Before Counting Insider Buyers, Read the Transaction Code on Every Form 4

Two rows in an insider-transaction screener can carry the same one-word summary — bought — and describe events with almost nothing in common. One is a director moving personal money into an open-market order at whatever price the book showed that morning. The other is the same director receiving shares on a vesting schedule fixed eighteen months earlier, with no order entered, no price chosen, and no cash leaving anyone's account. The screener prints the same word twice.

That gap is why the single-buy-versus-cluster question is usually asked in the wrong order. Counting buyers first and inspecting filings second produces a number that cannot be interpreted, because a count only means something when every row inside it is the same kind of event. Five names on a five-row list is not a cluster if four of those rows are equity awards dated to a compensation calendar. The work that separates a genuine cluster from a single purchase happens in the transaction code column, and it happens before any counting.

The letter in the code column outranks the share count

Section 16 of the Securities Exchange Act determines who files. It reaches directors, officers, and beneficial owners of more than ten percent of a registered class of equity securities. The word officer is narrower than a company's org chart implies. Rule 16a-1(f) defines it as the issuer's president, principal financial officer, principal accounting officer or controller, "any vice-president of the issuer in charge of a principal business unit, division or function (such as sales, administration or finance)," and any other officer who performs a policy-making function. Plenty of people carrying a vice-president title never appear in these filings at all. The absence of a name is not evidence about that person's opinion of the company; it is usually evidence about their job description.

What each filer reports is a coded row. The general instructions to Form 4 carry a table of twenty transaction codes, and that table is the entire interpretive apparatus. A handful of the codes carry the meaning most readers assume the word "buy" carries. Most do not.

P	"Open market or private purchase of non-derivative or derivative security"
S	"Open market or private sale of non-derivative or derivative security"
A	"Grant, award or other acquisition pursuant to Rule 16b-3(d)"
M	"Exercise or conversion of derivative security exempted pursuant to Rule 16b-3"
F	"Payment of exercise price or tax liability by delivering or withholding securities"
G	"Bona fide gift"
C	"Conversion of derivative security"
D	"Disposition to the issuer of issuer equity securities pursuant to Rule 16b-3(e)"
V	"Transaction voluntarily reported earlier than required"
J	"Other acquisition or disposition (describe transaction)"

Only P puts two things in the same row: a decision the insider made freely, and a price the market set on the day. A M row records an acquisition, but the consideration is the strike price fixed when the derivative was awarded, which may bear no relationship to the quote on the exercise date. An A row records an acquisition in which nothing was decided by the recipient at all. An F row is a tax mechanic attached to vesting, and reading it as a sale decision double-counts a single compensation event. A J row is an instruction to go read the footnote, because the form itself has given up on classifying what happened.

What a group of filings has to clear before it is a cluster

Once the codes are read, the count can be attempted. A cluster is a claim that several people independently reached a similar conclusion at a similar time. Four conditions have to hold before a group of filings supports that claim, and each one removes candidates.

• One code, not a mixture. The count runs on P rows. Mixing an A row into the tally does not weaken the signal — it changes what is being measured.
• Separate decisions, not one corporate event. A grant date produces many A rows across many filers on the same day. A vesting date produces many F rows the same way. Those are one decision by a committee, replicated across a roster. They look like consensus and are the opposite of it.
• Size relative to the filer's existing position. Form 4 reports securities beneficially owned following the reported transaction. A purchase that moves that figure by a rounding error is a different observation from one that moves it materially, even when the dollar amounts look similar across two filers with very different existing stakes.
• A window chosen before looking. Widening a window until enough filings fall inside it manufactures clusters at will. The window has to be fixed in advance, and the same window has to be applied to periods that produced nothing.

Section 16(b) adds a structural reason to treat P rows as slow signals rather than fast ones. The statute allows the issuer to recover "any profit realized by him from any purchase and sale, or any sale and purchase" of the issuer's equity securities "within any period of less than six months." A covered person who buys in the open market and sells inside six months hands the profit back. Whatever else a P row is, it is rarely the opening leg of a short-horizon trade, because the person filing it cannot close that trade quickly without giving up the gain.

The plan checkbox changes what the trade date means

The face of Form 4 carries a checkbox: "Check this box to indicate that a transaction was made pursuant to a contract, instruction or written plan that is intended to satisfy the affirmative defense conditions of Rule 10b5-1(c)." Amendments effective 27 February 2023 tightened the conditions behind that box and required the checkbox and adoption-date disclosure on Forms 4 and 5. The same amendments moved reporting of bona fide gifts onto Form 4 rather than Form 5.

The conditions matter for reading dates. Under Rule 10b5-1(c)(1)(ii)(B), a director or officer relying on the affirmative defense must wait through a cooling-off period running to the later of ninety days after adoption of the plan, or two business days following disclosure of the issuer's financial results in a Form 10-Q or Form 10-K — subject to a maximum of 120 days after adoption. For persons other than the issuer and other than directors and officers, the cooling-off period is thirty days. Directors and officers must include a representation certifying that on the adoption date they are not aware of material nonpublic information and are adopting the plan in good faith. Overlapping plans are restricted, and the affirmative defense for a single-trade plan is limited to one such plan in any twelve-month period for everyone other than the issuer.

The consequence for anyone reading dates off a filing is direct. A transaction with that box checked was set in motion at least thirty days earlier, and for a director or officer at least ninety days earlier in most cases. The trade date on the row is an execution date produced by an instruction written a quarter earlier. Whatever the insider knew or believed in the week the trade printed, the plan could not legally have been adopted in response to it.

