say ADF test?

1. There is some confusion here. The opposite of stationary is not integrated of order 1. Structural breaks also mean something is not stationary.
2. If something is integrated, a VAR can still be in levels. See Sims Stock Watson (1992) - Inference in linear time series models with some unit roots.
3. You need to be careful with unit root tests. They usually have low power. Failure to reject the null in an ADF-test does not mean the null of a unit root is actually true.
4. The correct approach often depends on what you are trying to achieve. For some applications, the distinction between an exact unit root and something very persistent is nothing to worry about.
1 Like