Console / Research / Optimization
Parameter selection engineered to resist overfitting.
Walk-forward optimization and held-out evaluation replace the single in-sample fit, so chosen parameters reflect a robust region of performance rather than a fragile peak.
The Console's optimizer searches a parameter space with walk-forward re-estimation, evaluating every candidate on windows held out of the search. What it returns is a region of stable performance and the record of how it was found.
Parameters are chosen on periods the optimizer never saw
Parameters are re-estimated on a rolling basis and evaluated on periods the optimizer never saw, rather than optimized against the entire history at once, which rewards memorization over genuine structure. The performance that counts is the performance on the unseen windows.
See the whole parameter space
The Console surfaces performance across the parameter space, so a broad plateau where nearby settings all perform well is distinguishable from a lone spike. Stability diagnostics flag when a top result sits on a narrow peak.
- Held-out evaluation on data withheld from the search, so selection cannot borrow from the test.
- Parameter landscapes that show the robustness of a region, not only the ranking of individual results.
- Stability diagnostics that flag when a top result sits on a fragile peak.
For institutions
Overfitting is controlled by construction.
A model-risk review turns on two questions: how the parameters were selected, and what kept the selection away from the data it was tested on. Both are answered by the method itself, and the full search is retained alongside the chosen configuration.
Next step
Optimization, live on your strategies and venues.
Response within two business days