Working Paper: NBER ID: w28010
Authors: Michael Boutros; Itzhak Bendavid; John R. Graham; Campbell R. Harvey; John W. Payne
Abstract: Using 14,800 forecasts of one-year S&P 500 returns made by Chief Financial Officers over a 12-year period, we track the individual executives who provide multiple forecasts to study how their beliefs evolve dynamically. While CFOs’ return forecasts are systematically unbiased, their confidence intervals are far too narrow, implying significant miscalibration. We find that when return realizations fall outside of ex-ante confidence intervals, CFOs’ subsequent confidence intervals widen considerably. These results are consistent with a model of Bayesian learning which suggests that the evolution of beliefs should be impacted by return realizations. However, the magnitude of the updating is dampened by the strong conviction in beliefs inherent in the initial miscalibration and, as a result, miscalibration persists.
Keywords: Miscalibration; CFOs; Bayesian Learning; Confidence Intervals
JEL Codes: D03; D83; D84; E03; G30; G41
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| CFOs' beliefs about future stock market returns evolve based on observed return realizations (G41) | CFOs widen their confidence intervals (C46) |
| Realized returns fall outside a CFO's confidence interval (G17) | CFOs widen their confidence intervals significantly (C46) |
| CFOs miss their confidence intervals (C46) | CFOs learn less with each subsequent miss (G41) |
| Initial miscalibration predicts persistent miscalibration (C53) | CFOs widen their intervals less in response to missing the interval (G40) |