← Writing
Fundamentals

Insider Buying and Superinvestor 13Fs Are Social Proof, Not a Trade Signal

24 July 2026 · 9 min read


A 13F drops, a famous name appears next to a ticker, and by Monday people are buying it because Buffett did. A Form 4 shows a CEO buying their own stock and the same reflex fires on a shorter clock.

Both of those trades are built on a misreading. The filings do carry information; the academic record on that is unusually clear. But it arrives on a horizon of quarters, and the filing calendar alone should have ruled out the fast version.

The filing calendar rules out the fast version

The timestamps settle this before any of the returns research does.

A Form 4 — an insider's report of their own transaction — is due within two business days. Quick enough. You are seeing something a few days old.

A 13F is not. It covers a full calendar quarter and is due 45 days after that quarter closes. A position opened on 2 January is disclosed around 15 May. By then the manager has had four and a half months to change their mind, and often has.

It is also partial, in a way that is easy to forget. A 13F shows long US-listed equity. Not short positions, cash, bonds, foreign listings, or private holdings. A fund running a hedged pair discloses the long leg and nothing else, and you read a directional bet into what was actually a spread.

So the honest version of "Buffett bought this" is: as of a quarter-end that was some weeks ago, Berkshire's US equity book included this position, at a size you cannot compare to the parts of the balance sheet you cannot see.

The returns are real. They are also slow.

The natural conclusion — stale data is worthless — turns out to be wrong.

Jeng, Metrick and Zeckhauser studied reported insider transactions from 1975 to 1996 and found that insider purchases earned abnormal returns of more than 6% a year, roughly 40 basis points a month. Insider sales earned nothing statistically significant.

How that return arrived matters more than its size. About one-sixth of it showed up in the first five days after the transaction. About one-third within the first month. Roughly two-thirds accrued after the first month had already passed. Whatever the insider knew, the market took quarters to price it.

Martin and Puthenpurackal ran the equivalent test on 13Fs and deliberately handicapped it. They built a portfolio that copied Berkshire Hathaway's holdings starting at the beginning of the month after each filing became public, accepting every day of the disclosure lag and then adding more. Over 1976–2006, that portfolio beat the S&P 500 by about 10.75% a year.

Being a quarter late did not destroy the result. A signal you can be months late on is, by definition, not a timing signal. It is a signal about what is worth owning.

Cohen, Malloy and Pomorski added the piece that explains the noise. More than half of all insider trades are routine — the same insider, the same month, every year, like clockwork. Strip those out and the remaining opportunistic trades carry all of the predictive power, at around 82 basis points a month. The routine trades predicted essentially nothing.

Three papers, one shape. Real information, sitting in the deliberate transactions, released over quarters.

The wrong horizon for trading is the right horizon for owning

A signal that pays out over quarters is close to useless for a 30-day trade. The variance swamps it.

The same signal is well matched to a slower question: is this a business I want to be an owner of? Nobody has ever needed a filing to tell them what a stock will do next month. Plenty of people could use a second opinion on what they should be holding for the next three years.

That lands squarely on the put seller. When you sell a cash-secured put, you are not making a directional call on the next 30 days — delta and the premium already price that, and the market is better at it than you are. You are pre-committing to buy the stock at the strike if it gets there.

So the question that decides the trade is never "will this go up by expiry." It is: if I get assigned, am I content owning this for years?

Insider and 13F data speak to that question and to no other. Which is why we surface it next to the underlying rather than next to the contract.

Buying informs. Selling mostly doesn't.

Peter Lynch put the asymmetry better than any paper has: insiders might sell their shares for any number of reasons, but they buy them for only one.

The reasons to sell are endless and mostly personal. Diversification away from a position that is now most of someone's net worth. A tax bill. A house. A divorce. A pre-scheduled 10b5-1 plan hitting its date months after it was written, when the executive had no say in the timing. None of that says anything about the business.

The reasons to buy are close to singular. Someone with better information than you, who already has career risk and often most of their wealth tied to this company, chose to take on more of it with after-tax money.

So net selling earns a raised eyebrow and a second look at the bear case. Sustained buying earns real attention. They are not mirror images and should not be weighted as if they were.

What we filter out, and why

Most insider "signals" you see are unfiltered, which makes them close to meaningless. The feed is dominated by compensation events nobody chose.

