45 Days Late: Why Copying 13F Filings Doesn't Work — and What They're Actually Good For

August 14 was the filing deadline for Q2 13Fs. Over the past few days, your feed has probably been taken over by the same kind of headline: "Here's the list — what Buffett's successor bought with $23.5 billion." The quick ones may have already placed orders straight off the list.

If you haven't placed an order yet, congratulations — being a step slow is actually the right call here. Those numbers are all real; the problem is what you do with them. What makes a 13F genuinely valuable isn't only what's written on it.

Start With a Case You Couldn't Have Copied

After the close on May 15, 2024, Berkshire's Q1 13F went up on the SEC's site, and out of nowhere the list contained a property-and-casualty insurer that had never appeared before: Chubb. It wasn't a small position — nearly 26 million shares, roughly $6.7 billion by market value — and the moment it surfaced, it took a seat among Berkshire's top ten holdings.

That position had actually been getting built since July 2023. For nine months, across three full quarters of 13Fs, there wasn't a word about it. The market could only tell from a single line of small print on the cover page that he was hiding something — and after guessing for most of a year, barely anyone guessed it was this.

But this wasn't a failure to report. Buffett had applied to the SEC for confidential treatment. His stated reason was that the position wasn't finished being built. Berkshire's orders are too big; taking in a company takes several quarters, and if he shows his cards halfway through, the copycat flows pour in and every remaining share has to pay extra for his reputation. The SEC accepted it and approved.

In after-hours trading on the reveal day, Chubb's price was pushed to an all-time high. By the time you saw the name in a push notification, though, the new price was already sitting there.

For nine months nobody knew about the position. The second anyone knew, the price knew too. This isn't a loophole; the rules themselves allow it. Institutions managing over $100 million in 13(f) securities must file quarterly, with a deadline 45 days after quarter-end — that's the lag out in the open. Confidential treatment then adds another layer on top of that lag, hiding a position for several quarters at a stretch. For big money, filing and showing your hand on the spot have never been the same thing.

By Design, Half the Table Is Missing

Even without a confidentiality request, what gets submitted in a 13F isn't a complete holdings sheet. The SEC's reporting scope explicitly excludes the following:

Shorts aren't reported at all, and you're not allowed to net shorts against longs. A fund's 13F might show 5 million shares of some company while it's simultaneously short three other names in the same industry, or has simply bought puts on that stock as a hedge. You're seeing one leg; it's standing on two.

Cash doesn't make the table either. A manager might have moved 30% of the portfolio into cash this quarter, and you'd see nothing of it on the 13F. In asset management, cutting down to cash is often a far more important decision than what got bought.

Bonds and non-US-listed equities aren't in scope either. A global macro fund's real risk exposure might sit in Japanese government bonds and European equities; the 13F only photographs its small US-equity slice.

Options are reported, but not the strike, not the expiry, not what they're hedging. A call position with $500 million notional might be a heavy bet, or it might be covered calls written to collect premium — on the filing they look exactly the same.

But the more important question now is that even the numbers that are honestly reported aren't necessarily saying anything.

8.02%, 7.97%, 8.18%

These are BlackRock's ownership percentages in Nvidia, Apple and Microsoft (reporting period June 30, 2026). State Street's figures are 4.17%, 4.21%, 4.25%.

The three companies are in completely different businesses. Yet the same institution's ownership percentage across all three is nearly identical.

The answer is index funds. The bulk of the money these two manage is passive, allocated mechanically by index weight — whatever weight the S&P gives Nvidia, they have to buy a position in proportion. So news like "BlackRock added 16.39 million shares of Nvidia this quarter" has nothing to do with being bullish or bearish; it's just the arithmetic result of tracking an index.

The same table has other traps buried in it. Some names are tagged 2026-06-30, some are still stuck at 2026-03-31, and some aren't 13Fs at all. Jensen Huang's 3.55% on the shareholder list comes from insider filings — a different system entirely from institutions' quarterly snapshots. In mid-June, about 445,000 shares in his name disappeared over two days, and plenty of pages logged it straight as "reduced holdings." But open the original Form 4 and 400,000 of those shares were a gift, while 46,000 were withheld for taxes on vesting restricted stock — not a single share was sold on the open market. Different data sources, different reporting periods, different definitions: mix them into one table and compare across, and your conclusion is guaranteed to be wrong.

A more subtle trap is in market value — what you actually want to look at is share count. A stock rises 50% during the quarter, the institution didn't touch a single share, and the position's market value still rises 50%. That's where a lot of "Fund X massively added to its position" headlines come from.

There's another kind of trap that academics call window dressing. Fund managers have an incentive to buy the quarter's winners and clear out the losers in the final days of a quarter, so the filing they turn in looks more respectable. Researchers have shown that the worse a manager's performance, the more pronounced these quarter-end moves are. That June 30 snapshot may itself have makeup on. That said, most long-only value institutions don't do this sort of thing; their filings are basically genuine. But from the outside, you can't tell which ones are made up.

So What Is It Actually Good For?

After all that, if the conclusion were "13Fs are useless," you could just swipe away right now.

They're useful. You just have to use them in reverse.

The most worthwhile thing to look at is exits, not entries. An entry might just be a starter position — 1% of the book, and being wrong doesn't cost much. An exit is different: that's completely rejecting a company you've already researched and owned, a much higher-cost decision. Same kind of data point, and the signal-to-noise ratio on an exit is clearly higher.

A brand-new position from zero to one carries more weight than incremental adds and trims. A position going from nothing to something means some research team did the full work and reached a conclusion. Adding or trimming 5% on an existing position might just be a byproduct of subscriptions and redemptions, rebalancing, or a risk-control adjustment.

Disagreement is worth more than consensus. When two institutions of comparable calibre make completely opposite moves on the same stock in the same quarter — one clears out, one initiates — that stock is worth your time to think about. Disagreement means there's a key variable here that the market hasn't priced yet. There's no excess return in the consensus zone; that's true for everyone.

Slow data has to be verified with fast data. 13Fs lag 45 days. But since February 5, 2024, 13D filings have been shortened to within 5 business days of the trigger, with material changes to be amended within 2 business days. Insiders' Form 4s are faster still — 2 business days ever since Sarbanes-Oxley in 2002. Use the slow quarterly data to find a direction, then use these two fast ones to check whether it still holds.

The most important use of all is as a search-space compressor. There are over five thousand US-listed stocks; you couldn't get through them in a lifetime. The real value of a 13F is letting you pull out twenty names you've never heard of that a serious institution has bet heavily on. Which of those twenty you can buy, though — it won't tell you a word about that.

But it's not that it won't tell you; it's that this layer of information has a shelf life, and academics have already run the numbers for you. One study using a 1994–2010 sample found that after funds started submitting 13Fs, their own seven-factor alpha fell by 3% to 4% a year — being watched has a cost, and the people taking that excess return move far faster than you. Another study found that disclosed names do still show a bit of positive excess return in the two days after disclosure, but for buy-and-hold long-term investors, copying 13Fs earns no extra money.

Two days. That's the entire information dividend in this document. Four days have now passed since August 14 — that dividend was carved up long ago.

So, back to those lists flooding your feed these past few days.

A 13F is a photo taken 45 days ago, with half of it cropped out, and makeup on.

Use it as an answer, and you won't win by copying.

Use it as a clue, and it can compress five thousand names down to twenty. And what about those twenty?

Go read the filings yourself. There is no second path.