How to Track Insider Transactions With Form 4 Data

Why Form 4 is the fastest window into insider activity

Section 16 of the Securities Exchange Act requires a company's officers, its directors, and any beneficial owner of more than 10 percent of a class of its equity to report personal trades in that company's stock. The reporting vehicle is Form 4, due within two business days of the transaction. Trade on Monday, file by Wednesday. No other routine US ownership disclosure runs on a clock that tight.

Compare the rest of the disclosure calendar:

A 13F tells you what a hedge fund held weeks ago, in aggregate, with no timestamps on individual trades. Form 4 tells you what a named executive did with personal money this week, at a stated price, in a stated quantity. That asymmetry is why insider data earns a permanent place in any flow-tracking stack, next to the slower institutional picture covered in 13F vs 13D vs 13G vs Form 4.

The raw filings are free. The SEC's investor education site has a plain-language overview of Forms 3, 4, and 5, and you can pull any individual document through EDGAR full-text search. The hard part is not access. It is filtering, deduplication, and structure, which is where most homegrown pipelines quietly go wrong.

What to filter: code P, the non-derivative table, the plan checkbox

Two filters remove most of the noise before you apply any judgment at all.

First, transaction codes. Every Form 4 row carries a one-letter code explaining why shares moved. P is an open market purchase: the insider chose to spend personal cash at the market price. S is an open market sale, which is ambiguous on its own (diversification, taxes, a house). Nearly everything else is compensation mechanics.

Second, the form has two tables. Table I covers non-derivative securities, meaning actual common stock. Table II covers derivatives: options, RSUs, warrants. For conviction tracking you want Table I. Table II rows are overwhelmingly comp-cycle artifacts, and counting them as trades is the classic beginner error behind misleading "insider selling surges" headlines.

Third, check the Rule 10b5-1 indicator. A relatively recent amendment to the form added a checkbox flagging trades executed under prearranged trading plans. A purchase or sale running on autopilot, scheduled months in advance, carries far less information than a discretionary trade placed that week.

Across the 51,000+ insider transactions Arkolith tracks, applying just these filters (code P, Table I, not plan-flagged) collapses a firehose of filings into a short, readable list. You can see the filtered result on a per-ticker surface like the TSM insider page, where each row links back to its source filing.

Cluster logic: one buyer is noise, three are a pattern

A single insider purchase is weak evidence. People buy their own stock for morale, optics, or because a board norm expects it. The signal strengthens when independent actors converge. Academic work on insider trading has generally found that purchases carry more predictive content than sales, and that clusters, meaning several distinct insiders buying the same name inside a short window, are more informative than lone trades.

Practical weighting heuristics, in rough order of importance:

• Count distinct insiders, not rows. One CFO filing three Form 4s in a week is one decision. Three officers filing once each is three decisions.
• Weight the role. A CFO or CEO sees the P&L daily. An outside director sees board decks quarterly. Operating insiders buying is the stronger read.
• Size against the person, not the market. A purchase that is large relative to the insider's existing stake and likely compensation means more than a big absolute number from a billionaire founder.
• First buys beat routine adds. An insider who has never bought in the open market, suddenly buying after a drawdown, is the textbook setup.
• Tight windows matter. Purchases clustered within days of each other, near the same price, suggest a shared view rather than coincidence.

None of this is mechanical alpha. It is a prioritization scheme for human or agent attention. For a worked example of cluster logic applied to a specific month of filings, see insider buying clusters.

An API workflow for tracking insider transactions

A reasonable agent workflow has three steps: resolve the entity, pull the filtered insider stream, and cross-reference the institutional picture.

Start by resolving the company to its identifiers (CIK, ticker):

curl -H "Authorization: Bearer YOUR_KEY" "https://arkolith.com/api/v1/search?q=Taiwan%20Semiconductor"

With the entity resolved, your agent can pull the insider transaction stream through Arkolith's MCP tools, pre-filtered to non-derivative open market trades, with each row carrying the EDGAR accession number of the filing it came from. That provenance matters more than it sounds: an agent that can cite the actual accession number cannot invent a trade, which is the whole argument in stopping AI from hallucinating market data.

Then cross-reference institutional positioning, because an insider buy is more interesting when big holders are accumulating too and less interesting when they are distributing into it:

curl -H "Authorization: Bearer YOUR_KEY" "https://arkolith.com/api/v1/funds/0001067983/holdings"

curl -H "Authorization: Bearer YOUR_KEY" "https://arkolith.com/api/v1/funds"

The institutional side has real depth to check against: Arkolith's Q1 2026 13F dataset covers 1,824 institutional filers and 1.87 million long positions representing $53.7 trillion in reported value. The same data is exposed both as REST and as MCP tools for agents running inside Claude; see connecting market data to Claude and the quickstart to get a key minted and the first call made in a few minutes.

Limitations: what the two-day lag does and does not give you

Honesty about the edges keeps this signal useful.

Two business days is disclosure speed, not real time. The market frequently reprices on the filing timestamp rather than the trade date, and systematic readers parse filings within seconds of publication. If your plan is to race the reaction, you are competing with infrastructure. The durable use is slower: building conviction context, screening, and confirming theses.

Amendments rewrite history. A Form 4/A supersedes the original filing. A naive pipeline that appends every document double counts trades or preserves corrected errors. Your data layer needs supersession logic, the same class of problem discussed in how accurate is 13F data.

Footnotes carry the meaning. Indirect ownership through trusts and family entities, share counts adjusted by footnote, and multi-line breakdowns of a single economic decision all live in the fine print. Summed naively, they distort position sizes.

Form 5 surfaces stragglers late. Certain exempt transactions only appear on the annual Form 5, due 45 days after fiscal year end, so the near-real-time stream is not quite the complete record.

Sales are nearly uninformative alone. Without the plan flag, the insider's remaining stake, and the vesting calendar, a sale tells you almost nothing. Treat S rows as context, not signal, and pre-register your filters so you are not discovering "clusters" wherever you happen to look. Pair the insider view with the institutional one (the investors leaderboard is the other half of that picture) before any conclusion.

Frequently asked questions about tracking insider transactions

How quickly are insider transactions made public?

Form 4 is due within two business days of the trade, so open market purchases and sales by officers, directors, and 10 percent owners hit EDGAR within days. Form 3 (initial holdings) allows 10 days after becoming an insider, and Form 5 (the annual catch-up for exempt transactions) allows 45 days after fiscal year end. In practice the discretionary trades you care about arrive on the two-day clock.

Is it legal to trade based on insider transaction filings?

Yes. Once a Form 4 is filed it is public information, and reading public disclosures is the entire purpose of the reporting regime. What remains illegal is trading on material nonpublic information, which is a different thing from analyzing trades that insiders have already lawfully disclosed. As always, how you act on any signal is your own decision and your own risk.

Which Form 4 transactions are worth tracking?

Open market purchases (code P) in the non-derivative table, not flagged as Rule 10b5-1 plan trades, are the core signal. Sales need heavy context before they mean anything, and grant, exercise, and tax-withholding codes are compensation mechanics rather than trading decisions. Clusters of purchases by multiple distinct insiders within a short window are the strongest pattern.

Can AI agents track insider transactions automatically?

Yes. An agent connected to an MCP server or REST API can resolve a company, pull its filtered insider stream, apply cluster logic, and cite each datapoint back to its EDGAR accession number so nothing is hallucinated. Arkolith exposes its 51,000+ tracked insider transactions this way, through both MCP tools and REST endpoints.

This article explains public filings and data concepts. It is not investment advice.