Rolling Information Coefficient
Quick Reference
| Property | Value |
|---|---|
| Dimension | quality |
| Category | statistics |
| Version | v1.0 |
| Output Column | information_coefficient |
Rolling Information Coefficient: rank correlation between signal and forward return over window
Formula
rolling_rank_corr(signal, forward_return, window)
CDM Inputs
| Column | CDM Table | Description |
|---|---|---|
signal | cdm_* | CDM source table |
forward_return | cdm_* | CDM source table |
Parameters
| Parameter | Type | Default | Description |
|---|---|---|---|
window | integer | 300000 | Window for IC estimation |
Output
Column: information_coefficient
Rolling rank correlation (Spearman) between signal and return
Market Intuition & Trading Rationale
Rolling Information Coefficient (IC) measures the rank correlation between a signal's predictions and actual forward returns: rolling_rank_corr(signal, forward_return, window). Unlike Pearson correlation (which assumes linearity and is sensitive to outliers), Spearman rank correlation captures monotonic relationships — if higher signal values consistently correspond to higher forward returns, IC is high, regardless of whether the relationship is linear.
IC is the standard metric in quantitative finance for evaluating signal quality. An IC of 0.05–0.10 is considered good in equity markets; an IC of 0.02–0.05 is typical. The rank-based approach makes IC robust to fat-tailed return distributions and outlier signals — a few extreme signal values won't distort the estimate.
IC is directional — positive IC means the signal correctly predicts direction, negative IC means the signal consistently predicts the wrong direction (which is also useful — just flip the sign). IC near zero means the signal has no predictive power. The square of IC (IC²) approximates the R² of a cross-sectional regression of returns on signals.
Usage Cases
- Signal validation: Before deploying a new FeatureType, compute its rolling IC against forward returns. If IC is consistently positive and statistically significant, the feature captures genuine predictive information. If IC oscillates around zero, the feature is noise.
- Signal decay monitoring: Track rolling IC over time. Declining IC means the signal's predictive power is eroding — the market may have adapted, or the alpha is being crowded out. Trigger a review when IC drops below its historical confidence interval.
- quality assessment: Used as the quality dimension in feature sets. A signal with high SNR but low IC has stable output but no predictive power — it's consistently wrong. A signal with low SNR but high IC is noisy but directionally correct — worth keeping with proper risk management.
YAML Definition
name: rolling_information_coefficient
description: 'Rolling Information Coefficient: rank correlation between signal and
forward return over window'
category: statistics
dimension: quality
version: v0.9.0 (Beta)
required_inputs:
- signal
- forward_return
output_column: information_coefficient
output_description: Rolling rank correlation (Spearman) between signal and return
tags:
- quality
- ic
- correlation
parameters:
window:
type: integer
description: Window for IC estimation
required: false
default: 300000
formula: rolling_rank_corr(signal, forward_return, window)