Tony Wang17 min readWe Read 1,934 SEC Insider Trades. Not One Was a Real Purchase.
Across the 15 most insider-active US companies, zero of 1,934 Form 4 trades were open-market buys. Plus a busted Chime lockup theory and a fake 'net buyer.'
Financial media treats "insiders are buying" and "insiders are selling" as if they're reading the same kind of signal in opposite directions. They aren't. Academic work on this goes back decades: insider purchases are a real, information-rich signal, because an executive has to spend their own cash to make one. Insider sales are mostly noise — tax bills, diversification, and, increasingly, sales that were scheduled months in advance and never touched a human decision on the day they executed.
So we pulled the raw SEC Form 3/4/5 data for the 15 US operating companies with the most insider transaction activity in the trailing ~90 days and read every single trade's transaction code, not just the net share count everyone quotes. What we found is that the "buying" side of the ledger is close to fictional.
Not one real purchase, in 1,934 trades
Every SEC Form 4 carries a one-letter transaction code. S is an open-market sale. M is an option exercise. C is a conversion between share classes. A is a grant or award. F is tax withholding. G is a gift. And P — the one that actually means "an insider spent their own money to buy shares on the open market" — never showed up.
91.3% of all activity is a plain sale. The rest is mechanics — option exercises, tax withholding, internal share-class conversions — not anyone choosing to buy. If you've ever seen a headline claim "insider buying" at scale, this is worth remembering: the dataset behind it almost certainly nets these mechanical "acquisition" codes against sales, and calls whatever's left over "buying."
Show the full 15-company transaction-code breakdown
| Company | Total | S | M | C | A | G | F | J | P |
|---|---|---|---|---|---|---|---|---|---|
| CrowdStrike (CRWD) | 198 | 198 | — | — | — | — | — | — | 0 |
| United Therapeutics (UTHR) | 178 | 158 | 20 | — | — | — | — | — | 0 |
| CoreWeave (CRWV) | 176 | 150 | 15 | 10 | 1 | — | — | — | 0 |
| AppFolio (APPF) | 167 | 159 | — | — | 8 | — | — | — | 0 |
| Hershey (HSY) | 151 | 148 | — | — | 2 | — | 1 | — | 0 |
| Datadog (DDOG) | 145 | 129 | — | 16 | — | — | — | — | 0 |
| Cloudflare (NET) | 118 | 108 | 2 | 7 | — | — | 1 | — | 0 |
| Chime Financial (CHYM) | 113 | 91 | — | 4 | 6 | 6 | 6 | — | 0 |
| BeOne Medicines (ONC) | 109 | 89 | 20 | — | — | — | — | — | 0 |
| Arista Networks (ANET) | 103 | 102 | 1 | — | — | — | — | — | 0 |
| TransDigm (TDG) | 103 | 96 | 6 | — | — | — | — | 1 | 0 |
| Robinhood (HOOD) | 97 | 94 | — | 3 | — | — | — | — | 0 |
| Airbnb (ABNB) | 94 | 75 | — | 10 | — | 4 | 3 | 2 | 0 |
| Seagate (STX) | 92 | 86 | 6 | — | — | — | — | — | 0 |
| Snowflake (SNOW) | 90 | 83 | 4 | — | — | 3 | — | — | 0 |
Chime's "biggest seller" story isn't about its lockup — it's DST Global
Chime Financial (CHYM) posted the largest net insider selling of any company we pulled: -11,206,405 shares. The obvious explanation is a post-IPO lockup expiring — insiders sell in a wave the moment they're legally allowed to. It's a clean, common story.
It's also wrong here. Chime priced its IPO on June 12, 2025. A standard 180-day lockup would have expired around December 9, 2025. The selling we're looking at is dated May through August 2026 — eight-plus months later.
