spb/wp15_uqo

1\section{Introduction}2\label{sec:introduction}34When one security suffers a violent price move, does the shock stay local or5does it spread? The answer shapes almost everything that happens inside a6trading day: how fast prices aggregate information, how much liquidity7providers must charge for standing in the market, how execution algorithms8should schedule parent orders, and whether an idiosyncratic event can grow9into a market-wide dislocation. Regulators confront the same question from the10other side: episodes such as May 6, 2010 showed that modern equity markets can11move from normal functioning to synchronized collapse within minutes, faster12than any human intervention. Yet the empirical record on intraday shock13transmission is built almost entirely from short samples, small panels, or14single episodes. Whether the propagation mechanism itself has changed over the15three decades in which trading moved from exchange floors to fragmented,16algorithmic, ETF-intermediated electronic markets is, to our knowledge,17undocumented.1819The tension motivating this paper is that theory predicts change in both20directions. Faster, cheaper, more competitive markets should absorb ordinary21idiosyncratic shocks more efficiently: liquidity provision reacts within22seconds, arbitrage capital is abundant in normal times, and information is23incorporated nearly continuously. But the same institutions that supply this24efficiency---high-speed market making with tight inventory limits, cross-asset25arbitrage, index and basket intermediation---withdraw or transmit pressure26precisely when many securities move together. Modern markets may therefore27absorb ordinary shocks more efficiently while transmitting extreme shocks more28rapidly and discontinuously.2930We ask how the speed, breadth, persistence, and asymmetry of intraday shock31propagation across U.S.\ equities evolved over the past three decades, and32under which market conditions an initially idiosyncratic shock becomes33sectoral or market-wide.3435Our laboratory is a panel of \DataRows{} one-minute OHLCV bars, originally36sourced from Finnhub.io, covering37\DataTickers{} tickers---including delisted securities---from January 199238through March 2026. The data are audited exhaustively before use: observed39trading dates match the official exchange calendar exactly over thirty-four40years, timestamp conventions are verified on both sides of every41daylight-saving transition, and integrity violations are negligible42(\DataOHLCViol{} impossible OHLC bars in \DataRowsBn{} billion). Because the43source provides bars only for minutes with trades, and identifies securities44by ticker alone, the empirical design works with liquidity-screened,45lagged-rank universes, ticker spells rather than tickers, and detected46corporate-action windows; Section~\ref{sec:data} documents every screen.4748Three findings organize the paper. First, \emph{markets absorb isolated49idiosyncratic shocks---and we can measure how completely}. We detect50\NEventsRealMM{} million one-minute events in which a single top-500 stock51moves by at least six robust standard deviations of its own residual return52while the rest of the market is calm. Relative to matched within-stock53placebos, the abnormal share of other securities responding within 5 to 12054minutes is economically zero (point estimates of $-1$ to $-3\times10^{-4}$55with day-block bootstrap standard errors of about $1\times10^{-4}$), in every56decade, for positive and negative shocks alike. Idiosyncratic events die where57they land.5859Second, \emph{the transmission of common information has become nearly60instantaneous}. In 1996 the average pairwise correlation of one-minute returns61among the hundred largest stocks was \EppsNinetySixOne{}---economically zero,62the classical Epps effect---while thirty-minute returns already comoved at63\EppsNinetySixThirty. By 2015 the one-minute correlation was64\EppsFifteenOne{} against \EppsFifteenThirty{} at thirty minutes: the Epps65curve has flattened almost completely. Information that once took half an hour66to spread across the cross-section now spreads within the minute.6768Third, \emph{tail risk has migrated from the security to the minute}. Ordinary69extreme moves became rare: the frequency of one-minute returns exceeding 1\%70fell from roughly \ExtremeGtOneNineties{} per 10,000 in the 1990s to71\ExtremeGtOneTens{} per 10,000 in the 2010s---a factor of about72\ExtremeDropFactor. But synchronized extremes did not: minutes in which more73than 2\% of the universe jumps beyond $4\sigma$ simultaneously---systemic74co-jump minutes---spike by orders of magnitude in 2008 and 2020, and the75frequency of $|r|>5\%$ minute returns rose from \ExtremeGtFivePreCovid{} per7610,000 in 2017--2019 to \ExtremeGtFiveCovid{} in 2020, a factor of77\ExtremeCovidFactor. The joint pattern---absorption of the idiosyncratic,78synchronization of the systemic---is the modern form of market fragility.7980Identification receives explicit attention throughout. The central threat is81that securities respond to common information rather than to each other. Our82shock definition residualizes against a leave-one-out market factor with83strictly lagged loadings and scales; isolation filters require a calm market84and fewer than 2\% simultaneous extremes at the event minute; every event is85compared to a placebo in the same stock at the same minute of day on a calm86day; and the pre-event window shows no abnormal response, ruling out87anticipation or mechanical selection. What the isolation filters remove---the88simultaneous-extreme minutes---we do not discard but study separately as the89systemic co-jump sample. Common-information contamination is thus measured,90not assumed away.9192Mechanically, we ask \emph{where} propagation should hide if it exists:93responses concentrated among the shocked stock's closest lagged-correlation94neighbors, cumulative rather than point-in-time responses, and state95dependence across calm and stressed markets and across shock magnitudes.96Section~\ref{sec:mechanisms} reports these probes.9798The paper contributes to four literatures. To the contagion literature, which99measures transmission at daily or lower frequency for small panels, we100contribute event-level evidence from the full U.S.\ cross-section at the101minute scale over thirty-four years. To the market-microstructure and102high-frequency-trading debate, which contrasts normal-times liquidity gains103with episodic fragility, we contribute the missing time series: propagation104and synchronization measured uniformly across the entire structural105transformation of U.S.\ trading. To the co-jump econometrics literature, we106contribute a long-sample account of how systemic co-jump frequency and107severity evolved. And to the measurement literature, we document what OHLCV108bars can and cannot identify about liquidity and transmission, with every109limitation quantified.110111Section~\ref{sec:data} describes and audits the data.112Section~\ref{sec:measurement} defines shocks, outcomes, and networks.113Section~\ref{sec:design} presents the empirical design.114Section~\ref{sec:results} reports the descriptive transformation of the market115and the baseline propagation results. Section~\ref{sec:mechanisms} examines116networks, cumulative responses, and co-jumps. Section~\ref{sec:robustness}117summarizes robustness, and Section~\ref{sec:conclusion} concludes.118