spb/wp15_uqo

1\section{Data}2\label{sec:data}34% Every statement below is verified in docs/DATA_AUDIT.md (Stage A pilot5% audit and Stage C full-sample audit, 2026-07-13).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 exact14byte-level snapshot is pinned and checksum-verified in the replication15package, which fixes the data used throughout. The archive carries no explicit16licence; we use it for research with attribution and do not redistribute the17minute data, only the aggregates, event-level estimates, and code. Each row18records the open, high, low, and close price (U.S.\ dollars, unadjusted for19corporate actions), share volume, and ticker for one minute of trading in one20security. Timestamps are UTC and denote the start of the bar minute; we verify21the convention empirically by mapping bar times to the Eastern-time session in22months on both sides of daylight-saving transitions.2324A bar exists only for minutes with at least one reported trade: there are no25zero-volume rows in the source. Missing minutes are therefore informative about26trading activity, and we treat them explicitly rather than filling a regular27grid. Two properties of the source cannot be established from the data and are28carried as limitations. The archive does not document whether bars aggregate29consolidated-tape trades or a single venue; the presence of sparse low-volume30prints after the 13:00 close on early-closing days suggests off-exchange31trades are included, but volumes could not be reconciled against a32consolidated benchmark for more than a few securities. And securities are33identified by ticker alone, without a permanent identifier or security-type34label, so that ETFs, ADRs, preferred shares, and warrants are present and can35be screened only heuristically.3637\subsection{Full-sample audit}3839The archive contains \DataRows{} bars. Observed trading dates match the40official XNYS calendar exactly---zero missing sessions and zero dates outside41the calendar over thirty-four years---and the regular-hours share of bars is42flat across calendar months, which rules out any daylight-saving mishandling43(an offset error would displace roughly 15\% of session rows in March and44November). Early-closing sessions are truncated at the true exchange close.45Integrity failures are negligible: \DataOHLCViol{} impossible OHLC bars46(concentrated in 1993--1995, flagged and excluded) and \DataDupKeys{}47duplicate ticker--minute keys (1.3 per million), which we merge under a logged48deterministic rule. No bar has a non-positive price or a zero or negative49volume. A pilot audit of three months (June 1992, June 2009, December 2025)50conducted before the full download, summarized in the Online Appendix,51established the same facts on 58.8~million bars and verified volume units52against public daily volumes for benchmark securities.5354The panel spans \DataTickers{} distinct tickers forming \DataSpells{}55continuous spells, a spell being a run of trading days without a gap longer56than thirty calendar days; \DataMultiSpellTickers{} tickers have multiple57spells, so spells, not tickers, are the identity unit of every statistic58that follows a security through time. Entries and exits occur continuously,59with exits running between 800 and 1,600 per year: delisted securities are60present throughout, and only 22--26\% of the tickers active in June 199261appear in 2009 or 2025. The data are therefore not a backfilled survivor62universe, although we cannot verify that every delisting is captured to its63final trading day. There is one discrete source-coverage expansion in 202064(about 11,000 entries, largely over-the-counter names) that fixed-rule lagged65universes absorb (Figure~\ref{fig:coverage}). A conservative overnight-ratio66screen (canonical split ratios within $\pm1.5\%$, both-day dollar volume of67at least \$50,000, prior close of at least \$1) detects \NSplitEvents{}68corporate-action events (\NSplitFwd{} splits, \NSplitRev{} reverse splits);69event days $\pm1$ trading day are excluded from all return-based analyses.70Dividends and other distributions are not detected; their overnight effect71never enters an intraday return because returns are computed within the72trading day only. Table~\ref{tab:sample_construction} tracks the sample from73raw bars to the regular-hours analysis panel.7475\input{tables/table1_sample_construction}7677\begin{figure}[t]78\centering79\includegraphics[width=\textwidth]{figE1_coverage.pdf}80\caption{\textbf{Coverage of the full panel.} (a) Distinct symbols per year81and the screened (lagged top-3000-eligible) universe; (b) regular-hours bars82per year; (c) median and 10th-percentile bars per stock-day on the lagged83top-1000 universe.}84\label{fig:coverage}85\end{figure}8687\subsection{Universes}8889Coverage widens from roughly 8,300 symbols per month in 1992 to about 23,10090in late 2025, and much of the recent expansion reflects over-the-counter91instruments rather than listed common equity. All analyses therefore operate92on universes defined by strictly lagged rules: a security is eligible in93month $m$ if in month $m{-}1$ it traded on at least ten days with a median94close of at least \$5 and a median of at least 100 regular-hours bars per95day, and the top-500, top-1000, and top-3000 universes rank eligible96securities by month-$(m{-}1)$ regular-hours dollar volume. Membership never97uses information from month $m$. A ticker-shape heuristic (fifth-letter98suffixes and warrant, unit, and preferred suffixes) flags likely non-common99instruments; it does not identify ETFs, which trade under ordinary tickers.100101\subsection{Known limitations}102103The archive contains no quotes, spreads, depth, or trade direction; all104statements about liquidity in this paper rest on dollar volume and the105high--low range. Security types are unlabeled, prices are unadjusted, venue106scope is undocumented, and the licence is unspecified. Pre-2001 minute107returns are discrete at eighths and sixteenths, and roughly half of the108one-minute returns of top-1000 securities in the mid-1990s are exactly zero109(\ZeroRetNinetySix\% in 1996 against \ZeroRetTwentyFour\% in 2024,110Online Appendix); any second-moment or tail statistic compared across the111decimalization boundary inherits this discreteness. We flag the tick-regime112caveat wherever it binds and anchor every secular comparison on a113within-era placebo contrast.114