This paper develops a novel approach that leverages the information contained in expectations datasets to derive empirical measures of beliefs regarding economic shocks and their dynamic effects. Utilizing a panel of expectation revisions for a single variable across multiple horizons, we implement a time-varying factor model to nonparametrically estimate the latent shocks and their associated impulse responses at every point in time. The method is designed to accommodate small sample sizes and relies on weak assumptions, requiring no explicit modeling of expectations or assumptions about agents’ forecasting models, information sets, or rationality. Our empirical application to consensus inflation expectations identifies a single perceived shock that closely aligns with observed inflation surprises. The time-varying impulse responses indicate a significant decline in the perceived persistence of this shock, suggesting that inflation expectations have become more “anchored” over time.