So who's actually selling? Of Chime's 91 sale transactions, 88 (96.7%) belong to just two SEC filer entities, and both are the same investor. Zinonos Despoina — a signatory of Cardew Services Ltd, which is DST Global Advisors Ltd's corporate trustee vehicle — filed 51 sales totaling 7.91M shares. DST Managers VII Ltd, a separate DST Global fund entity, filed another 37 sales totaling 4.67M shares. We cross-checked both against SEC's full-text search: every one of their Form 3/4 filings lists DST Global Advisors Ltd and its DST Global VI/VII fund vehicles as co-filers.
That's not a lockup wave and it's not a broad loss of employee confidence. It's DST Global — Yuri Milner's venture firm, an early Chime investor — running a multi-month, systematic distribution of its stake, more than a year after the company went public.
Datadog's "net buying" runs backward when you decompose it
Datadog (DDOG) is one of exactly two companies in our sample that shows net insider buying rather than selling: +9,888 shares. On its face, that's a mildly bullish data point from a well-known company's leadership.
Break it down by transaction code and the story reverses. Datadog's 145 transactions are 129 sales and 16 conversions — zero purchases, zero option exercises. The conversions are Class B-to-Class A share reclassifications, filed same-day, immediately ahead of a sale, by the same four insiders: CEO Olivier Pomel, co-founder Alexis Le-Quoc, CFO Amit Agarwal, and one additional insider.
631,938 minus 622,050 is 9,888 — an exact match to the headline figure. The dataset counts a same-day, non-economic Class-B-to-Class-A conversion as an "acquisition" alongside a genuine purchase, so it partially offsets the much larger open-market sale in the net-shares math. Strip the conversion out and what's actually happening is Datadog's CEO, CFO, and co-founders selling 622,050 shares on the open market — the opposite of what "net buying" implies.
TransDigm's "buying," too, is just option mechanics
TransDigm (TDG) is the other net "buyer" in our sample, at +9,415 shares — and it's an even smaller-scale version of the same pattern. Its 103 transactions break down to 96 sales, 6 option exercises, and one non-economic 4/A amendment. The option exercises belong to two insiders — a director who exercises roughly 10,132 shares on a recurring schedule, and a co-COO who exercises 3,900 — both of which get sold down within days across the 96 separate sale transactions. A few thousand shares' worth of routine, small, recurring option grants nets out against much larger daily sales to leave +9,415. It isn't a buy signal by any reading that matters.
One data-quality note worth flagging for anyone building on this feed directly: TransDigm's Form 4 history includes duplicate/amended filings that re-report the same June 18, 2026 transactions under two separate accession numbers. Raw transaction counts from this endpoint can double-count via 4/A amendments; the net-share totals aren't affected, since the duplicates cancel, but a raw count shouldn't be cited as a unique-transaction count without a dedup pass.
The real base rate: selling is a mega-cap phenomenon, not a market-wide one
Everything above is about the most insider-active companies — the names that show up in "top insider sellers" lists because they generate the most Form 4 filings. That's a selection effect. Zoom out to every US operating company with any recent insider activity — 3,894 of them — and the picture is close to the opposite of "insiders are selling."
The 111 companies running ≥50 insider transactions in the window — mega-caps with big executive teams and active 10b5-1 programs — are 67.56% net sellers. That's where the "insiders are dumping stock" headlines come from, and where all six companies profiled above live. But that tier is 2.9% of the population. The other 3,783 companies — 97.1% of everyone with any insider activity — lean net-buying, from 58.0% in the medium tier to 63.7% in the low tier. Overall, 59.98% of the full population is a net buyer.
So does real insider buying exist anywhere? Yes — just not at the top 15.
The 15 companies above are the most insider-active in the dataset — the ones generating the most Form 4 filings, all sitting at the extreme edge of the "high activity" tier (≥50 transactions in 90 days, 111 companies total). That raised an obvious follow-up: is code P actually extinct, or is zero-of-fifteen just an artifact of looking at the 15 most extreme cases?
