Measuring a Signal's Decay Curve

A signal doesn't just have a level of skill — it has a shelf life. A decay curve measures how that skill fades with time, and it is often more useful than the headline information coefficient, because it tells you the one thing a single IC number can't: how long you're allowed to wait before trading on it, and how often you need to refresh it.

What you're actually plotting

Take a signal computed at time zero for every name in the universe. Instead of correlating it with only the next day's return, correlate it separately with the return over day 1, day 2, day 3, and so on out to whatever horizon is plausible. Plotting IC against the number of days since the signal fired gives the decay curve. A fast-decaying signal — most short-term reversal and order-flow signals — drops toward zero within a few days. A slow-decaying signal — most valuation and quality signals — stays above zero for months.

The shape matters as much as the speed. A curve that decays smoothly can be traded with a simple fixed holding period. A curve that dips negative before recovering (common in reversal signals crossed with momentum) means a naive fixed hold will trade straight through a bad patch.

A worked example

A researcher measures a value signal's IC at each horizon: day 1 it's 0.030, day 5 it's 0.026, day 20 it's 0.018, day 60 it's 0.009, day 120 it's 0.002. The curve is still clearly positive at 60 days and only flattens out near zero around day 120. That says the signal can be traded on a monthly or even quarterly rebalance without giving up much of the day-1 edge, and that a weekly rebalance would mostly be paying turnover costs to capture a curve that barely moved.

Compare that to a short-term reversal signal: day 1 IC of 0.05, day 3 IC of 0.01, day 5 IC of −0.005. Here the curve crosses zero and goes negative by day 5 — holding past day 3 doesn't just waste the edge, it actively fights it. The right holding period for this signal is two to three days, full stop.