Volatility Regime-Adaptive Spread Widening for ETF Option Market Making: Evidence from a 2024 SPY, QQQ, and IWM Panel

Abstract

This paper evaluates a volatility-regime classifier and spread-widening rule for ETF option market making using a 2024 panel built from Cboe historical options volume, Cboe symbol data, Cboe VIX history, and Stooq ETF prices. The empirical sample contains 756 symbol-day observations for SPY, QQQ, and IWM across 252 U.S. trading sessions. The study uses daily ETF returns, 21-day realized volatility, five-day VIX changes, and Cboe-style option volume features to classify each symbol-day as Calm, Normal, or Stressed. The experimental design uses a strict chronological split: January through September for training and October through December for testing. Five models are evaluated: a threshold-rule baseline, multinomial logistic regression, linear support vector machine, random forest, and gradient boosting. The best model is Multinomial logistic, which achieves a macro F1 score of 0.918 on the out-of-sample test period and materially outperforms the deterministic threshold baseline. The resulting spread rule widens quotes by regime and applies a VIX-shock add-on and an option-volume liquidity credit. In the execution proxy, the regime-adaptive rule lowers the objective to 4.957 bps, improves stressed-day coverage, and reduces under-widening relative to static, VIX-only, and realized-volatility-only controls. The paper also reports an LLM-style market-maker risk memo that converts model outputs into desk actions.