We went back through the next 20 companies by net insider share accumulation — still ≥50 transactions in the window, still the same high-activity tier, just not in the absolute top 15 by transaction count — and checked each one specifically for genuine code-P purchases dated within the same trailing-90-day window (2026-05-01 through 2026-08-25).
Zero companies with any real purchase
25% have at least one genuine code-P purchase
| Company | P txns | Who bought | Approx. value | Character |
|---|---|---|---|---|
| Liberty Latin America (LILA) | 12 | 4 insiders | ~$8.1M | 10% owner + 3 officers/directors, sustained 2 months |
| MapLight Therapeutics (MPLT) | 6 | 10% owner + 1 director | ~$65.5M | Large strategic stake-building, pre-revenue biotech |
| RENN Fund (RCG) | 56 | 10% owner + 2 others | ~$122K | Mechanical: ~756 shares almost every trading day, 2.5 months |
| Cuentas (CUEN) | 67 | CEO + 1 LLC | under $50K | Sub-$1M-revenue penny stock, sub-$1/share lots |
| Carvana (CVNA) | 2 | 1 director | ~$1.5M | Single modest, genuine director purchase |
The clearest case is Liberty Latin America. Between June 25 and August 18, four different insiders bought real shares with their own money: 10% owner and director John Malone — the closely-watched investor behind Liberty Media and its sister companies — across nine separate transactions in Class A common and Series A preference shares; Executive Chairman Michael Fries and SVP/CLO John Winter, both in late June; and director Charles Bracken in mid-August. Four people, three different roles (10% owner, executive chairman, general counsel, director), no offsetting sale on the same filing, spread across two months. That's about as close as this dataset gets to the textbook "informed insider conviction" signal the academic literature describes.
MapLight Therapeutics is the single largest purchase in the sample by dollar value: 10% owner Catalyst4, Inc. bought roughly 5.6 million shares across five block trades between July 28 and August 14 — around $65.5M at the reported prices — plus a smaller purchase from a director. MapLight is a clinical-stage biotech with no reported annual revenue; a major backer building a position at that scale reads like strategic conviction, not routine trading.
The other three cases complicate the picture rather than reinforce it. RENN Fund's "buying" is institutional holder Horizon Kinetics Asset Management purchasing almost exactly 756 shares on nearly every trading day for 2.5 straight months — a pattern that looks as programmatic as any 10b5-1 sale, just running in the buy direction. Cuentas is a sub-$1M-revenue penny stock whose CEO bought in $100–$5,000 increments, dozens of times, at $0.30–$0.70 a share — genuine skin in the game, but trivial in dollar terms. Carvana's is the cleanest small case: one director, two trades, about $1.5M, no complications.
Zoom out and the pattern holds together with the base-rate finding above: real insider buying isn't extinct, it's just concentrated where the academic literature would predict — smaller, less-followed companies and major shareholders building positions, not the mega-caps that dominate "insider activity" headlines by sheer Form 4 volume. And even where it's real, it isn't uniformly meaningful — a daily 756-share purchase for ten weeks straight is a different kind of "buying" than four different executives independently putting real money in over two months.
What it's for
Insider Form 3/4/5 data is public and free from SEC EDGAR, but the transaction codes are the whole story, and most tools that summarize it collapse those codes into a single net-shares number before you ever see them. That number can mean "the CEO sold half a million shares" or it can mean "someone converted a share class right before lunch." They look identical in a "net buying" headline. They are not identical in what they tell you.
Company-level insider transactions, financials, and filing history are queryable at /datasets/sec over one REST API. If you take one thing from the numbers above: before you cite "insider buying" from any dataset, ask whether it's actually counting code P.
Frequently asked questions
Does 'insider buying' in SEC data mean an executive spent their own money to buy stock?
Often not. Across the 15 most insider-active US operating companies' Form 3/4/5 filings over roughly the last 90 days (1,934 transactions total), zero carried code P, the code for a genuine open-market purchase. The rest were open-market sales (91.3%), option exercises, share-class conversions, grants, tax withholding, and gifts — mechanical codes that a naive 'net shares' figure can net against sales and report as 'buying.'
