Concepts / IC Decay

IC Decay

5 min read · Concept

The single best diagnostic for whether your factor is real edge or rear-view-mirror data fitting. Cheap to compute, hard to fake.

What IC is

The Information Coefficient is the Spearman rank correlation between your factor score at time t and the forward return over the next horizon (typically 1 day, 1 week, 1 month). A factor that perfectly ranks future returns has IC = 1; a coin flip has IC = 0; a factor that ranks them backwards has IC = -1.

The framework comes from Grinold & Kahn, Active Portfolio Management (1999). Spearman (rank) rather than Pearson (linear) because factor models are about ordering the universe, not predicting return magnitudes — you care that the top decile beats the bottom decile, not that you got the basis points exactly right.

The formula

# At each rebalance date t:
factor_score_t  = your factor value for each asset
forward_return  = return from t to t + horizon

IC_t = spearman_rank_corr(factor_score_t, forward_return)

# Lifetime IC = mean(IC_t across all rebalance dates)
# IC standard error = std(IC_t) / sqrt(n_periods)
# IC information ratio = lifetime_IC / IC_standard_error

Reading the lifetime IC

|IC| > 0.05: meaningful for a single-name equity factor. Quality, momentum, value all live in roughly this range over long samples.
0.02 < |IC| < 0.05: marginal — could be edge, could be noise depending on sample size and decay shape. Don't deploy on this alone; needs corroborating evidence from walk-forward.
|IC| < 0.02: noise. The factor is not predictive at any horizon you care about. Stop tuning it.
Negative IC: the factor is anti-predictive. Either flip the sign and trade the inverse, or recognise that what you thought was alpha is the opposite. Negative IC at lifetime is usually a sign you've mislabelled the direction.

Decay: the most diagnostic part

Compute IC at multiple forward horizons (1 day, 5 days, 20 days) and look at how IC changes. The decay ratio is IC(t+1) / IC(t) — how much of the signal survives one period forward.

Decay ratio > 0.9: persistent signal. Quality and value factors look like this — slow-moving fundamentals generate slow-decaying ICs. These are the factors with the longest shelf life.
Decay ratio 0.7–0.9: normal for cross-sectional momentum. The signal lives 1–3 months and you need to rebalance inside that window.
Decay ratio < 0.7: fading fast. Either the alpha is short-horizon (intraday or microstructure) and you need execution speed to capture it, or it's crowded out — by the time your data arrives, the trade is gone.
IC flips sign at later horizons: classic mean-reversion fingerprint. A momentum signal that decays positive then negative is actually a short-horizon overshoot indicator — useful, but not what you thought you were trading.

What to watch in the result card

IC IR > 0.5. The IC information ratio (lifetime IC / standard error of IC) is the noise-adjusted version. IR above 0.5 means the IC is statistically distinguishable from zero.
Decay shape consistent with the rebalance frequency. If you rebalance monthly but your IC is gone by horizon t+5 days, you're holding through the dead zone. Either rebalance faster or redesign the signal.
IC by sub-period stable. A lifetime IC of 0.04 made up of 0.10 in the first half and -0.02 in the second is a broken factor. Check the rolling IC chart, not just the headline.

Common mistakes