Identifying Stock Option Mispricing via Machine Learning

This paper identifies implied volatility (IV) mispricing across a large cross section of options using the
Instrumented Principal Component Analysis (IPCA) framework. We incorporate short- and long-term
historical volatility components together with 15 stock- and option-related risk factors as characteristics
within the IPCA model. Different from targeting on traditional 1-month ATM IV misvaluation, our
analysis covers 16 option categories across four maturities (1,2,3,and 6 months) and four deltas (-25%,
-50% for put, 50%, 25% for call). Comparing model performance across factor dimensions, we determine
that the IPCA 4-factor model achieves the optimal trade-off between explanatory power and parsimony.
Applying a long–short 10-1 delta-hedged options trading strategy, the 4-factor IPCA model delivers a high
average information ratio (IR) of 3.399. Overall, our approach provides a powerful and implementable
predictive signal for option returns, remaining robust to transaction costs and consistently outperforming
conventional volatility-based benchmarks in double-sorting analyses.