Iterative detection of global factors near the BBP phase transition

Battle Count: 75

The paper addresses a fundamental problem in quantitative finance: determining the number of global factors in high-dimensional asset return data. This is directly relevant to factor investing, portfolio construction, and risk model estimation. The IGF algorithm provides a more nuanced factor detection than the Onatski test (median 7 vs. 1 factor for S&P 500), which could improve factor-based trading strategies. However, the paper is primarily methodological/theoretical and does not directly propose trading strategies or backtest performance. The detected factors are statistical and require further economic interpretation before direct trading application.

The algorithm requires: (1) eigendecomposition of correlation matrices, (2) iterative median-based noise estimation, (3) Marčenko-Pastur edge computation, (4) participation ratio calculation for each eigenvector, and (5) sequential testing. The core steps are computationally straightforward (O(p^3) for eigendecomposition, O(p) for PR). However, understanding the RMT background (BBP transition, MP law, Tracy-Widom), calibrating the threshold tau via synthetic experiments, and correctly implementing the iterative recalibration require significant expertise in random matrix theory and high-dimensional statistics.

The paper provides detailed algorithmic steps, parameter settings for Monte Carlo simulations (p grid, n=400, sigma_b^2=0.08, sigma_f^2=0.000158, sigma_e^2=0.0045, b=1), and the empirical S&P 500 data setup (p=417, q=1/2, n=834, 185 moving windows). However, no code repository is provided, and data is available only upon request. The threshold tau=0.3 is calibrated via synthetic moving-window experiments described in the paper.