Skip to main content
Factor optimization runs several factor analyses with different parameter combinations. It is useful for sensitivity checks, but it does not replace out-of-sample validation.

​
How optimization works

1

Save the current factor

Qube saves the current code and uses the latest factor version as the optimization baseline.
2

Design the search grid

Ask Qube to identify sensible code or analysis parameters. Keep the grid at 64 combinations or fewer so you can inspect each result.
3

Run the combinations

Each combination creates a child factor analysis. The task keeps progress, status, metrics, and failure reasons for every child.
4

Review the best and nearby results

Compare the objective, IC metrics, return curves, and sensitivity across neighboring combinations.
5

Apply and validate

Apply the best parameters to create a new factor version, then run a fresh analysis on a separate period.
The default objective is ic_mean. Applying the best parameters changes the analysis parameters, not the factor code.
Do not select an isolated historical peak as proof of quality. Prefer a stable region of parameter values and confirm it outside the optimization period.
After validation, use Turn a factor into a strategy to add position and risk rules.