Do Cryptocurrency Markets Differentiate Infrastructure from Regulatory Shocks? A Multi-Moment Event Study with Dependence-Robust Inference

Version 4.0.0 — merged multi-moment edition. This version consolidates and supersedes the two predecessor preprints: the return-level companion (Same Returns, Different Risks, v3, arXiv:2602.07046) and the conditional-variance study (Infrastructure vs Regulatory Shocks, Research Square 10.21203/rs.3.rs-8323026). It tests the infrastructure-versus-regulatory question at both moments of the return distribution under one dependence-robust design and, correcting the inference of the variance predecessor, reports an honest dual failure-to-reject. Abstract Do cryptocurrency markets process infrastructure failures (exchange outages, protocol exploits, network disruptions) differently from regulatory shocks (enforcement actions, policy changes)? An earlier version of this work reported that they do, with infrastructure events generating a roughly fivefold larger conditional-variance response, apparently significant. We re-test the claim properly at both moments of the return distribution over a single shared sample (50 events, six assets, January 2019–August 2025), using one dependence-robust design instantiated twice: an event-level block bootstrap with the Ibragimov–Müller few-cluster test at the first moment, and a Student-t-copula CCC-GARCH-X bootstrap at the second. Neither moment supports a statistically distinguishable asymmetry. Cumulative abnormal returns differ by +7.19 percentage points (block-bootstrap p=0.283, Ibragimov–Müller p=0.295; 95% CI [-5.97, +20.48]); the variance point estimate is directional (≈2.6–5×) but the Student-t-copula bootstrap returns p=0.323. The apparent significance of the earlier variance result was an inference artefact: treating six strongly cross-correlated per-asset coefficients (ρ̄≈0.69) as independent, and drawing bootstrap innovations from a Gaussian rather than the fitted Student-t, both manufactured precision the data do not contain. The lead contribution is methodological: a four-part inference cautionary tale (pseudoreplication, cross-asset dependence, heavy-tailed bootstrap innovations, block-versus-parametric resampling) and a Monte-Carlo size study showing the naive i.i.d. rule rejects a true null 51% of the time at the nominal 5% level, while only the Student-t-copula bootstrap controls size (3.9%). Two findings survive: a scope condition (the directional variance differential concentrates in curated, high-salience event identification, ≈5×, versus a mechanical impact screen, 1.3–1.6×, neither significant) and a weekly, litigation-concentrated GDELT sentiment lead on volatility (10 of 18 pairs, 7 surviving FDR). Replication materials: github.com/studiofarzulla/crypto-event-study. v4.1.0 (2026-07-11): Dataset truth-up: the event catalog (events.csv, 50 events) and a full archive of the analysis code repository (github.com/studiofarzulla/crypto-event-study at commit 1baf97c) are now deposited alongside the paper, completing the data-availability statement. Journal correspondence is excluded from the code archive. Paper PDF and LaTeX source unchanged from v4.0.0.