Multi-Scale Stress Accumulation as a Predictor of Elevated Volatility Across Financial Markets

We introduce a multi-scale stress accumulation framework for identifying regimes of elevated future volatility in financial time series. The model constructs a binary stress signal based on extreme return events and aggregates it using exponential weighting across fast, intermediate, and slow temporal scales. To ensure adaptability across changing market conditions, stress regimes are defined using rolling quantile-based thresholds. We evaluate the relationship between stress regimes and future realized volatility using Bitcoin (BTC-USD) as a primary case study, with additional validation across Ethereum (ETH-USD), U.S. equities (SPY), and foreign exchange (EUR/USD). Across all assets, periods classified as high stress are consistently associated with elevated short-term volatility. Out-of-sample testing demonstrates stable performance with minimal degradation between training and test periods. Across all tested asset classes, stress regimes are associated with a 20–60% increase in forward realized volatility. Unlike traditional volatility models that directly estimate variance dynamics, the proposed framework identifies a structural signal based on the accumulation of extreme events across time scales. The results suggest that multi-scale stress accumulation captures a generalizable property of financial markets related to volatility clustering and regime transitions. All computations and empirical results were obtained using a Python-based implementation of the proposed framework.