Was Chime Financial's big insider sell-off caused by its IPO lockup expiring?
No. Chime priced its IPO on June 12, 2025; a standard 180-day lockup would have expired around December 9, 2025. The heaviest insider selling is dated May through August 2026, 8+ months later. 88 of Chime's 91 sale transactions (96.7%) trace to two SEC filer entities that are both DST Global — the venture firm and an early Chime investor — distributing its stake through a corporate trustee vehicle, confirmed via SEC full-text search cross-referencing both filers' Form 3/4 co-filer listings.
Why does Datadog show net insider buying when its CEO and CFO were selling stock?
Datadog's reported +9,888 'net buying' is a share-class-conversion artifact. Its 145 transactions are 129 sales and 16 Class-B-to-Class-A conversions (zero purchases). The same four insiders — CEO, CFO, two co-founders/officers — converted 631,938 shares (counted as an 'acquisition') and sold 622,050 of them on the open market. 631,938 minus 622,050 equals exactly 9,888. The real activity is those insiders net-selling 622,050 shares, the opposite of what the headline number implies.
Are most US companies' insiders net buying or net selling stock right now?
Net buying, once you look past the most active companies. Across all 3,894 US operating companies with any recent SEC insider activity, 59.98% are net buyers. Net selling is concentrated almost entirely in the 111 highest-activity companies (67.56% net sellers) — the mega-caps that generate the most Form 4 filings and dominate 'top insider sellers' headlines. The other 3,783 companies, 97.1% of the population, skew net-buying, from 58.0% in the medium-activity tier to 63.7% in the low-activity tier.
Can you tell from public SEC data whether an insider sale was a pre-scheduled 10b5-1 plan or a discretionary decision?
No — we checked. The SEC's Form 3/4/5 transaction schema (accession number, code, filing/transaction date, owner name/title, shares, price) has no field indicating a Rule 10b5-1 trading plan. Trade patterns that look automated — many same-day sales at steadily rising prices in round share increments, as seen at several mega-caps in this data — are consistent with a scheduled plan but not provable from the transaction data alone; confirming one requires reading each Form 4's own footnote text.
How many SEC filers show any insider trading activity?
15,003 of 976,941 total SEC filer records (1.5%) show any Form 3/4/5 transaction in the trailing 90 days. Of those, 11,104 (74.0%) are individual insider filers who get their own CIK, not companies, and only 3,894 (26.0%) are entity_type=operating companies. Any company-level insider-activity ranking has to filter to entity_type=operating first, or it returns people's names instead of tickers.
Does genuine insider buying (transaction code P) exist anywhere in SEC data?
Yes — just not at the most insider-active mega-caps. Checking the next 20 companies by net insider share accumulation (still the same ≥50-transaction, high-activity tier as the original 15, just not the absolute top 15 by transaction count), 5 of 20 (25%) show at least one genuine code-P open-market purchase within the same trailing-90-day window, versus zero of the original 15.
What does real insider buying look like when it actually happens?
It splits into different patterns. The strongest signal is Liberty Latin America (LILA): four different insiders — 10% owner and director John Malone, Executive Chairman Michael Fries, SVP/CLO John Winter, and a fourth director — independently bought roughly $8.1M in stock across two months, no offsetting sales. MapLight Therapeutics (MPLT) saw the single largest purchase: a 10% owner buying a ~$65.5M stake in a pre-revenue biotech. But not every real purchase is meaningful signal — RENN Fund's (RCG) buying is an institutional holder purchasing almost exactly 756 shares on nearly every trading day for 2.5 months, a pattern as programmatic as any scheduled sale, and Cuentas' (CUEN) is a sub-$1M-revenue penny stock CEO buying in sub-$1/share, sub-$5,000 increments.