Working Paper: NBER ID: w12513
Authors: Michael W. Brandt; David A. Chapman
Abstract: We construct a simple reduced-form example of a conditional pricing model with modest intrinsic nonlinearity. The theoretical magnitude of the pricing errors (alphas) induced by the application of standard linear conditioning are derived as a direct consequence of an omitted variables bias. When the model is calibrated to either characteristics sorted or industry portfolios, we find that the alphas generated by approximation-induced specification error are economically large. A Monte Carlo analysis shows that finite-sample alphas are even larger. It also shows that the power to detect omitted nonlinear factors through tests based on estimated risk premiums can sometimes be quite low, even when the effect of misspecification on alphas is large.
Keywords: No keywords provided
JEL Codes: G0; G10; G12; G14
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| omitted variables bias (C20) | significant pricing errors (alphas) (G12) |
| measurement error in betas (C51) | low power to detect omitted nonlinear factors (C51) |
| model misspecification (C52) | biases in estimated alphas (C51) |
| finite samples (C34) | larger pricing errors (D49) |
| standard linear conditioning (C51) | significant pricing errors (alphas) (D43) |
| nonlinear factors (C39) | pricing errors (D49) |