TeX 53.6%
Python 46.2%
1\section{Related literature}2\label{sec:literature}34% All 110 citations verified at DOI level via OpenAlex/Crossref on 2026-07-13;5% see docs/LITERATURE_MAP.md. Novelty statements are qualified by the documented6% search of the same date.78This paper measures how quickly, how broadly, how persistently, and how9asymmetrically idiosyncratic intraday shocks travel across the U.S.\ equity10cross-section, and how that transmission evolved between 1992 and 2026. The11question sits at the intersection of several literatures: theories of12cross-asset shock transmission; empirical work on fire sales, flows, and13ownership linkages; the measurement of comovement, commonality, and liquidity14dynamics; lead--lag effects and information diffusion; network approaches to15contagion and systemic risk; the transformation of U.S.\ market structure and16the high-frequency trading debate; the econometrics of jumps, co-jumps, and17scheduled announcements; and the measurement toolkit for high-frequency and18OHLCV data. We take each in turn and close with the paper's positioning.1920\subsection{Theories of cross-asset shock transmission}2122Theory identifies two broad families of channels through which a shock to one23security moves others. The first is informational. \citet{king1990transmission}24model rational agents who infer information from \emph{other} markets' price25changes, so that an idiosyncratic ``mistake'' in one market transmits broadly---26a mechanism designed to explain why nearly all markets crashed together in27October 1987. \citet{kodres2002rational} show that cross-market portfolio28rebalancing of shared macro risk exposures transmits idiosyncratic shocks even29without common fundamentals, with severity increasing in information asymmetry.30The second family operates through balance sheets and funding.31\citet{kyle2001contagion} generate contagion as a wealth effect: convergence32traders who lose money in one market liquidate positions in others, raising33volatility and correlations exactly when capital is scarce. \citet{yuan2005asymmetric}34shows that borrowing constraints make crises, contagion, and asymmetric price35movements likelier after intermediate losses. \citet{brunnermeier2009market}36link market liquidity to funding liquidity in mutually reinforcing margin37spirals that explain sudden, correlated liquidity dry-ups and flight to38quality. Most directly related to our intraday setting,39\citet{cespa2014illiquidity} show that when liquidity providers learn from40\emph{other securities' prices}, illiquidity becomes contagious and the market41admits liquidity-crash equilibria---a mechanism they connect explicitly to42May~6, 2010. These theories jointly predict that transmission should be43state-dependent (stronger when volatility is high and capital constrained),44asymmetric (stronger for negative shocks), and increasing in the degree to45which pricing and liquidity provision are cross-security activities. Our46measures of propagation speed, breadth, asymmetry, and market-state dependence47are designed to test exactly these comparative statics at the minute scale.4849\subsection{Fire sales, flows, and ownership networks}5051A complementary empirical literature documents non-fundamental transmission52through investor portfolios. Mutual funds facing outflows engage in forced53sales concentrated in their holdings, generating price pressure and reversals54\citep{coval2007asset}; predictable flows generate predictable price pressure55\citep{lou2012flow}; and uninformative trading imbalances in one stock spill56over to correlated stocks through risk-averse liquidity providers57\citep{andrade2008trading}. Ownership structure itself predicts fragility:58stocks with concentrated or correlated-flow ownership have higher volatility59and comovement \citep{greenwood2011stock}, and common active ownership60forecasts cross-sectional correlation beyond fundamentals-based similarity61\citep{anton2014connected}. At the institutional level, overlapping portfolios62create fire-sale spillovers that aggregate into measurable systemic63vulnerability \citep{greenwood2015vulnerable,duarte2021fire}. This work64establishes that idiosyncratic demand shocks \emph{do} propagate, but its65resolution is coarse---monthly or quarterly holdings and flows. Whether the66same propagation operates within the hour, and whether its speed has changed as67intermediation became algorithmic, is the margin our data address.6869\subsection{Comovement, commonality, and liquidity dynamics}7071The commonality literature establishes that liquidity has a systematic72component: spreads and depths co-move with market- and industry-wide liquidity73\citep{chordia2000commonality}, common factors in order flow explain much of74the commonality in returns \citep{hasbrouck2001common}, and commonality is75supplied in part by common