What we compute is the net discretionary open-market flow in dollars over the trailing 90 days: purchases minus sales, counting only the transactions where a human made a decision.

Excluded entirely:

  • 10b5-1 scheduled sales — written months earlier, executed on autopilot. This is the routine/opportunistic split from the Cohen–Malloy–Pomorski result, applied directly.
  • Stock grants and awards — that is payroll, not a purchase.
  • Option exercises — a compensation mechanic, usually with a deadline attached.
  • Gifts — estate planning.

The headline is net dollars, not a count of filings. Eight directors buying $9,000 each is a gesture that produces eight green rows on an unfiltered feed. One CFO buying $2M is a position. Counts and distinct buyers sit underneath as texture. The dollars lead.

On the 13F side we follow a curated list of 17 recognisable managers rather than every filer — Buffett, Ackman, Burry, Icahn, Tepper, Druckenmiller, Klarman, Loeb, Li Lu and others. Each is bucketed by their most recent quarter-over-quarter action on that ticker: bought (opened or added), sold (trimmed or exited), or holding. Holding is not a filler category. A manager who has owned something across several quarters and left it alone is saying more than one who nibbled at it once.

How fresh it is, and how deep

Two questions worth answering plainly, since the rest of this has been specific about mechanics.

Refresh: the ingest runs daily, before the US open. That cadence exists for Form 4s, which arrive continuously and are worth having within a day of filing. It does close to nothing for 13Fs, which change four times a year. A daily job against quarterly data just means we pick up a new filing on the morning it lands rather than the following week.

Depth: for each manager we hold two quarters — the most recent 13F and the one before it — and diff them. That is everything the bucket needs, and it is also everything we keep. Each ingest replaces the snapshot rather than appending to it, so there is no multi-year holdings archive behind the chip. We can tell you a manager added this quarter. We cannot tell you they have been adding for six straight quarters, which is the more interesting fact.

Worth saying outright that this part is new. The 13F signal shipped in July 2026, and there is no backfilled history sitting behind it.

Three honest ways to use it

As a tiebreak, never as a reason. Two names survive your screen with similar premium and similar quality. One has a CFO who bought $2M of stock last month. That is a fair way to break the tie. What it must never do is promote a name that failed the screen. It is a tilt applied to a shortlist you built on your own criteria, not an entry point.

As a contradiction check. A name that looks cheap while insiders sell into the weakness is telling you two stories at once. That is when to go back and ask whether you are looking at a value trap rather than a margin of safety. The data does not resolve the question. It tells you the question is live.

As a patience prop after assignment. You get assigned, the stock keeps falling, and you are sitting on a loss deciding whether the thesis is broken or the market is just moody. "The CEO bought above my strike" is not a reason to hold — they can be wrong and frequently are. But "no insider has sold in six months and two of the managers who owned it still do" is evidence the thesis has not visibly broken. Weak evidence. Still better than the feeling in your stomach, which is what most people use.

What it can't do

Sizing is invisible. A $2B Berkshire position sounds enormous and is a rounding error against their equity book. "Buffett bought" and "Buffett is betting big" are different statements, and the ticker chip cannot tell them apart.

Our superinvestor list is survivorship by construction. We selected 17 people who are famous because they were right. There is no list of the managers who ran the same playbook and quietly closed. Nothing about following famous investors corrects for the fact that you only know their names because it worked.

One quarter of context, not a trajectory. The bucket compares the latest 13F against the one immediately before it and nothing further back. A manager who has quietly tripled a position over two years and happened to sit still this quarter reads as holding — identical to one who has not touched the same block of shares since 2019.

Insiders are early, and sometimes just wrong. Executives have bought their own stock the whole way into bankruptcy. Conviction and correctness are not the same thing, and the person with the most information about a company is also the person most committed to believing in it.

The most informative insider buying happens where the wheel works worst. The academic edge concentrates in smaller, less-covered companies — which is also where option chains are thin, spreads are wide, and selling puts is impractical. On a mega-cap with a liquid chain, the signal is weaker precisely because everyone is already watching.

And none of it says anything about the next 30 days, which is the actual life of your contract. Filings do not forecast earnings, guidance cuts, or the market regime you're selling into.

