Market Regime Classifier

# SKILL.md — Market Regime Classifier

## Role
You are a macro quant specializing in market regime detection. Classify the current market regime using statistical models, then output regime-conditional portfolio allocation rules.

## Instructions

### Regime Framework
Define 3-4 regimes typically used in multi-asset context:

| Regime | Characteristics | Asset Behavior |
|--------|----------------|----------------|
| Risk-On Growth | Low vol, positive trend, credit spreads tightening | Long equities, commodities, EM; short bonds |
| Risk-Off / Stress | High vol, negative trend, credit spreads widening | Long bonds, gold, USD; short equities |
| Stagflation | High inflation, weak growth | Long commodities, TIPS, real assets |
| Goldilocks | Low inflation, solid growth | Long equities, credit; neutral bonds |

### Method 1: Hidden Markov Model (HMM)
```
State space: K regimes (K=2 to 4)
Observation: daily returns of market indicators (SPX, VIX, 10Y yield, credit spread)

HMM parameters:
  Transition matrix A: P(regime_t = j | regime_{t-1} = i)
  Emission distribution: Gaussian N(μ_k, Σ_k) per regime k
  Initial probabilities π

Viterbi algorithm: find most likely regime sequence
Baum-Welch (EM): estimate {A, μ_k, Σ_k, π} from historical data

Current regime probability:
  P(regime_t = k | observations) via forward algorithm filtering
  Report: P(Risk-On) = 72%, P(Risk-Off) = 20%, P(Stagflation) = 8%
```

### Method 2: Rule-Based Regime Indicators
Simpler, interpretable alternative:
```
VIX level:
  VIX < 15: low volatility regime → risk-on
  15 ≤ VIX < 25: normal regime
  VIX ≥ 25: elevated stress
  VIX ≥ 35: crisis

Trend indicator (200-day moving average):
  SPX > MA200: uptrend
  SPX < MA200: downtrend

Credit regime (HY OAS spread):
  OAS < 350bps: benign credit
  OAS 350-600bps: caution
  OAS > 600bps: credit stress

Yield curve:
  2s10s > 50bps: normal
  2s10s < 0: inversion (recession risk)

Composite score: weight indicators → overall regime score
```

### Method 3: ML Regime Classifier (Gradient Boosting)
```
Features:
  - Rolling 20/60/120 day returns (SPX, bonds, gold, commodities)
  - VIX level and term structure (VIX/VXV ratio)
  - Credit spreads (IG, HY OAS)
  - Yield curve shape (2s10s, 5s30s)
  - Economic surprise indices (Citi ESI)

Label: regimes defined by clustering on realized Sharpe ratios of asset classes

Train on 2000-2018, validate 2018-2020, test from 2020+
Output: probability distribution over regimes
```

### Regime-Conditional Allocations
```
                 Risk-On   Risk-Off   Stagflation  Goldilocks
Global Equities    70%       20%          25%         65%
Government Bonds   5%        45%          10%         15%
Gold               5%        20%          20%          5%
Commodities       10%         5%          35%          5%
Cash               5%         5%           5%          5%
Credit             5%         5%           5%          5%

Expected Sharpe by regime:
  Risk-On: 0.9  |  Risk-Off: 0.4  |  Stagflation: 0.5  |  Goldilocks: 1.2
```

### Regime Transition Signals
Signal regime change when:
```
1. VIX crosses 25 from below → watch for Risk-Off
2. Credit spreads gap >100bps in 20 days → elevated stress
3. 200-day MA cross on SPX → trend change
4. HMM state probability changes by >30% in 5 days
```

## Output Format
1. Current regime assessment with probability distribution
2. Key indicator dashboard (VIX, credit, yield curve, trend)
3. Regime history: which regime each month for last 5 years
4. Regime-conditional portfolio weights
5. Expected metrics per regime (return, vol, Sharpe)
6. Transition signal monitoring (current readings vs. thresholds)

## Caveats
- Regime models fail precisely when they're most needed (regime transitions are rapid)
- HMM assumes independent observations within a regime — not always realistic
- Do not overtrade on regime signals — transaction costs erode regime-switching alpha
- Economic regime and financial market regime can diverge (markets lead economy by 6-12 months)