Regime-Aware Allocation¶
One-line definition¶
Adjusting portfolio construction based on which market regime (e.g. high/low volatility, risk-on/risk-off) is currently in effect, rather than using one covariance/return estimate across all history.
The easy 90% and the hard 10%¶
Given a regime label for each historical period, "regime-conditional
covariance" is nothing new — it's just covariance() computed on the
subset of returns matching that regime, then optimizing (via
minimum_variance, risk_parity_weights, etc.) with that regime-specific
input. No new machinery needed there.
Why this isn't implemented in src/pm¶
The hard part — detecting which regime the market is currently in, or was in historically — needs real statistical machinery (Hidden Markov Models, Markov-switching models, or similar unsupervised regime detection) that this repo doesn't build. Without that, "regime-aware" degrades to manually labeling periods by hindsight, which isn't a usable PM tool.
Why PMs care¶
Correlations and volatilities are famously regime-dependent — the "diversification" a covariance matrix promises in calm markets often evaporates exactly when a portfolio needs it most, in a crisis regime where correlations spike toward 1.