SEC Filing Dates Can Put After-Hours Data Into the Wrong Trading Session

A filing dated Tuesday may have arrived after the market closed. A backtest that trades at Tuesday's close based on that filing has used information before its assumed trade. The filed date in SEC company facts is useful for screening observations, but a date alone cannot establish intraday availability. Consider a hypothetical company filing accepted at 4:05 p.m. New York time on a Tuesday when the regular session closed at 4 p.m. A strategy that executes at the next regular-session open can first use it on Wednesday, provided Wednesday is a trading day. An acceptance at 3:55 p.m. is not automatically safe for a 4 p.m. closing auction order either: your pipeline needs time to receive, parse, validate, and submit the order before the broker's applicable cutoff. Keep two clocks Store the reporting period end separately from the time when the information became usable. The first answers what period a fact describes. The second controls what the strategy could know. Also keep

SEC Filing Dates Can Put After-Hours Data Into the Wrong Trading Session

A filing dated Tuesday may have arrived after the market closed. A backtest that trades at Tuesday's close based on that filing has used information before its assumed trade. The filed date in SEC company facts is useful for screening observations, but a date alone cannot establish intraday availability.

Consider a hypothetical company filing accepted at 4:05 p.m. New York time on a Tuesday when the regular session closed at 4 p.m. A strategy that executes at the next regular-session open can first use it on Wednesday, provided Wednesday is a trading day. An acceptance at 3:55 p.m. is not automatically safe for a 4 p.m. closing auction order either: your pipeline needs time to receive, parse, validate, and submit the order before the broker's applicable cutoff.

Keep two clocks

Store the reporting period end separately from the time when the information became usable. The first answers what period a fact describes. The second controls what the strategy could know. Also keep the accession number, so a selected fact can be traced back to its filing.

For a filing-level timestamp, SEC submissions data includes an acceptanceDateTime field in its recent filings arrays. Join the fact's accn to the corresponding submissions accession. Older filing history can be listed in additional submissions files referenced by the main response; do not assume the recent array contains every accession. The SEC documents these endpoints in its EDGAR API guide.

An acceptance timestamp is a useful input, not a measured feed delivery time. If you have your own archived receipt timestamps, use the later of acceptance and receipt, then include processing latency. A production simulation should also account for the actual source and any information released earlier through another channel.

Map availability to an executable session

The following example deliberately accepts a calendar of session opens. It uses timezone-aware datetimes and a processing delay; the calendar provider must supply actual exchange sessions, including holidays and daylight-saving changes. These sample dates and values are fictional.

from datetime import datetime, timedelta
from zoneinfo import ZoneInfo

NY = ZoneInfo("America/New_York")

def first_eligible_open(accepted, opens, processing_delay):
    if accepted.tzinfo is None or accepted.utcoffset() is None:
        raise ValueError("accepted must be timezone-aware")
    ready = accepted + processing_delay
    for session_open in sorted(opens):
        if session_open.tzinfo is None or session_open.utcoffset() is None:
            raise ValueError("session opens must be timezone-aware")
        if session_open >= ready:
            return session_open
    return None

opens = [
    datetime(2025, 3, 11, 9, 30, tzinfo=NY),
    datetime(2025, 3, 12, 9, 30, tzinfo=NY),
]
accepted = datetime(2025, 3, 11, 16, 5, tzinfo=NY)
assert first_eligible_open(accepted, opens, timedelta(minutes=5)) == opens[1]
assert first_eligible_open(opens[0], opens, timedelta(minutes=5)) == opens[1]

The second assertion captures a boundary that often disappears in date-only joins: a filing accepted exactly at the open cannot be used at that same open if processing takes five minutes. If the calendar ends before the next eligible session, the function returns None; extend the calendar instead of silently assigning the last available date.

This rule models orders placed for the next eligible open. A strategy trading intraday needs a different execution model, including order cutoffs and realistic fill assumptions. For a daily close strategy, use a proper session calendar and its early-close times, then define when the order must be ready. A hard-coded 4 p.m. close fails on shortened sessions.

Preserve the selection policy

A point-in-time dataset can retain period_end, accession, filed_date, accepted_at, received_at, ready_at, and first_eligible_session. Store timestamps with offsets or in UTC; keep the exchange calendar's local timezone for session calculations. Record the processing-delay assumption alongside the backtest version.

A later filing may repeat or revise an earlier period. Select only observations whose availability falls before the simulated decision time, and retain the selected accession with the result. If the timestamp or the join is missing, mark availability unknown and exclude the observation from that point-in-time run until the source is resolved. Falling back to midnight on the filed date creates a precise-looking timestamp that the source did not provide.

For the companion problem of repeated company facts and filing selection, see SEC Company Facts Duplicates Can Leak Future Filings into Your Backtest.

Share