Backtests
Select a system to see its complete performance record against its own benchmark. Every figure is hypothetical and net of costs. The rules that produced them are not published — see why.
| Measure | COS | Nasdaq-100 | Difference |
|---|---|---|---|
| Annualised return (CAGR) | 22.78% | 10.85% | +11.93 pts |
| Annualised volatility | 22.2% | 26.9% | −4.7 pts |
| Sharpe ratio | 0.94 | 0.44 | +0.50 |
| Sortino ratio | 1.35 | 0.65 | +0.71 |
| Maximum drawdown | −35.3% | −83.0% | 47.7 pts shallower |
| Return / max drawdown (MAR) | 0.65 | 0.13 | 4.9× |
| Time below a prior high | 89% | 93% | — |
| Growth of $10,000 | $2,747,418 | $167,719 | 16.4× the index |
Log scale: equal vertical distance is equal percentage change. Both series start at $10,000 on 1999-03-11.
The index lost 83.0% in the dot-com unwind and did not reclaim that high for more than twelve years. The worst comparable loss for COS was 35.3%. Return is what gets advertised; this is the chart that decides whether an investor is still holding the position when the return arrives.
Every parameter in COS was chosen using data through 2015 and then frozen. The period below was held back entirely — it was never used to select, tune or filter anything. It is the closest thing a backtest has to an honest forward test, and it contains a bear market, a crash and a melt-up.
| Measure · 10.6 years out of sample | COS | Nasdaq-100 |
|---|---|---|
| Annualised return | 27.21% | 20.13% |
| Sharpe ratio | 0.99 | 0.83 |
| Maximum drawdown | −31.1% | −35.1% |
| Growth of $10,000 | $127,493 | $69,548 |
The two most recent complete months, measured against the index on the same dates and net of the same costs.
01
The edge in a systematic strategy is the ruleset. Publishing it transfers that edge to anyone who reads the page and, if enough capital follows the same signal, erodes it for the clients who paid for it. So this page shows the outcome in full detail and the mechanism not at all. Verify us on results, cost assumptions and out-of-sample discipline — the things that can be checked without giving the strategy away.
02
Daily data across 6,896 sessions from 1999-03-11 to 2026-08-10. Commission, spread and financing are deducted from every position, and returns are reported net. The benchmark is the Nasdaq-100 total return with dividends reinvested, measured on identical dates. Sharpe and Sortino are computed against the prevailing short-term Treasury bill rate as the risk-free reference.
03
Beyond the held-out decade, the research ran parameter sensitivity sweeps, bootstrap resampling of the return series, decay tests across sub-periods, and a cost sensitivity analysis that identifies the execution price at which the edge stops existing. Results that only worked at one parameter setting were discarded rather than reported.
04
These are hypothetical results computed with hindsight over a period in which US large-cap technology was the best-performing asset class in the world. No backtest can reproduce a live fill, an outage, a halt, or a regime it has never encountered. Treat every figure here as the ceiling of what to expect, not the estimate.
Access
If the drawdowns above are ones you could hold through, the next step is a short conversation about your account and whether this belongs in it.