TeX 53.6%
Python 46.2%
1\section{Introduction}2\label{sec:introduction}34When one security suffers a violent price move within a trading minute, does5the shock stay local or does it spread? The answer shapes how fast prices6aggregate information, how much liquidity providers must charge for standing7in the market, how execution algorithms should schedule parent orders, and8whether an idiosyncratic event can grow into a market-wide dislocation.9Regulators confront the same question from the other side: May~6, 2010 showed10that modern equity markets can move from normal functioning to synchronized11collapse within minutes, faster than any human intervention12\citep{kirilenko2017flash,menkveld2019flash}. Yet the empirical record on13intraday shock transmission is built almost entirely from short samples,14small panels, or single episodes. Whether the propagation mechanism itself15changed over the three decades in which trading moved from exchange floors to16fragmented, algorithmic, ETF-intermediated electronic markets is, to our17knowledge, undocumented.1819Theory predicts change in both directions. Faster, cheaper, more competitive20markets should absorb ordinary idiosyncratic shocks more efficiently:21liquidity provision reacts within seconds, arbitrage capital is abundant in22normal times, and information is incorporated nearly continuously23\citep{hendershott2011algorithmic,menkveld2013high,brogaard2014high}. But the24same institutions that supply this efficiency---high-speed market making with25tight inventory limits, cross-asset arbitrage, index and basket26intermediation---withdraw or transmit pressure precisely when many securities27move together \citep{brogaard2018high,cespa2014illiquidity,bendavid2018etfs}.28Modern markets may therefore absorb ordinary shocks more efficiently while29synchronizing more tightly on common ones. The title asks whether U.S.\30equity markets became faster, wider, and more fragile at the intraday horizon.31The answer this paper documents is: faster, yes; wider, no---transmission32became sharper rather than broader; more fragile, only in a specific and33bounded sense that we make precise.3435Our laboratory is a panel of \DataRows{} one-minute bars covering36\DataTickers{} tickers, including delisted securities, from January 199237through March 2026. The data are audited exhaustively before use: observed38trading dates match the official exchange calendar exactly over thirty-four39years, timestamp conventions are verified on both sides of every40daylight-saving transition, and integrity violations are negligible41(\DataOHLCViol{} impossible OHLC bars in \DataRowsBn{} billion). Because the42source provides bars only for minutes with trades, identifies securities by43ticker alone, and does not document its venue scope, the empirical design44works with liquidity-screened, lagged-rank universes, ticker spells rather45than tickers, and detected corporate-action windows; Section~\ref{sec:data}46documents every screen and every limitation we could not remove.4748Three findings organize the paper. First, \emph{isolated idiosyncratic shocks49do not raise market-wide tail activity by more than a few percent of its base50rate}. We detect \NEventsRealMM{} million one-minute events in which a51top-500 stock moves by at least six robust standard deviations of its own52residual return while the market factor is calm and fewer than 2\% of other53names move simultaneously. Relative to matched within-stock placebos, the54difference-in-differences estimate of the share of other securities with a55standardized residual beyond two in absolute value is between $-1$ and56$-3\times10^{-4}$ at horizons of 1 to 120 minutes, against a base rate of57about \BaseBreadthShare{}, with day-block bootstrap standard errors near58$1\times10^{-4}$. We also show, however, that real events occur in minutes59of elevated ambient cross-sectional activity (pre-event breadth is60\PreBreadthDiff$\times10^{-4}$ higher than for placebos), so that the61difference-in-differences and a pre-level-stratified comparison bracket the62market-wide response between zero and roughly \StratShareOfBase\% of the base63rate. Both bounds are economically small, and the pattern holds in every64decade and for positive and negative shocks alike.6566Second, \emph{what transmission exists is concentrated and, within the limits67of our measurement, has intensified}. The shocked stock's twenty closest68lagged-correlation neighbors are \NetDiffFive$\times10^{-4}$ more likely than69other receivers to show a tail exceedance five minutes later, net of the same70contrast for placebos---a response of about