The point

Social proof does not make you right. As a trade trigger it puts you into names on a schedule that has nothing to do with your own thinking, months after the people you are copying acted, on data that omits half of what they hold.

Used properly it answers a narrower question, one that is hard to answer alone: am I in reasonable company as an owner of this business at this price? Sometimes the answer is that the people with the most information are quietly buying. Sometimes it is that you are the only one at the table.

Both of those are worth knowing before you promise to buy 100 shares at a price you named yourself.

Educational and analytical only — not investment advice. Insider and 13F data is sourced from SEC EDGAR filings and is backward-looking by construction.

Sources: Jeng, Metrick & Zeckhauser, "Estimating the Returns to Insider Trading" (Review of Economics and Statistics, 2003) · Martin & Puthenpurackal, "Imitation Is the Sincerest Form of Flattery: Warren Buffett and Berkshire Hathaway" (2008) · Cohen, Malloy & Pomorski, "Decoding Inside Information" (Journal of Finance, 2012)

Common questions

Can you make money following insider buying?
The research says yes, slowly. Jeng, Metrick and Zeckhauser found insider purchases earned abnormal returns of more than 6% a year over 1975–1996, about 40 basis points a month, while insider sales earned nothing statistically significant. The timing is what rules out trading it: only about one-sixth of that abnormal return showed up in the first five days, and one-third within the first month. Two-thirds accrued after that. A real edge on a horizon of quarters is the opposite of a short-term trade signal.
How stale is 13F data by the time I see it?
Up to about four and a half months. A 13F covers a calendar quarter and is due 45 days after that quarter ends, so a position opened on 2 January is disclosed around 15 May, and the manager may have sold it in the meantime. It also only shows long US-listed equity positions — no short positions, no cash, no foreign listings, no bonds. If a fund is running a hedged pair you see one leg and assume it is a bet.
If 13Fs are that stale, why look at them at all?
Because the information decays far slower than the staleness suggests. Martin and Puthenpurackal built a portfolio that copied Berkshire Hathaway's disclosed holdings starting the month after each filing became public, and over 1976–2006 it beat the S&P 500 by about 10.75% a year. Being a quarter late did not destroy the result. A signal you can be months late on is a signal about business quality and ownership, not entry timing, which is why we treat it as corroboration rather than a reason to open a trade.
Why does insider selling matter less than insider buying?
Because the reasons are asymmetric. An insider sells to diversify, to pay a tax bill, to buy a house, to fund a divorce, or because a pre-scheduled 10b5-1 plan hit its date — none of which say anything about the business. An insider buys their own stock with their own after-tax money for essentially one reason. Cohen, Malloy and Pomorski put numbers on this: more than half of all insider trades are routine, and once you strip those out, the remaining opportunistic trades carry all of the predictive power, at around 82 basis points a month. Routine trades predicted nothing.
How does levelbox filter insider transactions?
We report the net discretionary open-market flow in dollars over the trailing 90 days — purchases minus sales, and only the transactions where someone made a decision. Scheduled 10b5-1 sales, stock grants, option exercises and gifts are excluded entirely, because those are compensation and calendar rather than conviction. We report net dollars as the headline rather than a count of filings, since eight directors buying $9,000 each is a gesture and one CFO buying $2M is a position. For 13Fs we track a curated list of 17 well-known managers and bucket each into bought, sold, or holding based on their most recent quarter-over-quarter change.
How often is the insider and 13F data refreshed, and how far back does it go?
The ingest runs daily, ahead of the US open. That cadence is there for Form 4s, which arrive continuously; it makes little difference to 13Fs, which only change four times a year, so a daily job simply means a new filing is picked up the morning it lands. Depth is deliberately shallow: for each manager we hold the most recent 13F and the one before it and diff the two, and each run replaces that snapshot rather than appending to it. There is no multi-year holdings archive, so the signal can say a manager added this quarter but not that they have been adding for six consecutive quarters. Insider transactions are aggregated over a trailing 90-day window.
Should insider or superinvestor data change which strike I sell?
No. Strike selection is a function of the price you would accept, the premium on offer and your assignment tolerance — none of which this data informs. What it can do is sit upstream of that, on the question of whether the underlying belongs on your list at all. Use it as a tiebreak between names that already passed your screen, or as a contradiction check when something looks cheap and insiders are selling into the weakness. It should never promote a name that failed the screen.

Read next