Abstract
Controlling the false discovery rate (FDR) under arbitrary correlation remains challenging: standard methods either assume independence or sacrifice substantial power through conservative corrections. We propose a dueling double bootstrap (DDB) procedure that provides adaptive FDR control without parametric assumptions on the dependence structure. The outer ``dueling" layer bounds the proportion of true nulls with explicit finite-sample probability guarantees; the inner layer calibrates p-value thresholds using this bound. Simulations demonstrate that DDB maintains nominal FDR while achieving 2–4 times higher power than Benjamini–Yekutieli corrections under strong correlation. Applying DDB to 344 equity and bond portfolios with substantial cross-sectional dependence, we find that 68–87% of anomalies replicate under rigorous error control. The procedure offers a frequentist, nonparametric alternative to existing approaches for multiple testing under dependence.
Download
Citation
Bianchi, Daniele, Junpei Komiyama, Ken McAllin, and Leyong Yang. “Correlated False Discoveries and Asset Pricing Anomalies.” Working paper.
@unpublished{BKMY25,
author = {Daniele Bianchi and Junpei Komiyama and Ken McAllin and Leyong Yang},
year = {2025},
title = {Correlated False Discoveries and Asset Pricing Anomalies}}