News‎ > ‎

REVISED: Time-varying Forecast Stickiness w/Jim Nason

posted May 26, 2019, 7:07 AM by Elmar Mertens   [ updated Jun 4, 2019, 11:00 AM ]
Finally, together with Jim Nason, we just finished a thorough revision of our paper on time-varying stickiness --- i.e. a time/varying frequency of updating information -- in professional forecasts. The paper confirms earlier findings (e.g. Coibion and Gorodnichenko, 2015 AER) that SPF predictions are "sticky" in that their forecast errors are predictable, but we also find evidence that stickiness was actually quite low during the Great Inflation of the 1970s and the ensuing Volcker disinflation. Stickiness rose only once inflation became much less persistent, so that updating to more recent information adds much less predictive content.

The paper is here. See below for an abstract.

Here is a picture of the estimated stickiness parameter, as well as confidence bands around changes in the stickiness parameter.

Finally, here we show MSE loss of actual SPF forecasts compared against a counterfactual, where their stickiness is assumed to have been about as high as during the second half of our sample.


Our paper studies the joint dynamics of U.S. inflation and a term structure of average inflation predictions taken from the Survey of Professional Forecasters (SPF).  We combine an unobserved components (UC) model of inflation and a sticky information forecast mechanism to study these dynamics.  The UC model decomposes inflation into a trend and a gap component and measurement error. We innovate by endowing inflation gap persistence and the frequency of sticky information inflation forecast updating with drift. Stochastic volatility is imposed on the innovations to trend and gap inflation. The result is a non-linear state space model.  The model is estimated on a sample from 1968:Q4 to 2018:Q3 using sequential Monte Carlo methods that include a particle learning filter and a Rao-Blackwellized particle smoother. Our estimates show that (i) longer horizon average SPF inflation predictions inform estimates of trend inflation; (ii) inflation gap  persistence is procyclical before the Volcker disinflation and acyclical afterwards; (iii) by 1990 sticky information inflation forecast updating is less frequent than it was earlier in the sample; and (iv) the drop in the frequency of the sticky information forecast updating occurs at the same time persistent shocks become less important for explaining fluctuations in inflation. All told, the data calls for drift in inflation gap persistence and in the frequency of updating sticky information forecasts.