Volatility regime probabilities and the horizon profile of tail risk
FINANCE RESEARCH LETTERS, cilt.108, 2026 (SSCI, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 108
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.frl.2026.110425
- Dergi Adı: FINANCE RESEARCH LETTERS
- Derginin Tarandığı İndeksler: Social Sciences Citation Index (SSCI), Scopus, ABI/INFORM
- Orta Doğu Teknik Üniversitesi Adresli: Evet
Özet
We introduce Volatility Regime Risk (VRR), a real-time, model-implied regime risk measure that combines the filtered probability of being in the low-volatility regime today with the learned probability of moving to the high-volatility regime tomorrow. VRR is constructed from a Markov-switching GJR-GARCH model whose transition probabilities are learned from lagged macro-financial predictors using gradient-boosted trees. Using S&P 500 data over 2000-2024, we show that VRR has a sharply horizon-dependent relation with left-tail events. Univariately, VRR is positively associated with short-horizon tail risk. Conditional on the VIX and a lagged volatility proxy, however, its coefficient is negligible at daily and weekly horizons but significantly negative at the monthly horizon. The monthly result is robust to additional VIX-dynamics controls and to excluding the COVID-19 episode. Event-based evidence supports this interpretation: in the out-of-sample window, no monthly left-tail events follow the top-decile orthogonal-VRR event starts. The machine-learned transition structure also reveals a strongly right-skewed distribution of high-volatility regime persistence, while TreeSHAP attribution shows that VRR is most strongly associated with option-market signals, with secondary contributions from credit-market and macro indicators. VRR is best interpreted not as a universal crash-warning index but as a state-dependent measure of broad financial stress recognition, whose tail-risk implications differ across horizons.