A signal's half-life is the forward-return horizon T₁/₂ at which the signal's IC has fallen to half of its peak value (typically the IC at the shortest measured horizon). It operationalizes the speed of alpha decay into a single actionable number.
Measuring it empirically
- Compute the signal's IC at successive horizons: IC(1d), IC(5d), IC(10d), IC(21d), IC(63d), etc.
- Fit a decay model, e.g., exponential: IC(h) = IC₀ × e^(−λh)
- Solve for T₁/₂ = ln(2) / λ
Implications for strategy design
- A one-day half-life implies the signal is stale within a day. High-frequency execution at very low transaction costs is required — not viable for most institutional strategies.
- A 20-day half-life is compatible with weekly rebalancing, a much more cost-efficient implementation that is accessible to a wider range of portfolio sizes.
Regime changes (detected via rolling IC trends) can shorten a signal's effective half-life if competitors arbitrage it more aggressively or if a structural change reduces the persistence of the underlying information advantage.