Working Paper: NBER ID: w7162
Authors: A. Craig Mackinlay; Lubos Pastor
Abstract: Implications of factor-based asset pricing models for estimation of expected returns and for portfolio selection are investigated. In the presence of model mispricing due to a missing risk factor, the mispricing and the residual covariance matrix are linked together. Imposing a strong form of this link leads to expected return estimates that are more precise and more stable over time than unrestricted estimates. Optimal portfolio weights that incorporate the link when no factors are observable are proportional to expected return estimates, effectively using an identity matrix as a covariance matrix. The resulting portfolios perform well both in simulations and in out-of-sample comparisons.
Keywords: Asset Pricing; Expected Returns; Portfolio Selection; Factor Models
JEL Codes: G12
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| mispricing vector (D46) | residual covariance matrix (C10) |
| omitted risk factor (I12) | mispricing of asset returns (G19) |
| strong link between mispricing and residual covariance matrix (C10) | more precise expected return estimates (C51) |
| identity covariance matrix (C10) | outperform sample or true covariance matrices (C10) |
| expected return estimates incorporating strong link (G17) | more stable over time (D15) |
| restricted estimators (C51) | optimal portfolio weights proportional to expected return estimates (G11) |