\section{Data}
\label{sec:data}

% Every statement below is verified in docs/DATA_AUDIT.md (Stage A pilot audit,
% 2026-07-13). Full-sample statements will be added after the Stage C audit.

\subsection{Source and structure}

Our data are one-minute OHLCV bars for U.S.-listed securities originally
sourced from Finnhub.io (\url{https://finnhub.io}), a market data provider,
and aggregated from the provider's monthly archives into 411 monthly Parquet
files covering January 1992 through March 2026 (87.7~GB compressed) with no
missing months. We access the aggregation through a public archive whose exact byte-level
snapshot is pinned and checksum-verified in the replication package, fixing
the data used throughout for reproducibility.
Each row records the open, high, low, and close price (U.S.\ dollars, unadjusted
for corporate actions), share volume, and ticker for one minute of trading in one
security. Timestamps are UTC and denote the start of the bar minute; we verify the
convention empirically by mapping bar times to the Eastern-time trading session in
months on both sides of daylight-saving transitions.

A bar exists only for minutes with at least one reported trade: there are no
zero-volume rows in the source. Missing minutes are therefore informative about
trading activity, and we treat them explicitly rather than filling a regular grid.

\subsection{Pilot audit}

Before downloading the archive we audited three pilot months (June 1992, June
2009, December 2025; 58.8~million bars). Trading-day counts match the U.S.\
exchange calendar in all three months, including the December~24, 2025 early
close. No OHLC-consistency violations ($L_{i,t} > \min(O_{i,t},C_{i,t})$,
$H_{i,t} < \max(O_{i,t},C_{i,t})$, or $L_{i,t}>H_{i,t}$), null values, negative
volumes, or zero-volume bars occur in the pilot sample; 14 duplicate
ticker--minute pairs (all in inactive over-the-counter names) appear in December
2025. Liquid securities are densely covered throughout: the four most active
tickers in December 2025 have complete 390-bar regular sessions, and large
issuers already average 277--369 bars per day in June 1992.

Coverage widens over time, from roughly 8,300 symbols per month in 1992 to about
23,100 in late 2025; much of the recent expansion reflects over-the-counter
instruments rather than listed common equity, and our sample filters hold the
universe definition fixed using lagged dollar-volume ranks. Only 22--26\% of June
1992 tickers appear in the later pilot months, indicating that delisted
securities are retained in the historical files.

Figure~\ref{fig:pilot_coverage} summarizes raw data volume and per-day symbol
counts; Figure~\ref{fig:pilot_seasonality} documents the familiar intraday
patterns in absolute returns, volume, and range. Table~\ref{tab:pilot_summary}
reports pilot summary statistics for the 500 highest dollar-volume securities in
each pilot month.

\begin{figure}[t]
\centering
\includegraphics[width=\textwidth]{figP1_coverage.pdf}
\caption{\textbf{Dataset coverage.} Panel (a): compressed size of each monthly
source file, 1992--2026 (all 411 files of the pinned snapshot). Panel
(b): number of symbols with at least one bar per trading day in the three pilot
months.}
\label{fig:pilot_coverage}
\end{figure}

\begin{figure}[t]
\centering
\includegraphics[width=\textwidth]{figP2_seasonality.pdf}
\caption{\textbf{Intraday seasonality, pilot months.} Minute-of-day means during
regular trading hours for the top 500 securities by within-month dollar volume:
(a) absolute one-minute log return in basis points; (b) share of the stock-day's
volume; (c) high--low log range in basis points. Descriptive full-month
normalization; no corporate-action screens.}
\label{fig:pilot_seasonality}
\end{figure}

\input{tables/tabP1_pilot_summary}

\subsection{Full-sample audit and clean panel}

The full archive contains \DataRows{} bars. Observed trading dates match the
official XNYS calendar exactly---zero missing sessions and zero dates outside
the calendar over thirty-four years---and the regular-hours share of bars is
flat across calendar months, ruling out any daylight-saving mishandling.
Integrity failures are negligible: \DataOHLCViol{} impossible OHLC bars
(concentrated in 1993--1995) and \DataDupKeys{} duplicate ticker-minute keys
(1.3 per million), which we merge under a logged deterministic rule. The panel
spans \DataTickers{} distinct tickers forming \DataSpells{} continuous
spells; \DataMultiSpellTickers{} tickers have multiple spells, so spells, not
tickers, are the identity unit. Entries and exits occur continuously
(delisted securities are present throughout), with one discrete
source-coverage expansion in 2020 that our fixed-rule universes absorb
(Figure~\ref{fig:coverage}). A conservative overnight-ratio screen detects
\NSplitEvents{} corporate-action events (\NSplitFwd{} splits,
\NSplitRev{} reverse splits); event days $\pm1$ trading day are excluded
from all return-based analyses. Table~\ref{tab:sample_construction} tracks
the sample from raw bars to the regular-hours analysis panel.

\input{tables/table1_sample_construction}

\begin{figure}[t]
\centering
\includegraphics[width=\textwidth]{figE1_coverage.pdf}
\caption{\textbf{Coverage of the full panel.} (a) Distinct symbols per year
and the screened (lagged top-3000-eligible) universe; (b) regular-hours bars
per year; (c) median and 10th-percentile bars per stock-day on the lagged
top-1000 universe.}
\label{fig:coverage}
\end{figure}

\subsection{Known limitations}

The archive contains no quotes, spreads, depth, or trade direction; all liquidity
statements in this paper refer to OHLCV-based proxies. Securities are identified
by ticker only, prices are unadjusted, and security types are unlabeled; the
corporate-action detection, ticker-spell construction, and universe
classification underlying the final sample are described in
Section~\ref{sec:measurement} and the Internet Appendix. In the pilot months the
standard deviation of unscreened minute returns is an order of magnitude larger
than the 1st--99th percentile spread, reflecting split-like jumps and isolated
bad prints that the cleaning stage must remove.

% [Full-sample statistics, sample-construction table (Table I), and survivorship
% curves to be added after Stages B–D.]