Detailed reports / ML policy optimization
ML Policy Performance Report
Portfolio-scaled diagnostics for the frozen out-of-sample policy run. These views use weighted daily returns after portfolio sizing, ticker caps, strategy-family caps, and gross exposure limits, so headline Sharpe and drawdown reflect the allocated portfolio path rather than full-size raw trade compounding.
Portfolio path
Equity curve
Risk path
Portfolio drawdown
Calendar effects
Monthly return heatmap
Risk-adjusted trend
Rolling 12m Sharpe
Portfolio usage
Gross and net exposure
Trading intensity
Turnover
Sleeve contribution
Per-sleeve equity curves
Diversification
Sleeve correlation matrix
Edge map
Asset / timeframe / strategy diagnostics
| Scope | Trades | Net Profit | Scaled Return | Portfolio-Scaled DD | Daily Sharpe | PF |
|---|---|---|---|---|---|---|
| Loading report | ||||||
How Portfolio-Scaled Drawdown Works
Raw trade return is not the same as portfolio return. Each selected action receives a weight, then its contribution is scaled by that weight: weighted return equals trade return times portfolio weight.
The policy starts from predicted utility divided by squared ATR risk, clips and normalizes daily weights, then applies ticker, strategy-family, and gross exposure caps. Portfolio-Scaled DD is calculated from the daily sum of those weighted returns, compounded into an equity curve, and measured versus its running high.
Selection diagnostic
Randomized Entry & Path Risk
This section checks whether selected policy trades performed better than randomized entries from the same candidate universe, and whether the curve is robust to unfavorable trade sequencing.
Panel 1
Actual Selected-Trade Performance
These metrics summarize the realized trades selected by the policy for the chosen group.
Panel 2
Actual vs Randomized Entry
Randomized entry compares selected policy trades with random trades sampled from the same candidate universe.
Panel 3
Randomized Drawdown Risk
This shows whether the selected policy had better or worse drawdown behavior than random entries from the same opportunity set.
Panel 4
Markov Path Stress
Markov path stress keeps empirical win/loss behavior but resamples trade order using a simple two-state transition model.
This is a diagnostic curve using a fixed allocation fraction per action, not the final portfolio simulator. It is designed to test selection quality and path risk, not exact deployable PnL.