How much of “insider trading” data is actual trading?
Study #5 · window 2026-09-01 → 2026-09-30 · every number below regenerates from one script on a free key
Feeds and headlines present everything on a Form 4 as “insider trading”. The form itself is more precise: every transaction leg declares what KIND of event it reports via a one-letter transaction code — P and S are purchases and sales, A is a grant from the issuer, M an option exercise, F shares withheld to cover tax on vesting, G a gift. A CEO buying stock with their own money and a vesting schedule hitting its date are different events that happen to share a form. This page measures the actual mix, for every Form 4 filed in September 2026.
Method. All 10,826 ownership filings with filed_date in September 2026 via GET /v1/insider-trades, fetched one filed-date at a time (a fixed since=/until= window returns the same population forever). Original Form 4s only: 214 4/A amendments are excluded as re-filings, and 376 Forms 3/5 are excluded by name — their dataset coverage is live-forward since 2026-09-23 (mid-window), and Form 3 declares holdings, not transactions. 102 holdings-only Form 4s carry no legs and are counted here, not dropped. The unit below is the transaction leg — one row of the filing's transaction tables; 10,134 filings contribute 23,846 legs. As-filed dollar value = shares × price_per_share wherever a positive price is filed (14,412 legs); grants and exercises usually file no price or a non-market price, and this page never pretends otherwise.
What a month of Form 4s actually contains
All 23,846 legs by their own declared code. Codes with fewer than 25 legs are folded into other; nothing is dropped.
| code | meaning | legs | share | filings | |
|---|---|---|---|---|---|
S | Open-market or private sale | 7,829 | 32.8% | 3,096 | |
A | Grant / award from issuer | 4,782 | 20.1% | 3,712 | |
M | Exercise or conversion of derivative | 4,744 | 19.9% | 1,576 | |
F | Shares withheld for tax on vesting/exercise | 2,241 | 9.4% | 1,732 | |
P | Open-market or private purchase | 1,972 | 8.3% | 1,127 | |
D | Disposition back to the issuer | 669 | 2.8% | 219 | |
J | Other acquisition or disposition (see footnotes) | 640 | 2.7% | 369 | |
C | Conversion of derivative security | 476 | 2.0% | 151 | |
G | Bona fide gift | 384 | 1.6% | 263 | |
X | Exercise of in-the-money derivative | 56 | 0.2% | 23 | |
E | Expiration of short derivative position | 28 | 0.1% | 3 | |
other | All codes with fewer than 25 legs in the window (I, L, W) | 25 | 0.1% | 17 |
Direction is near-deterministic for most codes — 4,778 of 4,782 grant (A) legs are acquisitions, 2,239 of 2,241 tax-withholding (F) legs are dispositions — because the code already names the event. Exercises (M) split almost evenly (2,323 acquired / 2,421 disposed) for a structural reason: one exercise files paired legs — the option disposed on the derivative table and the shares acquired on the non-derivative table. 5,711 of 23,846 legs (23.9%) sit on the derivative table.
The same mix in as-filed dollars
14,412 of 23,846 legs (60.4%) carry both a positive as-filed price and a parseable share count; their shares × price sums to $48,925,940,045. Read the non-market codes with care: a grant's filed price (when there is one) is a plan or par-style price, an exercise's is the strike — these are what was filed, not market flow. The share of legs priced is itself part of the composition story.
| code | valued legs | of its legs | as-filed $ total | median leg |
|---|---|---|---|---|
S | 7,794 | 100% | $14,794,879,074 | $246,672 |
A | 1,065 | 22% | $19,613,438,884 | $11,500 |
M | 850 | 18% | $392,877,923 | $117,972 |
F | 2,204 | 98% | $7,627,349,010 | $73,023 |
P | 1,939 | 98% | $3,040,035,110 | $39,311 |
D | 203 | 30% | $2,472,984,534 | $503,625 |
J | 227 | 35% | $791,125,966 | $7,941 |
C | 74 | 16% | $77,386,630 | $209,607 |
G | 9 | 2% | $60,002,370 | $72,226 |
X | 25 | 45% | $20,209,240 | $10,667 |
E | 0 | 0% | $0 | — |
other | 22 | 88% | $35,651,303 | $1,518 |
The largest dollar line is not the open market: grants (A) sum to $19.6B from just 1,065 priced legs (22% of A legs carry a price at all), with a median A leg of $11,500 — a heavily skewed distribution where a handful of large priced grants carry the total. That is exactly why this page reports medians and priced-leg shares next to every sum.
