A Tool for Uncovering Forecasters’ Beliefs

That’s what the researchers’ tool—time-varying heteroskedastic principal component analysis, or tvHPCA for short—aims to provide. Heteroskedastic deals with varying levels of noise in data across horizons, and principal component analysis is a statistical technique for finding a signal or trend. Together, they isolate economic trends even as data patterns shift over time.

The tool treats monthly forecast revisions as a signal. When a forecaster substantially adjusts their one-year inflation outlook but barely changes their five-year projection, the tool infers that they do not expect the shock to persist. By contrast, if the forecaster revises inflation predictions across all horizons in the same direction, it suggests they expect the shock to linger. This information can help central banks assess whether stronger communication is needed or whether interest rates should be raised sooner.

The standard statistical tools that economists use to measure forecasters’ beliefs often struggle with noise (random measurement errors in expectations) that varies across forecast horizons and time periods. But Giacomini, Lu, and Smetanina suggest tvHPCA overcomes that obstacle, making it especially valuable during periods of economic uncertainty. It can also outperform standard tools when expectations data include forecasts for only a few horizons at each date, especially when the amount of noise differs across the horizons.

The researchers tested their method on simulated datasets, where the true shocks and impulse responses were known in advance, in order to confirm that the estimation performed well.

After that, they turned it on four decades of actual inflation forecasts made from February 1982 to July 2023, combining short- and intermediate-term expectations from the Blue Chip Economic Indicators survey with longer-horizon estimates from the Federal Reserve Bank of Cleveland. The resulting panel of data covers seven forecast horizons, from current-quarter predictions to 10-year outlooks.

Across this 40-year window of data, the analysis identifies what appears to be a single dominant shock driving inflation-forecast revisions at every horizon examined. This shock represents whatever new information caused forecasters to collectively revise their inflation predictions between one month and the next—the thing that surprised them. Notably, this statistically extracted measure of forecaster surprise closely tracks surprises derived independently from actual data from the Consumer Price Index using a separate econometric model. The strong correlation between these two independently constructed measures suggests that the extracted shock captures genuine price movements in the economy rather than phantom disturbances.

The analysis suggests that forecasters have come to believe that inflation shocks fade faster than they once did. During the mid-1980s, long-term expectations shifted quite a bit when shocks occurred. But in recent decades, that sensitivity appears to have diminished, and forecasters now see inflation disturbances as passing storms rather than permanent weather changes.

In central-banking circles, this shift toward thinking about shocks as transitory is called anchoring. When inflation expectations are well anchored, economic forecasters trust that inflation will return to target and that policymakers don’t need to be aggressive in raising interest rates. The beliefs extracted by the researchers’ tvHPCA method point to progressively stronger anchoring over the time studied.

“By focusing on expectation revisions rather than realized data, the method provides something policymakers rarely have access to, namely a real-time gauge of public beliefs about the size of a shock and how long it will last,” says Smetanina. The researchers say that their tool will enable policymakers to communicate more effectively, respond faster, and better understand the perceptions that shape economic outcomes.