2026/02/04 by Murad Farzulla · 1 voice · 1 citation
Economics, Econometrics and Finance · Mathematics · #q-fin.ST #q-fin.CP #stat.AP
arxiv published 2026/02/04 · arxiv updated 2026/07/11
Do cryptocurrency markets process infrastructure failures differently from regulatory shocks? We study both moments of the return distribution on one shared sample (50 events, six assets, 2019-2025), fitting a GJR-GARCH-X model under matched dependence-robust inference. We treat event inclusion as a measured design parameter: rather than asserting the selection-on-the-dependent-variable objection away, we trace the variance differential across the inclusion screen and measure the selection bias directly. The result is a scope condition -- under curated, high-salience identification the differential is sizeable (4.88×) but selection-conditional: a mechanical impact filter on a broad reconstructed pool collapses it to 1.3-1.6×. Identification is half the story; inference is the other. The curated multiplier is not distinguishable from zero once cross-asset dependence and heavy tails are respected: a Student-t-copula CCC-GARCH-X bootstrap (our inference of record) returns p ≈ 0.32, and because the six per-asset coefficients are strongly cross-correlated the contrast's effective sample size is nearer three than six (design-effect p ≈ 0.07-0.15). A naive i.i.d. test had reported an apparently decisive fivefold effect, but that significance was an artefact: pseudoreplication across correlated assets compounded by a heavy-tail-misspecified bootstrap. The first moment tells the same story -- a +7.19 pp cumulative-abnormal-return difference a block bootstrap cannot distinguish from zero (p = 0.283). Under correct inference the asymmetry is directional but unresolved. The contribution is a portable inference toolkit -- an inference ladder and a Monte-Carlo size study -- for diagnosing how cross-asset event studies in heavy-tailed markets manufacture significance, demonstrated where it dissolves a fivefold result the author had himself published.