alpha-factor-library

To benchmark cross-sectional alpha factors, you select a bundled alpha zoo and a target universe, then run IC/IR evaluation to generate aggregate metrics and an HTML report. This avoids starting factor research from scratch by evaluating prebuilt libraries.

An alpha zoo is a curated library of cross-sectional alpha factors. Factor benchmarking works by calculating the Information Coefficient and Information Ratio of these alphas across a specified universe and time period to produce aggregate performance metrics.

Yes, you can evaluate a custom factor by supplying a factor CSV for ad-hoc factor analysis. This runs the same IC/IR factor evaluation over your specified universe and period when your factor is not part of the bundled zoos.

You retrieve metadata for a specific alpha by requesting the registry entry for a named alpha identifier. This fetches the alpha's metadata from the zoo without needing to run a full backtest.

Constraints for IC/IR evaluation include maintaining mutual exclusivity between alpha_id and zoo parameters, and enforcing operator safety constraints such as no lookahead. Correct zoo and factor selection is required to produce valid aggregate metrics.

Yes, factor benchmarking supports evaluating an entire alpha zoo at once. You can bench all alphas within a zoo over an investable universe and time period, generating aggregate IC/IR metrics and an HTML report.