Perceived Taylor rule responses show limited time variation
Federal Reserve researchers Edward Herbst and Karen Page find that perceived monetary policy rules show little time variation once pooling restrictions are selected by data. The analysis evaluates 343 pooling patterns across the Blue Chip, SPF, and SEP forecast panels.
Intercepts absorb systematic forecast variation
Using Bayesian marginal likelihoods under a hierarchical hyper-g/n prior, the authors evaluate 343 restriction patterns across three forecast datasets: Blue Chip Financial Forecasts (119,435 observations), the Survey of Professional Forecasters (7,409 observations), and the Summary of Economic Projections (1,782 observations).
In all three panels, the data reject both a constant Taylor rule and the widely used period-by-period benchmark with respondent-date fixed effects.
The posterior distribution instead selects intercepts that vary across forecast horizons and survey dates.
Across all surveys, inflation and activity projections explain very little variation in projected policy rates.
Persistence replaces parameter volatility
Earlier literature attributed time variation in policy rules to shifting market perceptions by assuming respondent-date fixed effects.
When the model space expands to 2,401 specifications with inertial rules and partial pooling, systematic variation shifts into persistence parameters or grouped regimes.
In the Blue Chip and SPF panels, partial pooling compresses activity responses into five persistent date regimes and grouped horizons rather than high-frequency monthly fluctuations.
A statistical artifact unmasked
The paper demonstrates how ad hoc pooling created the illusion of volatile monetary policy reaction functions.
Data-driven model selection reveals that perceived response coefficients remain stable over time.
Researchers should discard period-by-period specifications that confuse horizon effects with shifting policy regimes.