μαMicro AlphasSignal Research

signal evaluation

Signal Aggregation

Combining multiple alpha signals into a single composite forecast to improve stability and reduce reliance on any one signal.

Signal aggregation combines predictions from multiple alpha signals into one composite forecast. The combined signal benefits from diversification: signals with low pairwise correlation but individually positive IC generate a composite with higher IC-to-noise than any single signal, improving the overall Information Ratio.

Common aggregation methods

  • Equal weighting — simple average of normalized (z-scored) signals. Robust when signals have similar IC and no strong views on relative quality.
  • IC-weighted averaging — weight each signal by its recent rolling IC. Gives more influence to whichever signal has been most predictive recently.
  • Shrinkage / Bayesian averaging — blend empirical weights toward a prior (e.g., equal weighting), reducing sensitivity to estimation error in IC estimates.
  • Machine learning ensembles — stacking, boosting, or learned nonlinear combinations. More expressive but more prone to overfitting.

A key insight from the Fundamental Law applies here: combining N uncorrelated signals of equal IC increases the composite IC by √N, directly improving the Information Ratio.

Related terms

Information Coefficient (IC)The correlation between a signal's forecasts and subsequent realized returns; the foundational measure of signal predictive skill.Rolling ICThe Information Coefficient computed over a rolling time window, used to track signal decay, stability, and regime-conditional behavior.Fundamental Law of Active ManagementGrinold & Kahn's formula: IR ≈ IC × TC × √N, linking Information Ratio to signal skill, transfer coefficient, and breadth.Ensemble MethodsTechniques that combine predictions from multiple models to reduce variance and improve out-of-sample robustness.BreadthThe number of independent investment decisions made per year; the N in the Fundamental Law of Active Management.

Related articles

← Back to glossary