Asset Pricing Using Partially Misspecified Models

This paper provides an estimator for the conditional expectation function using a partially misspecified model. The estimator automatically detects the dimensions along which the model quality is good (poor). The estimator is always consistent, and its rate of convergence improves toward the parametric rate as the model quality improves. These properties are confirmed by both simulation and empirical application. Application to the pricing of Treasury options suggests that the cheapest-to-deliver practice is an important source of misspecification.