The open-market slice: purchases vs sales
P and S legs together are 9,801 legs (41.1% of all legs; 53.0% of non-derivative legs) carrying $17,834,914,184 as filed. The two sides are far from symmetric:
| legs | filings | as-filed $ total | median leg | |
|---|---|---|---|---|
| Purchases (P) | 1,972 | 1,127 | $3,040,035,110 | $39,311 |
| Sales (S) | 7,829 | 3,096 | $14,794,879,074 | $246,672 |
Sales outnumber purchases 4.0:1 by legs and 4.9:1 by dollars, and the median sale leg ($246,672) is 6.3x the median purchase leg ($39,311). None of this is a signal: most insider compensation arrives as stock, so insiders are structural net sellers. It is the baseline shape of the data — the thing to know before reading any single month's “insider selling” as news.
Is the population complete?
A composition study over an incomplete population is quietly wrong, so the script cross-checks every day's Form 4/4A count against SEC EDGAR's own daily master indexes (distinct accessions bucketed by each row's own Date Filed column across every index in a buffered sweep, the same mechanics as studies #1–#4).
Form 4 side (forms 4 and 4/A): 21 of 21 index days match exactly; 9 days had no published index (weekends/holidays) and zero filings in the API — consistent. No mismatches to chase this time — zero repairs needed, the second study in a row.
Doing this with the API
Everything above is one list endpoint and a local loop: GET /v1/insider-trades with a fixed since=/until= window, then read each object's transactions[] — every leg carries transaction_code, acquired_disposed, is_derivative, shares and price_per_share exactly as filed. The free screener serves the P/S transaction-leg view directly; the insider trading API page walks the filters and query patterns covers windows and paging.
Reproduce this
The whole study is one standard-library Python script that runs on the free tier (the default one-month window is ~120 requests; at the free tier's 10 req/min expect about fifteen minutes). The --verify-coverage flag hits EDGAR directly and needs a contact identity in the User-Agent, per SEC fair-access policy — set your own.
# free key, no card: https://filingpulse.io/signup.html
export FILINGPULSE_API_KEY=fp_...
export EDGAR_CONTACT=you@example.com
curl -O https://filingpulse.io/examples/form4_transaction_codes_study.py
python form4_transaction_codes_study.py \
--since 2026-09-01 --until 2026-09-30 \
--out results.json --verify-coverage
The committed results this page renders from: form4_transaction_codes_2026-09.json. The window bounds are fixed, so the population never changes under you — that is what makes the study reproducible rather than a screenshot.
Limitations
- “Open market” is shorthand. The SEC defines P as open market or private purchase and S likewise for sales; the codes do not split the two, so neither can this page.
- As-filed dollars are not market flow. Sums count
shares × priceexactly as filed. Non-market prices (strikes, plan prices) are included where filed, unpriced legs contribute nothing, and a single large priced grant can dominate a code's total — the medians and priced-leg shares alongside every sum are the honest context. - Plan trades cannot be split out for this window. Form 4 carries a Rule 10b5-1(c) checkbox, but the API's
aff10b5_onefield is live-forward since 2026-10-11, after these filings were ingested —nullmeans not recorded, never “not a plan trade”. Scheduled plan sales are mixed into the S row. - A leg is a table row, not an order: one market order can file as several same-day legs at different prices, and one filing can report several insiders. Legs are the unit the form itself uses; filing counts are given alongside.
- Codes J (other, 640 legs this window) and the folded tail are explained in each filing's footnotes, which this page counts but does not parse.
- One month is one month. Grant and vesting calendars give the mix a seasonal rhythm (many issuers grant on fixed dates), so treat these shares as a measurement of this window, not a law of nature.
Sources
- SEC Form 4 and its General Instructions — the transaction-code table the codes above come from
- EDGAR daily index — the per-day ground truth the coverage check reads
- Study #1: How late is Form 4 data? · Study #2: When insider filings get corrected · Study #3: Which corporate events get disclosed on Fridays? · Study #4: Do insiders trade right after earnings?
- FilingPulse API docs · insider trading API · query patterns · free screener