A1185
Title: High-frequency option predictability
Authors: Sebastian Egebjerg - Aarhus University (Denmark) [presenting]
Christine Bangsgaard - Aarhus University (Denmark)
Abstract: The purpose is to study the high-frequency predictability of S\&P 500 index option returns using trade and quote data combined with machine learning methods. The models achieve substantial out-of-sample predictive power, with median $R^2$ around 20\% and directional accuracy above 53\% across the sample period. Predictability is concentrated in out-of-the-money and less liquid options, is stronger toward the end of the trading day, and quickly disappears beyond a few minutes. The most important predictors include short-term past returns and implied volatility, alongside microstructure variables such as bid-ask spreads and order-flow imbalances. Incorporating information from other parts of the option surface does not meaningfully improve forecasts, and retail trading activity is associated with reduced predictability. Overall, the results document consistent short-term predictability in SPX options, driven by market microstructure rather than systematic mispricing, and unlikely to yield exploitable trading strategies.
