spb/wp15_uqo

1\section{Data}2\label{sec:data}34% Every statement below is verified in docs/DATA_AUDIT.md (Stage A pilot audit,5% 2026-07-13). Full-sample statements will be added after the Stage C audit.67\subsection{Source and structure}89Our data are one-minute OHLCV bars for U.S.-listed securities originally10sourced from Finnhub.io (\url{https://finnhub.io}), a market data provider,11and aggregated from the provider's monthly archives into 411 monthly Parquet12files covering January 1992 through March 2026 (87.7~GB compressed) with no13missing months. We access the aggregation through a public archive whose exact byte-level14snapshot is pinned and checksum-verified in the replication package, fixing15the data used throughout for reproducibility.16Each row records the open, high, low, and close price (U.S.\ dollars, unadjusted17for corporate actions), share volume, and ticker for one minute of trading in one18security. Timestamps are UTC and denote the start of the bar minute; we verify the19convention empirically by mapping bar times to the Eastern-time trading session in20months on both sides of daylight-saving transitions.2122A bar exists only for minutes with at least one reported trade: there are no23zero-volume rows in the source. Missing minutes are therefore informative about24trading activity, and we treat them explicitly rather than filling a regular grid.2526\subsection{Pilot audit}2728Before downloading the archive we audited three pilot months (June 1992, June292009, December 2025; 58.8~million bars). Trading-day counts match the U.S.\30exchange calendar in all three months, including the December~24, 2025 early31close. No OHLC-consistency violations ($L_{i,t} > \min(O_{i,t},C_{i,t})$,32$H_{i,t} < \max(O_{i,t},C_{i,t})$, or $L_{i,t}>H_{i,t}$), null values, negative33volumes, or zero-volume bars occur in the pilot sample; 14 duplicate34ticker--minute pairs (all in inactive over-the-counter names) appear in December352025. Liquid securities are densely covered throughout: the four most active36tickers in December 2025 have complete 390-bar regular sessions, and large37issuers already average 277--369 bars per day in June 1992.3839Coverage widens over time, from roughly 8,300 symbols per month in 1992 to about4023,100 in late 2025; much of the recent expansion reflects over-the-counter41instruments rather than listed common equity, and our sample filters hold the42universe definition fixed using lagged dollar-volume ranks. Only 22--26\% of June431992 tickers appear in the later pilot months, indicating that delisted44securities are retained in the historical files.4546Figure~\ref{fig:pilot_coverage} summarizes raw data volume and per-day symbol47counts; Figure~\ref{fig:pilot_seasonality} documents the familiar intraday48patterns in absolute returns, volume, and range. Table~\ref{tab:pilot_summary}49reports pilot summary statistics for the 500 highest dollar-volume securities in50each pilot month.5152\begin{figure}[t]53\centering54\includegraphics[width=\textwidth]{figP1_coverage.pdf}55\caption{\textbf{Dataset coverage.} Panel (a): compressed size of each monthly56source file, 1992--2026 (all 411 files of the pinned snapshot). Panel57(b): number of symbols with at least one bar per trading day in the three pilot58months.}59\label{fig:pilot_coverage}60\end{figure}6162\begin{figure}[t]63\centering64\includegraphics[width=\textwidth]{figP2_seasonality.pdf}65\caption{\textbf{Intraday seasonality, pilot months.} Minute-of-day means during66regular trading hours for the top 500 securities by within-month dollar volume:67(a) absolute one-minute log return in basis points; (b) share of the stock-day's68volume; (c) high--low log range in basis points. Descriptive full-month69normalization; no corporate-action screens.}70\label{fig:pilot_seasonality}71\end{figure}7273\input{tables/tabP1_pilot_summary}7475\subsection{Full-sample audit and clean panel}7677The full archive contains \DataRows{} bars. Observed trading dates match the78official XNYS calendar exactly---zero missing sessions and zero dates outside79the calendar over thirty-four years---and the regular-hours share of bars is80flat across calendar months, ruling out any daylight-saving mishandling.81Integrity failures are negligible: \DataOHLCViol{} impossible OHLC bars82(concentrated in 1993--1995) and \DataDupKeys{} duplicate ticker-minute keys83(1.3 per million), which we merge under a logged deterministic rule. The panel84spans \DataTickers{} distinct tickers forming \DataSpells{} continuous85spells; \DataMultiSpellTickers{} tickers have multiple spells, so spells, not86tickers, are the identity unit. Entries and exits occur continuously87(delisted securities are present throughout), with one discrete88source-coverage expansion in 2020 that our fixed-rule universes absorb89(Figure~\ref{fig:coverage}). A conservative overnight-ratio screen detects90\NSplitEvents{} corporate-action events (\NSplitFwd{} splits,91\NSplitRev{} reverse splits); event days $\pm1$ trading day are excluded92from all return-based analyses. Table~\ref{tab:sample_construction} tracks93the sample from raw bars to the regular-hours analysis panel.9495\input{tables/table1_sample_construction}9697\begin{figure}[t]98\centering99\includegraphics[width=\textwidth]{figE1_coverage.pdf}100\caption{\textbf{Coverage of the full panel.} (a) Distinct symbols per year101and the screened (lagged top-3000-eligible) universe; (b) regular-hours bars102per year; (c) median and 10th-percentile bars per stock-day on the lagged103top-1000 universe.}104\label{fig:coverage}105\end{figure}106107\subsection{Known limitations}108109The archive contains no quotes, spreads, depth, or trade direction; all liquidity110statements in this paper refer to OHLCV-based proxies. Securities are identified111by ticker only, prices are unadjusted, and security types are unlabeled; the112corporate-action detection, ticker-spell construction, and universe113classification underlying the final sample are described in114Section~\ref{sec:measurement} and the Internet Appendix. In the pilot months the115standard deviation of unscreened minute returns is an order of magnitude larger116than the 1st--99th percentile spread, reflecting split-like jumps and isolated117bad prints that the cleaning stage must remove.118119% [Full-sample statistics, sample-construction table (Table I), and survivorship120% curves to be added after Stages B–D.]121