📊 Portfolio analytics
Combo deployed strategy vs SPY, QQQ, 60/40. Drawdown curves, rolling Sharpe, alpha & info ratio.
Plus strategy-level decay monitor (Citadel/Millennium pod-style kill triggers) and PBO/CPCV honesty check.
Combo (deployed)
Sharpe
| Sortino | |
| CAGR | |
| MaxDD | |
| Calmar | |
| Vol | |
Combo vs SPY
Annual alpha (CAPM)
| Beta to SPY | |
| Info ratio | |
| Tracking error | |
Rolling 60d corr (combo vs SPY)
🚨 Strategy decay monitor (multi-PM platform style)
Citadel/Millennium-inspired kill triggers: 60d return < -5% → cut capital; < -7.5% → terminate.
Watch for negative 60d Sharpe even if not at kill threshold.
| Strategy | Status | Health | 60d Sharpe | 60d Ret | Curr DD | Days since high |
|
|
|
|
|
|
|
🧬 Factor risk attribution (BARRA-lite)
Combo regressed on factor proxies (Market, Size, Growth, Term, Credit) built from ETFs.
Beta = factor exposure. Alpha = idiosyncratic = pure alpha. R² = how much of the combo is "just factors".
| Strategy | Factor / Stat | Beta | Annual contrib | R² | Alpha t-stat |
|---|
|
|
|
|
|
|
📉 Tail-risk metrics — VaR / CVaR / Calmar
Standard hedge fund tail-risk metrics. CVaR (Expected Shortfall) = average loss when you exceed VaR. Calmar = CAGR / |MaxDD|. Sorted by Calmar.
| Strategy | Sharpe | VaR 95% | CVaR 95% | Worst Day | Worst 5d | MaxDD | Calmar |
|---|
|
|
|
|
|
|
|
|
🎯 Entry timing + stop-loss test (leader strategy, 7y)
Three findings: open-entry slightly beats close-entry (+0.08 Sharpe);
weekly rebal hurts (Sharpe 1.08 vs quarterly 1.65 — leader signal needs time);
per-position stops on quarterly rebal HELP — -20% stop lifts Sharpe 1.65→1.82 and cuts MaxDD -46%→-33%.
| Test | Entry | Freq | Stop | Sharpe | Sortino | CAGR | MaxDD |
|---|
|
|
|
|
|
|
|
|
⚖️ Walk-forward MV-LedoitWolf weight evolution
How the optimal weights for each strategy evolved over time. Refit monthly using prior 252 days. Note: insider weight went to 0 by April matching the decay-monitor alert; cross-asset rose to 22%.
🏆 Stacked overlays (best risk-adjusted)
Combining halve/kill drawdown overlay + sortino-target sigma. Order matters!
Halve/kill THEN sortino15 wins: Calmar 6.26, CAGR 118%, MaxDD -19%.
| Variant | Sharpe | Sortino | CAGR | MaxDD | Calmar |
|---|
|
|
|
|
|
|
📐 Sortino-targeting overlay (combo)
Sortino-target scales by downside-vol instead of total vol — preserves upside spikes that vol-target clips.
Best variant (sortino-target 15%) lifts CAGR from 85% to 106% with Calmar 4.70.
| Overlay | Sharpe | Sortino | CAGR | MaxDD | Calmar |
|---|
|
|
|
|
|
|
🛑 Drawdown overlay test
Adding kill-switch overlays to the combo. Sorted by Calmar (CAGR / MaxDD). Best variant: Halve@-10%/kill@-20% — improves Sharpe AND lowers MaxDD vs baseline.
| Variant | Sharpe | Sortino | CAGR | MaxDD | Calmar |
|---|
|
|
|
|
|
|
🚨 Anomaly detection (last 30d)
Rolling 60d Z-score + Isolation Forest on per-strategy daily returns. Flags days where a strategy's return is statistically unusual.
| Date | Strategy | Type | Direction | Return | Z-score |
|---|
|
|
|
|
|
|
⚡ Stress test — historical crisis periods
How each strategy performed in past stress periods. Honest view of when each works/fails.
| Period | Strategy | Cum Ret | Sharpe | MaxDD | Days |
|---|
|
|
|
|
|
|
🎓 Honesty check: PBO + CPCV
Probability of Backtest Overfitting (López de Prado 2014). 12-block CPCV. Lower = more honest.
| PBO | |
| Verdict | |
| Median IS-best Sharpe | |
| Median IS-best OOS Sharpe | |
| Degradation | |
| P(IS-best OOS Sharpe > 0) | |
| CPCV combinations tested | |
Per-strategy IS vs OOS Sharpe
⚖️ Portfolio construction methods
Hierarchical Risk Parity vs Mean-Variance Ledoit-Wolf vs Risk-parity vs Equal weight. MV wins on Sharpe but concentrates; HRP gives more diversification.
| Method | Sharpe | Sortino | CAGR | MaxDD |
|---|
|
|
|
|
|
🔬 Marcenko-Pastur covariance denoising
Random-matrix-theory based eigenvalue clipping. Filters noise eigenvalues from sample covariance. For our 6-strategy combo (q=92, plenty of data) it changes little. For a 20-stock min-variance portfolio (q=13) it dramatically tames extreme weights.
| Test | Cov | Sharpe / Vol | Max wt | Min wt |
|---|
|
|
|
|
|