676767(placeholder)
Introduction
Over the past few days, I have been retraining my quant system after adding a new elimination policy. The idea is fairly simple: reduce the influence of weaker opportunities while keeping the stronger parts of the system active.
On paper, that sounds like a small adjustment. In practice, changing what the system excludes can affect the behaviour of the whole portfolio.
The system is not built around one fixed portfolio. Each combination pairs different defensive and offensive strategies and uses different asset allocations depending on the market regime, which is estimated using a statistical model. That means these are genuinely different setups, not the same strategy with a few numbers moved around.
The policy keeps track of the recent health of each combination and uses an eligibility gate before accepting new exposure. If a combination starts to weaken, the system can hold it back while continuing to observe it.
Results
Getting to the final results took several days of retraining and evaluation because the backtest covers eight years. I was less interested in finding one unusually high score and more interested in seeing whether the improvement held up across several combinations.
I focused on validation performance and compared only the top five results, since the old leaderboard contained five entries. The candidates and their rankings changed, so the averages below compare the two groups rather than matching each rank directly.
| Validation metric | Previous top-five average | New top-five average | Change |
|---|---|---|---|
| Sharpe ratio | 1.930 | 2.299 | +0.369 (+19.1%) |
| CAGR | 22.31% | 24.33% | +2.02 pp |
| Maximum drawdown | -8.84% | -7.50% | 1.34 pp shallower |
| Calmar ratio | 2.567 | 3.242 | +0.675 (+26.3%) |
That is a fairly clear improvement across all four metrics. The new top five were:
| Rank | Validation Sharpe | Validation CAGR | Validation max drawdown | Validation Calmar |
|---|---|---|---|---|
| 1 | 2.348 | 22.24% | -6.98% | 3.188 |
| 2 | 2.289 | 24.31% | -7.52% | 3.232 |
| 3 | 2.287 | 25.19% | -7.67% | 3.284 |
| 4 | 2.287 | 25.19% | -7.67% | 3.284 |
| 5 | 2.285 | 24.73% | -7.68% | 3.220 |
All five updated combinations improved their validation Sharpe relative to their previous versions, with gains ranging from +0.313 to +0.468. CAGR improved by 1.18 to 2.86 percentage points, while maximum drawdown improved by 0.97 to 2.28 percentage points.
What stands out to me is not just the new top score. Performance and risk improved together across the group: average validation CAGR rose by just over two percentage points, average drawdown became 1.34 percentage points shallower, and average Calmar increased by 26.3%.
Of course, these are backtested validation results rather than live returns, so they do not guarantee what happens next. Still, seeing the improvement across several different strategy and asset-allocation combinations is a good sign that the new policy is doing more than helping one lucky setup.