A1304
Title: Full of noise? A bootstrap confidence interval that controls size in Diebold-Yilmaz networks
Authors: Milan Csaba Badics - Corvinus University of Budapest (Hungary) [presenting]
Abstract: Over the past decade, Diebold-Yilmaz (DY) connectedness has become a workhorse for empirical network analysis, yet inference has received relatively little attention. The few papers that attempt it typically adopt Efron's percentile bootstrap confidence intervals but rarely evaluate whether those tests achieve nominal size, despite its importance for policy and trading. It is demonstrated that even under favorable conditions, long samples, low persistence, homoskedastic innovations, and unbiased impulse responses, the DY connectedness measures exhibit skewness and bias that inflate rejection rates. A Hall-type bootstrap confidence interval is proposed with coverage calibration via bootstrap-after-bootstrap, delivering size-correct inference for the connectedness table. In Monte Carlo simulations, the proposed interval achieves an empirical size close to the nominal level. The method reduces spurious link detections, supporting more credible policy analysis and more reliable minimum-connectedness portfolio construction.
