TeX 53.6%
Python 46.2%
1\section{Robustness}2\label{sec:robustness}34The Internet Appendix reports the full battery specified in the analysis5protocol. The main text highlights the checks that discipline6interpretation. \emph{Tick-regime}: pre-2001 minute returns embed7eighth/sixteenth-tick bid--ask bounce; all secular comparisons are re-run on8range-based measures, on 5- and 15-minute aggregation, and within9post-decimalization subsamples. \emph{Coverage}: the 2020 source-coverage10expansion is absorbed by fixed-rule lagged top-$N$ universes; results are11reported for $N\in\{500,1000,3000\}$. \emph{Universe composition}: ETFs and12likely non-common instruments (ticker-shape heuristic) are excluded in13robustness runs. \emph{Corporate actions}: split windows are excluded14throughout; sensitivity to window width and detection thresholds is15tabulated. \emph{Thresholds}: source shocks at $|z|\ge4,5,6$ and receiver16thresholds at $|z|>2,3$ leave the baseline null unchanged. \emph{Inference}:17day-block and month-block bootstrap agree. Table~\ref{tab:robustness}18collects the headline checks: the market-wide null and the $+64\times10^{-4}$19network differential are stable across the post-decimalization subsample,20mid-day-only events, announcement-window exclusions, calm and crisis years,21higher source thresholds, and non-common-ticker exclusions, and the neighbor22differential measured on placebo events alone is zero ($-2.7\pm2.8\times2310^{-4}$)---the permuted-benchmark falsification the design commits to. The24one systematic pattern is that the network differential \emph{declines} with25source magnitude ($+64$ at $|z|\ge6$ to $+34\times10^{-4}$ at $|z|\ge12$),26consistent with the largest events being firm-specific news that neighbors27correctly do not price.2829\input{tables/table8_robustness}30