intermediaries---stocks sharing a NYSE specialist76firm co-move in liquidity beyond the market factor77\citep{coughenour2004common}. Time variation is systematic as well: liquidity78falls after market declines, especially when funding is tight79\citep{hameed2010stock}; specialist inventories and trading losses forecast80spread widening \citep{comertonforde2010time}; and returns to liquidity81provision spike with the VIX as liquidity ``evaporates'' in turmoil82\citep{nagel2012evaporating}. Aggregate liquidity risk is priced83\citep{pastor2003liquidity,acharya2005asset}, and commonality strengthens in84downturns across forty markets \citep{karolyi2012understanding}. Relative to85this literature, we do not measure liquidity \emph{levels}; we measure the86\emph{transmission of extreme events} and use liquidity states---constructed87from OHLCV proxies---as conditioning variables. At very short horizons,88measured comovement is attenuated mechanically by asynchronous trading89\citep{epps1979comovements}, a bias any minute-level design must engineer90around; our breadth measures therefore condition on trading eligibility rather91than assuming synchronous observation.9293\subsection{Lead--lag effects and gradual information diffusion}9495Cross-autocorrelation in returns is among the oldest facts in empirical asset96pricing: large stocks lead small stocks \citep{lo1990contrarian}, predominantly97within industries, consistent with gradual information diffusion98\citep{hou2007industry}. Economic links sharpen the pattern: returns of99principal customers predict suppliers' returns \citep{cohen2008economic}, and100cross-predictability follows supplier--customer industry links broadly,101declining with analyst coverage and institutional ownership102\citep{menzly2010market}. At high frequency, tick-data studies document that103liquid assets lead illiquid ones with intraday-seasonal lead--lag structure104\citep{huth2014high}, and \citet{chinco2019sparse} show that one-minute-ahead105cross-stock predictability exists but is sparse, short-lived, and news-related.106These studies measure \emph{unconditional} predictive structure. Our object is107different: the conditional response of the cross-section to an identified108extreme idiosyncratic event, its horizon profile, and the secular evolution of109both. The lead--lag literature nonetheless supplies one of our falsification110benchmarks: propagation estimated through lagged-return networks should111reproduce known liquidity-ordered asymmetries if it is measuring genuine112transmission.113114\subsection{Networks: real linkages, financial linkages, and measured connectedness}115116Network theory shows when idiosyncratic shocks fail to diversify away:117sufficiently asymmetric input--output structures let micro shocks aggregate118\citep{acemoglu2012network}, and intermediate integration and diversification119maximize cascade susceptibility in financial cross-holdings120\citep{elliott2014financial}; \citet{glasserman2016contagion} survey the121contagion-network literature. Real-economy links propagate firm-level shocks122measurably---natural-disaster shocks to suppliers impose large value losses on123customers \citep{barrot2016input}---and network positions carry asset-pricing124implications \citep{herskovic2018networks}. On the measurement side,125connectedness is estimated from variance decompositions126\citep{diebold2012better,diebold2014network}, Granger-causality graphs among127financial institutions \citep{billio2012econometric}, sparse VARs128\citep{barigozzi2019nets}, and tail-risk networks \citep{hautsch2015financial}.129The contagion-measurement tradition warns that unconditional correlation130comparisons are biased in volatile regimes \citep{forbes2002no}; defining131contagion as comovement beyond fundamentals, \citet{bekaert2014global} find132``wake-up call'' transmission tied to weak fundamentals. Classic133volatility-spillover evidence---meteor showers in FX134\citep{engle1990meteor}, international open/close spillovers135\citep{hamao1990correlations}, and cross-market volatility transmission136\citep{king1990transmission}---anticipates our design at coarser resolution137and for a handful of markets. Across all of this work the networks are138estimated at daily or lower frequency on small panels. We estimate pre-shock139networks on minute data for thousands of securities and use them as lagged140conditioning information---asking whether network position predicts141transmitters, receivers, amplifiers, and absorbers---with permuted-network142placebos in the spirit of the falsification tradition.143144\subsection{Market-structure evolution and the high-frequency trading debate}145146Our sample spans essentially the entire modern transformation of U.S.