This is the Bridgewater "All Weather" insight: every asset class has a macro environment where it shines. A balanced portfolio holds assets that cover all four quadrants so something is always working.

You detect regimes from leading macro data. Key signals:

Rule-based approach (simpler, more interpretable):

• Define thresholds: PMI > 50 = growth, CPI YoY > 3% = inflation
• Combine into quadrant classification
• Pros: transparent, easy to override manually, no overfitting risk
• Cons: arbitrary thresholds, binary classification misses transitions

ML-based approach (HMM, K-Means, GMM):

• Let the data discover regimes from return patterns
• HMM adds temporal structure — regimes are "sticky" and transitions have probabilities
• Pros: discovers non-obvious regimes, handles gradual transitions, probabilistic
• Cons: regime labels aren't interpretable by default, sensitive to hyperparameters, can overfit in-sample

HMMs model the market as a system that switches between hidden states. You observe returns; the model infers which regime generated them.

Key concepts:

• States: The discrete regimes (e.g., 4 states for the growth/inflation quadrant)
• Transition matrix: Probability of moving from regime i to regime j — captures regime persistence
• Emission distribution: The return distribution in each state (mean, variance)
• Viterbi algorithm: Finds the most likely sequence of regimes given observed data

In practice, HMM regimes often map to:

• State 1: Low vol, positive returns (bull market / Goldilocks)
• State 2: High vol, negative returns (crisis / bear market)
• State 3: Moderate vol, low returns (transition / uncertainty)
• State 4: High vol, positive returns (recovery / reflation)

Once you identify the regime, shift portfolio weights. You're not going 100% into the "winner" — you're tilting:

w_tactical = w_strategic + Δw_regime

Example tilts from a 60/40 baseline:

Keep tilts modest (±5-15%). Regime detection isn't precise enough to justify massive bets.