Discover calibration¶
Generated by scripts/discover_calibration.py (budget 300, 19 shuffled markets per search, costs 0.5 bp per side).
Noise: how often a search names an indicator on a random walk¶
| Bars | Searches | Named | Rate | 95 % interval | Median best Sharpe of the search |
|---|---|---|---|---|---|
| 1500 | 60 | 0 | 0.0% | 0.0% - 6.0% | 6.24 |
The last column is what a plain search would report as its best strategy on the same data.
Power: how often a planted AR(1) momentum edge is found¶
| Bars | AR(1) coefficient | Searches | Found | Rate | 95 % interval |
|---|---|---|---|---|---|
| 1000 | 0.05 | 12 | 0 | 0% | 0% - 24% |
| 1000 | 0.10 | 12 | 1 | 8% | 1% - 35% |
| 1000 | 0.15 | 12 | 3 | 25% | 9% - 53% |
| 1000 | 0.20 | 12 | 5 | 42% | 19% - 68% |
| 3000 | 0.05 | 12 | 0 | 0% | 0% - 24% |
| 3000 | 0.10 | 12 | 3 | 25% | 9% - 53% |
| 3000 | 0.15 | 12 | 9 | 75% | 47% - 91% |
| 3000 | 0.20 | 12 | 12 | 100% | 76% - 100% |
| 10000 | 0.05 | 12 | 2 | 17% | 5% - 45% |
| 10000 | 0.10 | 12 | 12 | 100% | 76% - 100% |
| 10000 | 0.15 | 12 | 12 | 100% | 76% - 100% |
| 10000 | 0.20 | 12 | 12 | 100% | 76% - 100% |
Run time 504 s. Costs of 5 bp per side remove the planted edge (see the unit tests); a search that names nothing is a result.