Quant research · strategy due diligence
Auditing a +557% breakout strategy
I was handed an intraday opening-range breakout with a spectacular backtest and asked to evaluate and improve it. I rebuilt it from the spec, found where the returns really came from, stress-tested it in a different market regime, and proposed a fix that holds up out of sample.
- 97% trade-direction match on rebuild
- ~30% of reported P&L was fill optimism
- +78% Sharpe from vol-scaled trigger
- −35% max drawdown
The claim
One trade per session on a single high-volatility stock: wait for the opening range, enter on the breakout, stop and target fixed in percent. The backtest showed +557% over 98 sessions at 3× leverage, with an annualised Sharpe of 7.55. Numbers like that are usually wrong somewhere. The job was to find out where.
1 — Rebuild it independently
I rebuilt the strategy from the written rules on independently sourced 1-minute data. My trades matched the original’s direction on 97% of sessions, so the logic was real and reproducible. The Sharpe of 7.55 was also easy to decompose: it is a per-trade Sharpe of 0.49 multiplied by √252. That’s correct arithmetic, but it assumes a trade every day at that quality.
2 — Price the execution honestly
The original backtest filled at the most favourable price inside the trigger bar and assumed zero slippage. Re-filling at prices you could actually get removed about 30% of the reported P&L. The edge survived, but smaller.
3 — Ask what it is actually long
High-volatility days paid about twice as much as calm ones. The strategy was a long-volatility bet in disguise, not a trend or bull-market artefact. So I ran the identical rules on the same stock in its calm prior year.
2025 was calm (~2.4% average daily range). The edge disappears below roughly a 4% daily range: 49% win rate, profit factor 0.83, $85k drawdown.
4 — Fix it, and test the fix out of sample
Two changes, both picked on a training window and judged walk-forward:
- Volatility-scaled trigger. The breakout threshold scales with recent volatility, so noise on quiet mornings stops triggering entries.
- Volatility gate. Trade only when the prior 10-day realised volatility is above about 5%. The threshold was robust anywhere from 3% to 5%.
| Change | Effect (out of sample) |
|---|---|
| Vol-scaled trigger | Sharpe +78%, max drawdown −35%, whipsaw stop-outs 6 → 1, profit −9% |
| + vol gate, 2025 + 2026 combined | profit roughly flat (~$363k), Sharpe 0.22 → 0.50 |
The gate gives up almost no profit. It simply stops trading the regime where the strategy has no edge.
What didn’t work
Inverse-volatility position sizing made things worse. It shrinks exposure exactly when the strategy earns its money, fighting the long-vol edge. I kept it in the report, because a negative result is part of understanding the strategy.
A backtest is a claim about the future. The audit’s job is to work out which part of the number is edge, which is execution, and which is regime.