Nonparametric Test of Affine Option Models

Oleg Bondarenko* (University of Illinois at Chicago)


Affine option models are popular among both academics and practitioners, because these models are flexible, analytically tractable, and easy to estimate. In this paper, we develop a novel methodology which can determine whether or not option prices are consistent with the class of affine models and, if yes, how many state variable are needed. The methodology is model-free and computationally very simple. It allows one to extract latent state variables without need to estimate model parameters and independent of any particular specification.

In the empirical application, the new methodology is implemented on the S&P 500 index options. We find that the whole class of 2-factor affine models is too restrictive for explaining historical option prices. This result rejects many canonical models, including the state-of-the-art model of Duffie, Pan, and Singleton (2000) with stochastic volatility and jumps in both price and volatility, as well as various generalizations. We argue that it might be warranted to extend affine models to three and more factors.