\ equity147trading, and the literature documents each step. Decimalization in 2001 cut148spreads, particularly for active stocks \citep{bessembinder2003trade};149liquidity itself promotes price efficiency \citep{chordia2008liquidity};150turnover exploded in the 2000s alongside algorithmic trading151\citep{chordia2011recent}; the NYSE's 2006--07 Hybrid automation made prices152more efficient while raising the cost of immediacy153\citep{hendershott2011automation}; and fragmentation across venues did not154measurably harm quality \citep{ohara2011fragmentation}. On algorithmic and155high-frequency trading, the normal-times evidence is broadly positive:156algorithmic trading narrows spreads and improves price discovery157\citep{hendershott2011algorithmic,brogaard2014high}, modern HFT market makers158halve spreads on entry \citep{menkveld2013high}, low-latency activity improves159short-horizon quality \citep{hasbrouck2013low}, machines improve FX efficiency160\citep{chaboud2014rise}, and colocation-based evidence replicates161internationally, with higher short-term volatility as the caveat162\citep{boehmer2021algorithmic}; \citet{menkveld2016economics} surveys the163channel-by-channel evidence. Theory tempers the optimism: speed investment can164be socially excessive \citep{biais2015equilibrium}, fast news trading165concentrates volume and short-run price correlation \citep{foucault2016news},166toxic arbitrage degrades liquidity \citep{foucault2017toxic}, the continuous167limit order book hard-wires a latency arms race168\citep{budish2015high}---measured directly in message data as races won in169microseconds, a fifth of volume \citep{aquilina2022quantifying}---and speed can170crowd out information acquisition \citep{weller2018does,baldauf2020high}, with171HFTs eventually trading \emph{with} informed institutional orders172\citep{vankervel2019high}. \citet{ohara2015high} argues the machine age changes173what microstructure concepts mean. This literature motivates our central174hypothesis---normal-times absorption improving while tail transmission175accelerates---but it measures liquidity levels, price discovery, or single176events. We contribute the missing time series: propagation speed and breadth of177extreme shocks, measured uniformly across three decades of structural change,178with flexible year effects rather than imposed trend breaks.179180\subsection{Fragility and flash events}181182Episodic evidence shows modern liquidity can vanish abruptly. On May 6, 2010,183high-frequency traders did not trigger the crash but amplified selling once it184began \citep{kirilenko2017flash}; cross-market arbitrage broke down, leaving185the seller dependent on local liquidity \citep{menkveld2019flash}; and186fragmentation shaped which ETFs were hit \citep{madhavan2012exchange}. More187systematically, HFTs supply liquidity in single-stock extreme moves but their188provision weakens when \emph{many} stocks move together189\citep{brogaard2018high}---precisely the margin our breadth measure tracks.190Proposed early-warning metrics remain contested: VPIN signaled deteriorating191conditions before the Flash Crash \citep{easley2012flow}, but its power appears192mechanically tied to volatility and trading intensity193\citep{andersen2014vpin}. This debate cautions against fragility indicators194that repackage volatility; our propagation measures are therefore defined195relative to lagged conditional distributions and validated against matched196placebo times.197198\subsection{ETFs, indexing, and passive intermediation}199200A structural change specific to the later sample is the growth of index-linked201intermediation. ETF ownership raises non-fundamental volatility of underlying202stocks through the arbitrage channel \citep{bendavid2018etfs} and contributes203to return comovement, especially in small and illiquid names204\citep{da2018exchange}; index-product arbitrage flipped index return serial205dependence from positive to negative across twenty markets206\citep{baltussen2019indexing}. Effects on efficiency are contested: ETF207ownership is associated with weaker firm-specific information environments208\citep{israeli2017dark}, yet ETF activity speeds the incorporation of209systematic earnings information where information environments are weak210\citep{glosten2021etf}, and Russell-reconstitution evidence suggests price211efficiency survives passive growth because the remaining informed investors212adjust \citep{coles2022index}. For us, the ETF layer is both a candidate213mechanism for the \emph{widening} of propagation (basket arbitrage converts214idiosyncratic shocks into correlated flows) and a required robustness screen:215no headline result may be driven by ETFs themselves.216217\subsection{Jumps, co-jumps, and scheduled announcements}218219Our shock measurement builds directly on high-frequency econometrics. Intraday220volatility periodicity must be filtered before anything else221\citep{andersen1997intraday}; bipower variation separates jumps from