Insights
What a Corporate Filing-Activity Spike Actually Measures
Alphanume Team · July 30, 2026
A corporate filing-activity spike measures an unusual concentration of SEC submissions, not the materiality or direction of the information inside them.
A filing-activity spike is a count anomaly. Alphanume's SEC Filing Intensity dataset provides the daily number of SEC filings associated with each ticker. That count can surface issuer-days worth investigating, but it cannot say whether the documents are favorable, adverse, routine, duplicative, or tied to one economic event.
The Filing Intensity documentation defines ticker and date filters, count thresholds, pagination, and nightly refresh timing. A reliable workflow uses the count as a trigger, then reviews the filing types and primary documents before assigning a narrative.
Distinguish a high count from an unusual count
A raw count of five can be exceptional for one issuer and routine for another. Define a baseline from prior observations for the same ticker, or rank all covered tickers on the same date. Use only data available before the issuer-day being scored when estimating a historical baseline.
Measure | Definition | Failure mode |
|---|---|---|
raw_count | Served daily filing_count | Favors naturally frequent filers |
change | Count minus trailing mean | Depends on baseline window |
ratio | Count divided by trailing mean | Unstable near zero |
issuer percentile | Rank within ticker history | Requires enough prior observations |
daily percentile | Rank across tickers that day | Sensitive to eligible universe |
Publish the window, lag, minimum-history rule, and treatment of zero variance. A standardized score without those choices is not reproducible. If zero-count rows are served, retain them in the baseline rather than calculating only across active days.
Retrieve high-activity issuer-days
The API supports exact and range filters for filing_count. An exact-count filter cannot be combined with count ranges. For a market-wide screen, combine a completed observation date with a prespecified lower bound. For historical estimation, collect full pages using both cursor fields.
GET /v1/filing-intensity
date=2026-02-06
filing_count_gte=5
For historical baseline:
GET /v1/filing-intensity?ticker=<ticker>&date_lt=2026-02-06
spike = current_count - mean(prior_count_window)
first_eligible_trade = session after nightly dataset refresh
retain ticker, date, raw count, baseline, spikeThe documentation states that the dataset refreshes nightly after the trading day. A completed daily count cannot be treated as an intraday signal earlier that same day. Map the refreshed observation to the next eligible session under a written availability rule.
Inspect the administrative cluster
Several filings can represent one administrative sequence: a filing plus exhibits, related ownership forms, amendments, or submissions from multiple insiders. A single highly material 8-K can produce a lower count than a routine cluster. Retrieve form types and documents from an SEC source before interpreting the spike.
- Count activity first and classify content in a separate table.
- Group related submissions when the research question is economic events.
- Keep form type, accession, acceptance time, and filing URL for review.
- Flag earnings, financing, and corporate-action dates as confounders.
The count dataset does not include filing form or textual content. Any form-level explanation is an external join and should retain its own provenance. This separation prevents a convenient activity feature from being mistaken for document understanding.
An issuer-day review can distinguish at least three mechanisms: many independent forms, several filings tied to one transaction, and amendments or exhibits that expand an earlier disclosure. Those categories require accession-level evidence outside the intensity feed. Store the classification beside the raw count, never in place of it. That preserves the original measurement when analysts disagree about whether a cluster represents one economic event or several.
Avoid direction and materiality claims
The main failure mode is calling a high count bullish or bearish without inspecting the filings. Another is selecting a threshold after seeing returns. Define candidate thresholds, windows, and outcome horizons before testing, then report all prespecified variants with event counts.
Ticker changes and coverage differences can distort issuer histories. Use dated identifier mappings for long samples and keep missing observations distinct from confirmed zeros. A missing row should not be inserted as zero unless the data contract supports that interpretation.
Outcome design matters as much as feature design. A spike can coincide with a large absolute move without predicting its sign. Test signed return, absolute return, volume, and volatility as different outcomes rather than describing them all as reaction. Use non-overlapping or clustered inference when the same issuer produces several adjacent spike days, and show issuer-level contributions so a serial filer does not dominate the result.
For deployment, log the dataset refresh time, response retrieval time, and intended order time. If the nightly process is late or a filing arrives after its cutoff, the signal belongs to the next update. A historical backtest that assumes perfect nightly availability should state that operational assumption and test the effect of an additional-session lag.
Audit the top issuer-days
As a concrete next action, request one completed date with a fixed raw-count floor and select twenty returned issuer-days for document review. Collect every underlying form, accession number, acceptance time, and exhibit relationship, then classify each row as independent filings, one transaction cluster, or amendments and exhibits. Export the raw count beside that evidence map and the first eligible trade session. This audit explains what a high count contained without duplicating the API z-score screen.