Methodology: Insider Transactions (Form 4) Chip

What the chip says and what it means

On any stock page, near the ticker price, you may see one of four chips: INSIDERS BUYING, INSIDERS SELLING, INSIDERS FLAT, or INSIDERS — (em-dash). Each label encodes the result of a specific 90-day calculation against Form 4 filings reported to the SEC. This page documents exactly which calculation, what the thresholds are, when the chip is hidden, and when it has changed.

The chip is informational, not advice. Cluster insider buying is one input among many; insider selling can be driven by tax planning, diversification, or compensation plans rather than a negative view. Read the underlying Form 4 transactions on the insider tab before drawing a conclusion.

1. Data source

The chip reads GET /api/insiders?ticker=<T>&period=90d. The backend (routes/routes_analysis_fastapi/options_insiders.py) returns the Form 4 transactions for the ticker — the chip’s counts and totals aggregated over the last 90 calendar days, the transaction list itself newest-first and unbounded (see the note below) — plus a float field carrying the most-recent reported shares-outstanding count from the company’s annual filing. The transactions come straight from SEC EDGAR via the platform’s Form 4 parser; no commercial data vendor sits between EDGAR and the chip.

Form 4 is the SEC’s reporting form for changes in beneficial ownership by company insiders — officers, directors, and ten-percent shareholders. It must be filed within two business days of the transaction. The chip’s 90-day window means it reflects roughly the last quarter of named-insider activity.

Filings the company makes under its own SEC ID. Some Form 4s name the issuer itself as the reporting owner (the filing’s owner field carries the company’s CIK, not a person’s; the footnotes may still name the officer). Since 2026-09-24 those rows are listed in the transaction feed with the label “Filed under the company’s own SEC ID”, but they are excluded from everything that counts insiders: this chip’s buy/sell totals, the net figure, and cluster detection. An unnamed filer cannot be counted as a distinct insider. The test is a comparison of the two CIKs on the filing and never a guess: a row with either CIK missing is treated as a normal insider row.

2. Transaction-code mapping

Bucketing is deliberate. We count only open-market purchases (P) and open-market sales (S) — the transactions where the insider made a directional choice with their own money. We do NOT count A (RSU vests, option grants, other compensation-issued shares — non-discretionary) because counting them painted “INSIDERS BUYING” on routine compensation events. RSU vests are excluded for that reason: receiving a scheduled equity grant is not a market signal. We also do not count D (dispositions to the issuer, including tax withholding), G (gifts), F (tax-withholding), or M (option exercise without sale) for the same reason — none of those carry a directional signal.

The bucketing change shipped 2026-05-22 (BOARD CYCLE DATA_CZAR P1). Before that date the chip counted A as a BUY and D as a SELL; that was wrong — an officer’s RSU vest is not a buy, and a forced disposition to cover taxes is not a sell. The current mapping matches the route’s own summary (insiders.py: net90dayValue sums P/S only) and pg_get_recent_insider_buys (WHERE transaction_code='P').

The code mapping lives in static/js/ticker-renderer.ts in _renderInsiderNetFlowChip (lines 1873–1888).

3. The 1%-of-float threshold

Aggregate share counts cross the 90-day window:

• buy_shares — sum of shares across all BUY-bucketed transactions
• sell_shares — sum of shares across all SELL-bucketed transactions
• net_shares = buy_shares − sell_shares

The chip is then classified as:

• INSIDERS BUYING if net_shares > 0 AND buy_shares ≥ 1% of float
• INSIDERS SELLING if net_shares < 0 AND sell_shares ≥ 1% of float
• INSIDERS FLAT in every other case where float is known

Honesty disclosure on “float”. The “float” metric on this surface uses shares_outstanding (total shares issued, from the most recent annual filing) — not true free float (outstanding minus restricted and insider-held blocks). Oxford Ledge does not currently carry a separate free-float figure; true free float requires SEC Form 144 and Schedule 13D/G data that is not yet integrated. Shares outstanding is always greater than or equal to free float, so using it as the denominator makes the 1% threshold slightly more conservative than a true-float threshold would be (the chip is marginally less likely to fire). This is the safe direction: no false “INSIDERS BUYING”. The wire field is named float for frontend compatibility, but the numeric value is shares_outstanding. This honesty disclosure was ratified in the 2026-05-22 BOARD CYCLE P2 sweep and is documented in the route docstring at routes/routes_analysis_fastapi/options_insiders.py lines 92–99.