diffusion222\citep{barndorffnielsen2004power}; formal jump tests exist at the observation223level \citep{lee2008jumps,aitsahalia2009testing}. Two cautions discipline our224thresholds: at ultra-high frequency much of what coarser methods call jumps is225friction or ``drift bursts'' \citep{christensen2014fact}, and multiple-testing226corrections eliminate a large share of spurious detections, with true jumps227rarer and news-related \citep{bajgrowicz2016jumps}. The co-jump literature is228the nearest empirical neighbor to our propagation object: index-level jumps are229systematic co-jumps linked to macro news while idiosyncratic jumps diversify230away \citep{bollerslev2008risk}; simultaneous multivariate jumps are frequent,231associated with liquidity dry-ups, and priced \citep{caporin2017systemic};232jumps coincide with liquidity-demand shocks and news233\citep{boudt2014intraday}; and simultaneous jumps across countries carry234portfolio consequences \citep{das2004systemic}. Self- and cross-excitation of235jumps has a canonical language in Hawkes processes236\citep{hawkes1971spectra,bacry2015hawkes}, estimated on international indices237by \citet{aitsahalia2015modeling}. Scheduled macro announcements provide the238identified common shocks against which idiosyncratic propagation must be239benchmarked: announcement surprises produce jump-like conditional-mean240responses in FX and global markets \citep{andersen2003micro,andersen2007real},241generate the cojumps documented by \citet{lahaye2011jumps}, and structure242expected returns around announcement times243\citep{savor2013how,lucca2015pre}. Accordingly, our baseline excludes244announcement windows and treats announcement-window propagation as a separate,245labeled object. Where the co-jump literature tests for \emph{simultaneity}, our246measures trace the \emph{sequence}---breadth and intensity at one through 120247minutes---and its secular change.248249\subsection{High-frequency measurement: factor structure, OHLCV proxies, and machine learning}250251Separating idiosyncratic from systematic movement at the minute scale requires252a factor structure: common factors exist in high-frequency returns and order253flow \citep{hasbrouck2001common}, high-frequency PCA is consistent with254low-frequency covariance structure \citep{aitsahalia2019principal}, latent255continuous and jump factors are estimable and economically interpretable256\citep{pelger2020understanding}, and betas decompose into continuous, jump, and257overnight components with distinct pricing \citep{bollerslev2016roughing};258signed semicovariances refine the decomposition \citep{bollerslev2020realized}.259Jump-tail risk itself commands a large, time-varying premium260\citep{bollerslev2011tails}, which is why our propagation outcomes separate261tail responses from ordinary ones.262Because our data are bars rather than quotes, liquidity enters only through263proxies with a long pedigree: spread estimators from return autocovariance264\citep{roll1984simple} and high--low ranges \citep{corwin2012simple},265range-based volatility \citep{parkinson1980extreme,garman1980estimation},266price-impact ratios \citep{amihud2002illiquidity}, and trade-based information267measures \citep{hasbrouck1991measuring}; \citet{goyenko2009liquidity} validate268such proxies against TAQ benchmarks. For estimation we use local projections269\citep{jorda2005estimation} for horizon-specific responses, and the270machine-learning extension follows the out-of-sample discipline of271\citet{gu2020empirical}, informed by evidence that microstructure features272carry universal predictive content \citep{sirignano2019universal} and remain273informative in the machine age \citep{easley2021microstructure}.274275\subsection{Positioning and contribution}276277Each literature above contributes a piece: theory predicts state-dependent,278asymmetric transmission; the flows literature demonstrates non-fundamental279propagation at monthly resolution; the network literature measures280connectedness on small panels at daily frequency; the market-structure281literature documents the transformation of trading but measures liquidity282levels and single events; the co-jump literature tests simultaneity at high283frequency over short samples. To our knowledge---based on a documented search284dated 2026-07-13 (replication package)---no prior study combines (i)285minute-level resolution, (ii) the full U.S.\ equity cross-section including286delisted securities, and (iii) a 1992--2026 sample spanning floor trading,287decimalization, Reg~NMS fragmentation, and ETF-intermediated, algorithmic288markets. The joint analysis of propagation speed, breadth, persistence,289asymmetry, secular evolution, and network conditioning at that resolution and290length---with placebo-timing, permuted-network, and announcement-exclusion291falsifications---is the contribution.292