Deep Surrogates for Finance: With an Application to Option Pricing

Abstract

We introduce ``deep surrogates'', a novel approach to speed up structural models' evaluation and estimation by orders of magnitude, thereby allowing for various compute-intensive applications that were previously intractable. As an application, we build surrogates for popular option pricing models. The surrogates make daily re-estimation of the model's parameters computationally feasible. With these re-estimations, we compare the out-of-sample performance to non-parametric alternatives. More importantly, we can show that parameter instability for the structural models is positively correlated with options market illiquidity. Furthermore, the surrogates make it easy to extract model-implied conditional tail risk measures and predictive distributions of option returns.

Keywords: Deep Learning, Structural Estimation, Option Pricing, Parameter Stability

JEL Classification: C45, C52, C58, C61, G17