Multi-Scale Regime Detection with Fuzzy Aggregation for Adaptive Sector Portfolio Management

Key results

Abstract

We propose a multi-scale regime detection framework that fuses four orthogonal stress detectors (CUSUM sequential analysis on z-score returns, rolling cross-sector correlation, market breadth, and rolling return skewness) into a unified stress probability via sigmoid-membership fuzzy inference. The aggregator uses a max-of-top-2 rule: it returns the second-largest membership, so at least two orthogonal detectors must agree before the system acts, structurally suppressing single-detector false alarms. The composite signal drives an adaptive sector portfolio that classifies eleven S&P 500 sector ETFs into three baskets by GARCH(1,1)-t conditional-volatility terciles and Ornstein-Uhlenbeck recovery half-life, with implicit cash allocation as the primary risk-reduction mechanism. In a walk-forward backtest (2009 to 2025, quarterly recalibration, 10 bps per trade, zero look-ahead), the framework is best understood as a capital-preservation overlay: it holds maximum drawdown to -6.2% (versus -41.7% for SPY) at 4.6% annualized volatility, for a Sharpe of 0.97 (versus SPY's 0.58) and Calmar 0.94. Benchmarked honestly, simple trend rules achieve higher raw Sharpe on the SPY universe; the architecture's distinctive value is generalization: applied unchanged to factor portfolios and international equities it scores Sharpe 1.21 and 1.04 against 0.51 and 0.11 for equal-weight.

Cite

@article{yadav2026regime,
  title   = {Multi-Scale Regime Detection with Fuzzy Aggregation for Adaptive Sector Portfolio Management},
  author  = {Yadav, Tanishk},
  year    = {2026},
  journal = {SSRN Electronic Journal},
  note    = {Preprint, abstract 6484679}
}