Among the pushbacks of using a trimmed-mean estimator of inflation is that you may be throwing the baby out with the bathwater. In more wonky language, you are trimming away information that is useful in spotting changes in the inflation trend. A recent example of this criticism was made in a May edition of the Richmond Fed’s Macro Minute by John O’Trakoun.

It’s a very common criticism. To the best of my recollection, this was first leveled at me in the early 1990s by the always thoughtful Charles Steindel, who claimed that a careful reading of the details inside the CPI report could give you insight into the future inflation trend. And if my memory serves me well (it was a long time ago), this was a primary motivation for the forecasting experiments done in Bryan and Cecchetti (1993) and elsewhere. To be clear, it was never our intent to argue that univariate or multivariate forecasts using the trimmed-mean inflation estimators is how inflation forecasting should be conducted (as hinted at by Bullard back in 2011.) It was our objective to show that whatever was being “trimmed” out of the estimator was not useful for predicting the future inflation trend. I think this observation still holds up well (as evidenced by a recent blog by Peter Rupert).

To those who say they can foresee a new emerging inflation trend by carefully reading the movements of the component price data, show it. Show that you can identify a transmission mechanism for inflation that runs through the component data. I’m betting you can’t. But if you can, that would be a huge contribution to central bankers.