1\section{Economic framework and hypotheses}2\label{sec:theory}34\subsection{Anticipation versus revelation}56Let $X_{i,q}$ denote an accounting flow of firm $i$ for fiscal quarter $q$ (operating income in our primary7specification), revealed to the public at the earnings-release date $A_{i,q}$ and structured in the SEC filing of date8$F_{i,q}\ge A_{i,q}$. A forecaster standing at a trading day $c<A_{i,q}$ holds an information set $\Info(c)$ that9contains, at most, prices and volumes up to the close of $c$ and the accounting statements that were \emph{available} at10$c$, i.e., quarters $j$ with $F_{i,j}+1\le c$. The quantity of interest is the conditional distribution of the surprise11\begin{equation}12 S_{i,q} \;=\; \frac{X_{i,q}-\E\!\left[X_{i,q}\mid \Acc(c)\right]}{\TA_{i,j^*(c)}},13 \qquad14 \E\!\left[X_{i,q}\mid \Acc(c)\right] \;=\; X_{i,q-4}+\frac{1}{4}\sum_{k=0}^{3}\bigl(X_{i,q-1-k}-X_{i,q-5-k}\bigr),15 \label{eq:surprise}16\end{equation}17where $\Acc(c)\subset\Info(c)$ is the accounting information alone, $j^*(c)$ is the latest available quarter, and the18expectation is a seasonal random walk with drift \citep{foster1977}, the classic honest benchmark when no point-in-time19consensus exists. The adverse event is $D_{i,q}=\ind{S_{i,q}\le -0.005}$. Semi-strong efficiency \citep{fama1970} implies that prices at $c$ impound all public20information, including $\mathcal{A}(c)$; the ``prices lead earnings'' literature \citep{beaver1980, kothari1992,21collins1994} documents that daily and annual returns indeed forecast accounting changes. None of this is the question22we ask. Our question is whether the \emph{resolution} at which the market is observed carries incremental information,23holding the information set, the universe, the calendar coverage, the horizon and the model capacity fixed.2425\subsection{Why resolution could matter, and why it might not}2627Three mechanisms make fine-grained activity informative beyond daily aggregates. First, in sequential-trade models28\citep{kyle1985, easley1987}, informed traders split orders and trade when liquidity is deep, so the \emph{composition} of29a day's volume and return (intraday autocorrelation, concentration of variance in a few minutes, volume on down-bars) can30reveal private information that the daily net return hides. Second, realized measures built from fine bars separate the31continuous and jump components of variation \citep{barndorff2004, andersen2003} and their sign32\citep{barndorff2010semivariance, patton2015}; negative semivariance and signed jumps have been linked to the33cross-section of expected returns \citep{bollerslev2020good} and may proxy for the arrival of adverse firm-specific news.34Third, limited attention \citep{hirshleifer2003, hirshleifer2009, dellavigna2009} predicts that the incorporation of35public but costly-to-process information is gradual, so that the time path of trading within the day (opening and closing36concentration, overnight versus intraday returns) carries information about ongoing processing.3738Against these channels stand two forces. The Epps effect and microstructure noise contaminate 1-minute returns39\citep{hansen2006, barndorff2008}, so that the finest sampling is not the most informative for variance measurement40\citep{liu2015}; and whatever information intraday activity contains about fundamentals may be fully summarised by the41daily close, in which case the frequency premium is zero. Algorithmic trading may further have reduced the amount of42pre-announcement information acquisition \citep{weller2018}. The sign and size of the frequency premium are therefore43empirical questions, and a precise null is as informative as a positive estimate.4445\subsection{Hypotheses}4647All hypotheses, endpoints and decision rules were frozen before the test period was opened (\texttt{docs/RESEARCH\_DESIGN.md}48in the replication package). Let49\begin{equation}50 \Loss(M)\;=\;-\frac{1}{N}\sum_{(i,q)}\Bigl[D_{i,q}\log \hat p^{M}_{i,q}+(1-D_{i,q})\log\bigl(1-\hat p^{M}_{i,q}\bigr)\Bigr]51 \label{eq:logloss}52\end{equation}53denote the out-of-sample log loss of model $M$ for the binary adverse outcome, and let B1--B5 be the nested information54sets of Table~\ref{tab:infosets}; the paired increment of interest is $\dlog(\mathrm{A},\mathrm{B})=\Loss(\mathrm{A})-\Loss(\mathrm{B})$.5556\begin{description}[leftmargin=1.5em]57\item[H1 (incremental information).] $\Loss(\mathrm{B5})<\Loss(\mathrm{B2})$ at the 20-trading-day horizon:58 one-minute features improve forecasts beyond point-in-time fundamentals and daily OHLCV. Null: no improvement.59\item[H2 (temporal resolution).] The gain of B5 over B2 exceeds the gains of B4 (5-minute) and B3 (30-minute) over B2,60 with the same feature families and the same number of inputs. Null: the gains are equal.61\item[H3 (heterogeneity).] The B5$-$B2 gain is larger for small, illiquid and highly levered firms, the groups in which62 information asymmetry and processing costs are highest. Null: homogeneous gain.63\item[H4 (timing).] The gain increases as the cutoff approaches the release ($h=40\to1$) and is absent at placebo cutoffs.64 Null: no gradient.65\item[H5 (pricing).] The ex-ante deterioration probability predicts the announcement-window abnormal return and66 post-cutoff factor-adjusted returns. Null: no return predictability after risk adjustment and costs.67\end{description}6869A positive H1 does not identify \emph{who} trades on what: observational prediction is not causal evidence of informed70trading, and we do not have quotes, trades or trader identities \citep{gregoire2022, hendershott2015}. The paper's claims71are confined to the forecasting content of anonymous bars and to the frequency at which that content appears.72