Ensemble option forecasting

Abstract

This thesis proposes a new model to forecast option prices with an ensemble of cutting-edge deep learning algorithms. We predict the entire volatility surface of 300 Technology stocks and use Rough Volatility (Gatheral, 2017) as benchmark. We demonstrate that our ensemble of deep learning techniques has also better stock log returns predictability power with over 400 factors reduced to 14 latent via auto-regressive auto-encoder. We predict option prices via an ensemble of linear regression, Rocket, Auto-ML and Temporal Fusion Transformer. We rank forecasts of option price returns in the universe and select options that satisfies an initial amount of capital. We show our strategy has satisfying performance metrics.

Date of Award	2021
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology
Supervisor	Kani CHEN (Supervisor)