Working Paper: NBER ID: w11978
Authors: Joseph Engelberg; Charles F. Manski; Jared Williams
Abstract: We use data from the Survey of Professional Forecasters to compare point forecasts of GDP growth and inflation with the subjective probability distributions held by forecasters. We find that SPF forecasters summarize their underlying distributions in different ways and that their summaries tend to be favorable relative to the central tendency of the underlying distributions. We also find that those forecasters who report favorable point estimates in the current survey tend to do so in subsequent surveys. These findings, plus the inescapable fact that point forecasts reveal nothing about the uncertainty that forecasters feel, suggest that the SPF and similar surveys should not ask for point forecasts. It seems more reasonable to elicit probabilistic expectations and derive measures of central tendency and uncertainty, as we do here.
Keywords: No keywords provided
JEL Codes: C42; E27; E47
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| point forecasts provided by forecasters (C53) | systematic bias in the predictions (C51) |
| favorable point estimates in one survey (C83) | favorable point estimates in subsequent surveys (C83) |
| point predictions (C29) | reveal nothing about the uncertainty forecasters feel (D84) |
| point predictions above upper bounds of subjective means, medians, and modes (C46) | inconsistency with underlying subjective distributions (D80) |
| point predictions below lower bounds of subjective means, medians, and modes (C46) | inconsistency with underlying subjective distributions (D80) |
| aggregating point forecasts across forecasters (C53) | conflates variation in beliefs with variation in reporting practices (D80) |