Using Machine Learning Methods to Predict Implied Volatility Surfaces for SPX Options

Abstract

The recent popularization of the derivatives market has led to increased interest in methods for modeling and predicting the implied volatility surface in different financial products. In particular, accurately predicting how the implied volatility surface changes with time can yield profitable trading strategies and may illustrate how different market influences may affect option pricing. The aim of this paper is to explore the abilities of different machine learning models to predict changes in the implied volatility surface over a one day to one month horizon in options on the S&P 500 index (SPX). First, we use historical option pricing data to extract an implied volatility surface and explore possible models which can be used to regularize the surface. Next, we evaluate the consistency, accuracy, and speed of different machine learning models in predicting surface moves. Then, we examine different market inputs as potential features for machine learning models and determine which ones have the highest predictive power. Finally, we explore potential applications of our models and potential further steps which can be taken to improve these models.