Deep Learning from Implied Volatility Surfaces

Abstract

We develop a novel methodology for extracting information from option implied volatility (IV) surfaces for the cross-section of stock returns, using image recognition techniques from machine learning (ML). The predictive information we identify shows minimal correlation with the existing option-implied characteristics, delivers a higher Sharpe ratio, and has a significant alpha relative to a battery of standard and option-implied factors. Large ML ensembles yield the best results, with performance improving as ensemble size increases. We introduce principal linear features, an analog of principal components for ML, and use them to show IV feature complexity: A low-rank rotation of the IV surface cannot explain the model performance. Our results are robust to short-sale constraints and transaction costs.

Keywords: Volatility surface, Convolutional neural networks (CNN), Machine learning, Cross-section of stock returns, Deep ensembles, Virtue of complexity