New Algorithm Improves Global Factor Detection in Asset Pricing
An iterative filtering method leverages eigenvector delocalization to identify weak market factors that traditional spectral tests miss.
Identifying the number of global factors in high-dimensional correlation matrices is critical for asset pricing, yet signal-to-noise separation often fails when the number of variables is comparable to the number of observations. This difficulty peaks near the BBP phase transition, where weak factors are frequently mistaken for random noise at the spectral edge.
Researchers have developed an Iterative Global Factor (IGF) algorithm to solve this by combining adaptive Marčenko–Pastur edge recalibration with a participation-ratio (PR) delocalization filter. While traditional methods rely on eigenvalues alone, IGF examines the structure of the eigenvectors. It distinguishes between leading coherent eigenvectors and idiosyncratic sample eigenvectors, retaining only those that meet specific extension criteria.
https://arxiv.org/abs/2607.06908
Monte Carlo simulations show that IGF recovers the true number of factors near the BBP transition where eigenvalue-only criteria remain ambiguous. When applied to S&P 500 returns via a moving-window calibration, the method detected a more dynamic set of global factors than the Onatski test, finding a median of 7 factors.
Integrating spectral separation with eigenvector delocalization allows for a more precise estimation of global factors in complex financial matrices. This approach reveals a richer underlying factor structure than previously detectable.