What Is Rank IC?

Rank IC is an information coefficient (IC) calculated from ranks rather than raw values. At each evaluation date, assets are ranked by their signal or model score and by their realised forward return; the correlation of those two rank vectors, usually Spearman rank correlation, is the period’s rank IC. Researchers normally summarise the series of period ICs with its mean, standard deviation, and distribution across market regimes.

Rank IC asks whether a signal orders assets correctly, rather than whether it forecasts the exact size of each return. It is therefore common in cross-sectional factor research and machine-learning stock selection, where portfolio construction often depends on relative ranking. A cross-sectional rank IC is undefined at a date when the feature has no variation across the tradable assets; a time-series evaluation or another suitable metric is needed for such a feature. Rank IC is less sensitive than a raw Pearson IC to extreme returns and nonlinear score scales, but it is still vulnerable to look-ahead bias, an inappropriate asset universe, and multiple testing.

See also

• Cross-sectional analysis

• Information coefficient (IC)

• ICIR

Research and literature

• Richard C. Grinold, The Fundamental Law of Active Management, The Journal of Portfolio Management 15(3), 1989.

• Shihao Gu, Bryan Kelly, and Dacheng Xiu, Empirical Asset Pricing via Machine Learning, The Review of Financial Studies 33(5), 2020.