Plan-flagged activity is not a corner case. Counting Form 4 filings whose text carries the phrase "10b5-1" across the last twelve months, every single month falls between 1,246 and 1,771 filings.

That series measures one specific thing and should not be stretched further. It counts filings whose document text contains the phrase, which is mostly footnote language, and it says nothing about how many of those transactions were purchases rather than sales. The structured checkbox itself is a form field, not free text, so the true count of plan-flagged transactions sits somewhere above these numbers. What the series does establish is that plan-driven flow is a standing feature of the filing stream in every month, not an occasional distortion to be ignored.

The filing date is not the trade date, and the gap is not constant

Section 16(a)(2)(C) of the Exchange Act requires the statement of changed beneficial ownership to be filed "before the end of the second business day following the day on which the subject transaction has been executed," and Rule 16a-3(g)(1) repeats that deadline. That is the version most descriptions stop at.

Rule 16a-3(g) then carves out the cases that matter most here. Where the transaction was made under a Rule 10b5-1(c) contract, instruction or plan, or is a discretionary transaction, and the reporting person did not select the date of execution, the deadline runs from the date the executing broker, dealer or plan administrator notifies the reporting person. If that notification arrives later than the third business day following the trade date, the execution date is deemed to be the third business day following the trade date. A plan trade can therefore surface on EDGAR up to three business days behind a self-directed trade executed the same morning.

Rule 16a-3(f) sets a third clock. Form 5 is due within 45 days after the issuer's fiscal year end and covers holdings and transactions not previously reported on Forms 3, 4 or 5, with no Form 5 required where everything reportable has already been disclosed before the due date. A row that appears on a Form 5 is by construction stale. Treating a batch of Form 5 rows filed in the same week as contemporaneous activity mistakes an administrative deadline for a trading decision.

Form 144 is a different document making a different statement

Form 144 gets folded into insider-activity feeds and does not belong there without a label. Rule 144(h) requires a notice of proposed sale when the amount to be sold in reliance on the rule during any period of three months exceeds 5,000 shares or units, or has an aggregate sale price above $50,000. That is a statement of intent filed ahead of selling, not a record of a completed transaction. The sale can come in smaller, come in later, or not happen.

The volume limitation in Rule 144(e) explains part of why the two documents diverge. An affiliate's sales in any three-month period are capped at the greatest of one percent of the class outstanding as shown by the issuer's most recent published report or statement, or the average weekly reported trading volume during the four calendar weeks preceding the filing of the notice. A large Form 144 often reflects a cap being measured, not conviction being expressed. When the corresponding Form 4 rows appear later, they are the record of what was executed, coded, and dated.

What Would Invalidate This

The frame above assumes that transaction codes carry information about intent. Several conditions break that assumption.

• Small filer populations. An issuer with three Section 16 filers cannot generate a statistically meaningful cluster at all. Two of three buying is a coincidence with a small denominator, not a consensus.
• Purchases driven by contract rather than conviction. A P row can reflect an employment agreement's ownership requirement, a purchase inside an employee stock purchase arrangement, or a director's fee taken in stock through a market purchase. The code is correct and the inference is still wrong. The footnotes usually say so, and skipping the footnotes is the failure mode.
• Signalling behaviour. Insiders know the filings are watched. A purchase intended to be read as a message is coded P exactly like any other. No coding scheme separates a genuine allocation decision from a communicative one.
• Anything that changes the reporting regime. Every threshold in this piece — two business days, 90 and 120 and 30 days, 45 days, 5,000 shares, $50,000, one percent — is a rule that has been amended before and can be amended again. The 2023 changes moved gifts from Form 5 to Form 4 and added the plan checkbox. A framework built on the current text needs re-checking against the current text.
• The step where the reasoning stops. Reading codes correctly narrows what a filing means. It does not establish that correctly-read insider purchases lead anywhere in particular. That is a separate question requiring separate evidence, and nothing above supplies it.

Concrete Framework

1. Pull the filing itself, not the summary. The screener's word for the event is not the filed code. Open the document.
2. Read the code column first. Discard everything that is not P before counting anything. Note separately what the discarded rows were, because a wall of A and F rows on one date is a compensation event worth knowing about on its own terms.
3. Check the Rule 10b5-1(c) box. If it is checked, treat the trade date as an execution date and look for the adoption date. The decision behind it is at least 30 days old, and for a director or officer normally at least 90.
4. Convert filing dates to trade dates. Build the window on transaction dates, not acceptance timestamps. Allow for the notification rule stretching plan filings up to three business days further out.
5. Test independence before calling it a cluster. Ask whether one committee decision, one vesting schedule, or one issuer event could have produced every row. If it could, the count is one, not many.
6. Scale each purchase against the holdings figure on the same form. Form 4 reports securities beneficially owned following the transaction. Use it as the denominator rather than comparing raw dollar amounts across filers.
7. Read every footnote before drawing a conclusion. Contractual purchases, sell-to-cover mechanics, indirect ownership through trusts and family entities, and J-coded oddities all live there.
8. Keep Form 144 in a separate column. Intent to sell and executed sales are different observations, and mixing them inflates both.
9. Record what the window would have shown in quiet periods. A cluster rule that has never been applied to periods it did not fire on has not been tested, only illustrated.
10. Re-verify the thresholds against the current rule text on a fixed schedule. The deadlines and cooling-off periods above are the ones in force now, not permanent features of the market.