The 1%-of-float threshold filters out single-employee transactions that are too small to be directionally meaningful on a company-wide basis. For a 200-million-shares-outstanding company, a single 15,000-share open-market buy by one director clears the threshold; a 500-share tax-withholding does not.

One percent is policy, not a tuning knob. It was chosen because it is high enough to filter routine compensation-plan activity and low enough that real cluster-buying still trips it on mid-cap and large-cap companies. The constant lives at static/js/ticker-renderer.ts as floatShares * 0.01.

4. The em-dash chip (fail-closed)

When the backend cannot determine the company’s float — usually because the most recent annual filing has not been indexed yet, or for a newly-listed ticker where shares-outstanding has not been reported — the chip renders as INSIDERS — (em-dash) rather than FLAT.

This is a deliberate visual distinction. Before 2026-04-27, a missing float silently collapsed the 1% threshold to zero, which painted every single-share net-buy as INSIDERS BUYING. The em-dash chip exists so that “we don’t know the float” reads differently than “we know the float and activity was balanced.” The aria-label and hover title on the em-dash chip carry the explanatory text.

5. Lookback window + caching

The lookback is a rolling 90 calendar days from the moment the page is loaded. Older Form 4 transactions are still readable on the insider tab; they just don’t affect the chip.

The 90-day window is now enforced in SQL (corrected 2026-09-08). This paragraph used to say the opposite, and it was accurate for about three and a half months: the chip’s figures were summed over the most recent 100 transactions with no SQL-side date filter, so a low-volume issuer could have its whole filing history counted into a number captioned “90-day”. It said so here while §2 above claimed the documented window — one page, both claims. The counts, share totals and net value are now aggregated by a windowed query that reads the entire window rather than a capped page, and the enforced number is shipped alongside the figures so this caption cannot drift from the query that produced it again.

One distinction the §2 wording above flattens. The chip’s figures are windowed. The transaction list on the insider tab is deliberately not — it stays newest-first without a date bound, because date-bounding it would empty the tab for exactly the thinly-filed issuers this correction was written for. So: older Form 4 transactions remain readable, and they do not affect the chip.

The chip caches its result for 5 minutes per ticker per browser session. A fresh fetch happens on every new ticker, on every page load after the cache expires, and on every hard refresh. The cache lives in memory only; closing the tab clears it.

6. Click behavior

Clicking the chip scrolls the page to #insider-feed-section — the table of underlying Form 4 transactions that produced the label. Keyboard users can use Enter or Space to trigger the same scroll. The chip never opens a separate page or modal; the full transactions are inline so a reader can audit the classification.

7. What we deliberately do not do

• No sentiment score on top of the chip. The label is mechanical: did transactions clear the threshold or not. We don’t layer a 1-to-10 conviction score because the underlying classification rule is already what we’d be scoring.
• No insider-by-name highlights on the chip itself. Some platforms surface “CEO bought” vs “director bought.” The chip stays aggregate; named-insider detail is one click away in the insider tab.
• No 10-percent shareholder reweighting. A 10-percent shareholder’s transactions count the same as an officer’s in the aggregate. The insider tab shows the role label per row so a reader can weight it themselves.
• No price-level information in the chip. We classify by share count, not dollar volume. A small-share, high-price transaction can clear the dollar bar without clearing the share bar; we use shares because the SEC filing’s ownership concept is share-denominated.
• No chip on ETFs. ETFs do not have insiders in the Form 4 sense; the chip is suppressed at render time for any ticker that the platform classifies as an ETF.

8. Changelog

Source code and references

Backend route: routes/routes_analysis_fastapi/options_insiders.py (api_insiders). Frontend chip renderer: static/js/ticker-renderer.ts (_renderInsiderNetFlowChip + _applyInsiderNetFlowChip). Contract tests: tests/test_routes_insiders_contract.py (pins the float field presence and the error-response shape).

This page mirrors the source files named above and is reviewed on the date shown at the top. How we keep every figure honest — the contract tests, freshness reviews, and public incident log behind the data — is documented in the Trust Dossier.

Corrections: oxfordledge@gmail.com.