News‎ > ‎

Time-varying Stickiness in Professional Inflation Forecasts

posted Mar 8, 2015, 7:20 PM by Elmar Mertens   [ updated Sep 15, 2016, 5:13 PM ]

In a new paper, co-authored with Jim Nason, we estimate a version of the Stock-Watson (SW) unobserved components (UC) model of inflation jointly with the Mankiw-Reis sticky information (SI) law of motion.

Jim  and I innovate on these models by adding time-varying persistence to the inflation gap of the SW-UC model in the form of a time-varying parameter AR(1) or AR(2) and to the SI model we let the frequency of forecast updating be time-varying. These time-varying parameters (TVPs) are assumed to follow independent random walks. As is standard in the SW-UC model, the innovations to trend and gap inflation are afflicted with stochastic volatility (SV) that follow log random walks. 

Our estimates (filtered and smoothed) of SV in the inflation trend are here, they display a pattern known from Stock & Watson (2007, JMCB):  

The joint model is estimated on real time U.S. GNP/GDP inflation and the associated average inflation predictions of the Survey of Professional Forecasters (SPF) on a sample running from 1968Q4 to 2014Q2. We estimate the joint model using a particle filter algorithm. The estimates show the data prefer a joint model with a TVP-AR(1) inflation gap. 

Here are estimates of our time-varying AR(1) coefficients, again filtered and smoothed, indicating substantial variation in inflation gap persistence as in Cogley and Sargent (2005, RED):

The joint model with time-varying inflation gap persistence also produces less sticky average SPF inflation predictions than with a fixed coefficient AR(1) inflation gap. We also find the SV of trend inflation exhibits negative comovement with the time-varying frequency of SI forecast updating while the SV and time-varying persistence of gap inflation often show positive comovement. Thus, the average SPF respondent is most sensitive to the impact of permanent shocks on the conditional mean of inflation.

Finally, here are the estimates of our stickiness parameter, represented as a weight on sticky vs current information. A value near zero is thus consistent with forecasts close to rational expectations:

The paper is here:  pdf

Slides are here: pdf (This is an updated version of our talk at the NBER SI) 

A shorter set of slides is here: pdf