Working Paper: NBER ID: w26105
Authors: Ivo Welch
Abstract: This paper proposes a robust one-pass estimator that is easy to code: Justified by the market-model itself and using a prior that market-betas should not be less than –2 and more than +4, the market-model is run on daily stock rates of return that have first been winsorized at –2 and +4 times the contemporaneous market rate of return. The resulting “slope-winsorized” estimates outperform (all) other known estimators in predicting the future OLS market-beta (on R² metrics). Adding reasonable age decay, suggesting a half-life of about 3 to 5 months, to observations entering the market-model further improves it. The estimates outpredict the Vasicek estimates by about half as much as the Vasicek estimates outpredict the OLS estimates.
Keywords: No keywords provided
JEL Codes: G10; G11; G31
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| Winsorization of returns (C46) | Estimation of market betas (C13) |
| Age decay (D15) | Accuracy of beta estimates (C51) |
| Slope-winsorized estimates (C51) | Predictive performance of future OLS market beta (G17) |
| Slope-winsorized estimators (C51) | Outperforming Vasicek estimates (G17) |