six percent of the neighbor base71rate, estimated with a standard error of \NetDiffFiveSE$\times10^{-4}$72(\NetDiffFiveTickerSE$\times10^{-4}$ with ticker clustering). The73differential is \NetDiffFiveNineties$\times10^{-4}$ in the 1990s,74\NetDiffFiveOughts{} in the 2000s, \NetDiffFiveTens{} in the 2010s, and75\NetDiffFiveTwenties{} in the 2020s. We present this increase with an76explicit caveat: neighbor sets are identified from one-minute residual77correlations in the preceding month, and the paper's own Epps evidence shows78that those correlations were close to zero in the 1990s, so part of the79secular rise reflects sharper identification of the network rather than80stronger transmission along it.8182Third, \emph{common information is impounded across the cross-section far83faster than it was, and tail risk within the post-decimalization era is84concentrated in market-wide minutes}. In 1996 the mean pairwise correlation85of one-minute returns among the hundred largest stocks was86\EppsNinetySixOne{} against \EppsNinetySixThirty{} at thirty minutes---the87classical Epps effect---whereas in 2015 it was \EppsFifteenOne{} against88\EppsFifteenThirty{}. The frequency of one-minute returns exceeding 1\% fell89from roughly \ExtremeGtOneNineties{} per 10,000 in the 1990s to90\ExtremeGtOneTens{} per 10,000 in the 2010s; part of that decline is the91mechanical disappearance of eighth and sixteenth ticks, and the decline92measured from 2002 onward is roughly tenfold. Minutes in which more than 10\%93of the universe moves beyond four robust standard deviations94simultaneously---severe co-jump minutes---are roughly an order of magnitude95more frequent in 2008 and 2020 than in surrounding calm years, and the96frequency of $|r|>5\%$ minute returns rose from \ExtremeGtFivePreCovid{} per9710,000 in 2017--2019 to \ExtremeGtFiveCovid{} in 2020. We do not claim that98such synchronization is secularly more frequent than in the 1990s: under the99same standardization the wide-tick 1990s produce severe-minute counts that100exceed 2008, which we attribute to discreteness rather than fragility and101which precludes a cross-regime comparison.102103Identification receives explicit attention throughout. The central threat is104that securities respond to common information rather than to each other. The105shock definition residualizes against a leave-one-out market factor with106strictly lagged loadings and scales; isolation filters require a calm market107factor and fewer than 2\% simultaneous extremes at the event minute; every108event is compared to a placebo in the same stock at the same minute of day on109a calm day; and the pre-event level of every outcome is reported rather than110assumed equal across arms. The minutes that the isolation filters remove are111not discarded but studied separately as the systemic co-jump sample. What the112design identifies is the predictive association between an isolated113idiosyncratic extreme and subsequent cross-sectional tail activity; it does114not separate fundamental information transmission from trading-channel115transmission, and we phrase every claim accordingly.116117The paper contributes to four literatures. To the contagion literature, which118measures transmission at daily or lower frequency for small panels, we119contribute event-level evidence from the full U.S.\ cross-section at the120minute scale over thirty-four years. To the market-microstructure and121high-frequency-trading debate, which contrasts normal-times liquidity gains122with episodic fragility, we contribute a uniform time series of propagation123and synchronization measured across the entire structural transformation of124U.S.\ trading. To the co-jump literature, we contribute a long-sample account125of how the frequency and severity of synchronized extreme minutes evolved,126with the comparability limits of such a series made explicit. And to the127measurement literature, we document what OHLCV bars can and cannot identify128about transmission, with every limitation quantified.129130Section~\ref{sec:literature} positions the paper. Section~\ref{sec:data}131describes and audits the data. Section~\ref{sec:measurement} defines shocks,132outcomes, and networks. Section~\ref{sec:design} presents the empirical133design and the inference. Section~\ref{sec:results} reports the descriptive134transformation of the market and the baseline propagation results, including135the design diagnostics. Section~\ref{sec:mechanisms} examines networks,136cumulative responses, and co-jumps. Section~\ref{sec:robustness} summarizes137robustness, and Section~\ref{sec:conclusion} concludes.138