How the optimization works
Every indicator with a period range is expanded into concrete instances. If there are several sources, one instance is created for each source × period combination.
Group A: indicator × threshold × operator. Group B: every pair of instances × operators. With MaxAnd ≥ 2, all the AND combinations on top of that.
Each combo is backtested over the In-Sample period. By default the work is spread across every CPU core; if you tick GPU and an OpenCL device is available, the sweep runs on the graphics card (one GPU thread per combo), typically 10-15× faster. Entry at the next Open; exit by intrabar TP/SL or by signal at the close.
Only strategies meeting the MinTrades floor get through. The results are ranked by the criterion you chose (Win%, PF, Net P. %…). Select any of them to see its full report.
🧮 Example — the combinatorial explosion
This is where you see why MaxAnd is so powerful… and so dangerous. Suppose the generation produces 500 single conditions (indicators × periods × operators × thresholds). The number of strategies to backtest, by MaxAnd:
From 500 to 20 million just by raising MaxAnd from 1 to 3 (and with more base conditions the jump goes into the billions). Two consequences: (1) this is why the Random mode exists — to sample at random when the space is astronomical; (2) with that many tests, some will win by pure chance — which is why MaxAnd=3 demands a high MinTrades, ranking by ProfitFactor and validation afterwards (PBO/DSR, Walk-Forward).
In-Sample vs Out-of-Sample
AniQuant splits the data. The In-Sample (IS) period is used to find the best conditions; the Out-of-Sample (OS) one checks whether the strategy generalizes to data the system never saw. Each trade counts in the segment where it closes: one opened at the end of IS and closed in OS counts in OS, both in the Lab grid and in Performance.
AniQuant can be tried free for 14 days, with every module and no card. If you have the book, 30.