1Verdict
Most of the headline result comes from choosing the best of many tries. A smaller, real effect remains, and it partly holds on data the strategy never saw. We would present it to an allocator at the restated figure, not the reported one.
2The numbers
| Sharpe ratio, as reported | 1.84 |
| Sharpe ratio, after the trials behind it | 0.52 |
| Share of the result explained by selection | 71% |
| Sharpe ratio on held-out data | 0.61 |
| Holds out of sample | partly |
3What was tried
Every one of the 400 recorded variations, by the Sharpe ratio it achieved. The promoted version sits at the far edge of what chance alone would produce from this many tries.
4What it means for an allocator
- Expect results nearer the restated figure than the reported one.
- The edge is real but modest; size and fees should reflect that.
- Ask for the next twelve months to be reported against this baseline.
5What we saw, and what we did not
We worked from the return series and the record of variations supplied. We did not see code, signals or positions, and none left the manager’s environment. The restated figure is only as complete as the record of what was tried.