\section{Conclusion}
\label{sec:conclusion}

Three decades of minute-level data on the full U.S.\ equity cross-section
allow the title's question to be answered in its three parts. Faster: common
information that in the mid-1990s took tens of minutes to appear in
cross-stock comovement now appears within the minute, whether because
incorporation accelerated or because the mechanical attenuation of
nonsynchronous trading receded, and most likely both. Wider: no. Isolated
idiosyncratic extremes---six-sigma one-minute moves in the largest five
hundred securities---raise market-wide tail activity by at most a few percent
of its base rate within two hours, in every decade and for either sign; what
transmission exists is concentrated among the shocked stock's twenty closest
lagged-correlation neighbors, and that concentrated channel became roughly
three times larger between the floor-trading and algorithmic eras, so
transmission became sharper rather than broader. More fragile: only in a
bounded sense. Within the post-decimalization era, ordinary extreme minutes
became an order of magnitude rarer while minutes in which a tenth of the
universe moves together are roughly ten times more frequent in 2008 and 2020
than in calm years; across the decimalization boundary the question cannot be
answered with threshold-based measures, because price discreteness
manufactures synchronization in the 1990s.

The evidence is deliberately associational where identification runs out,
and the paper reports the diagnostics that qualify its own headline numbers.
Real events arrive in minutes of elevated ambient activity, so the
market-wide response is bracketed rather than pinned at zero; the network
differential is a difference between two non-zero contrasts, and its secular
increase combines a change in transmission with a change in how precisely
one-minute correlations identify economic neighbors; and the co-jump series
conflates common news with endogenous synchronization. OHLCV bars do not
observe quotes, depth, order flow, or security type, so liquidity operates
through proxies and ETFs cannot be isolated; tickers are not permanent
identities, so long-run statements run within screened spells. The
extensions that would most sharpen the conclusions are, in order, neighbor
sets estimated on coarser-sampled or transaction-time correlations, the
pre-event network contrast, lower source and higher receiver thresholds on
wider universes, and the attribution of co-jump minutes to news using
timestamped announcement data.

For market design and risk monitoring the practical implication is direct.
Surveillance aimed at single-name events monitors the part of the system that
modern markets absorb, or transmit only to a handful of linked names; the
binding constraint for market-wide stability is the market-wide minute, and
the margin that moved over three decades is the sharpness of transmission
along statistical linkages, not its breadth.