Tools

Information Coefficient Calculator

Paste your signal's predictions and the realized forward returns for the same names. The tool computes the Pearson IC, the Spearman rank IC, and a t-stat for significance, and plots predictions against returns.

What it computes: the Information Coefficient (IC) is the cross-sectional correlation between a signal's forecasts and the returns that follow. The rank IC (Spearman's ρ) is the conventional measure because it's robust to outliers and only assumes a monotonic relationship; the Pearson IC assumes linearity. New to this? Start with the Information Coefficient guide.

Open source — view or fork the calculator on GitHub (MIT, single dependency-free HTML file). Related tools: Signal Decay Calculator · Backtest Overfitting Simulator · Cointegration & Pairs Trading Simulator · Signal Combination Simulator · Correlation Heatmap · VPIN & the Volume Clock · Signal Skill Explorer.

Predictions and realized returns

Paste two columns — your signal's predictions (forecast scores, any scale) and the realized forward returns for the same names. One value per line; rows are matched in order, so both columns must be the same length.

Frequently asked

What is the difference between Pearson and rank (Spearman) IC?
Rank IC correlates the ranks of forecasts and returns: it is robust to outliers, only assumes a monotonic relationship, and is the conventional measure in practice. Pearson IC assumes a linear relationship and is more sensitive to extreme values. Report rank IC unless you specifically need the linear version.
What do I paste in?
Two aligned columns for the same names at the same time: your signal's predictions (any scale — only the ordering matters for rank IC) and the realized forward returns that followed. The tool standardises and correlates them. Your data never leaves the browser.
How much data do I need for the IC to mean anything?
The t-stat the tool reports (approximately IC times the square root of N minus 2) tells you. With a handful of names an IC is almost meaningless; significance usually needs a few hundred observations, and a single cross-section's IC is far noisier than an IC averaged over many periods (the ICIR).
What IC should I expect from a usable signal?
A sustained rank IC of roughly 0.03 to 0.06 is a genuinely useful equity signal. A much higher figure on real out-of-sample data usually warrants a check for look-ahead bias or overfitting. See the Information Coefficient guide for context.
Is anything stored or uploaded?
No. The calculation runs entirely in your browser and nothing is sent anywhere. It is also open source on GitHub if you want to verify or self-host it.