Asset Pricing Using Partially Misspecified Models

This paper studies a robust pricing method using a misspecified model. It is particularly suited if the available model is multi-dimensional (which is increasingly popular in economics and finance), and possibly has good quality along some dimensions of the state variable yet poor quality along other dimensions. Application to the pricing of Treasury options shows a pricing error reduction of over 30% relative to either parametric or nonparametric prices (both in sample and out of the sample), and suggests that the cheapest-to-deliver practice is an important source of misspecification