How many managers are in the same trade

Methodology

Form 13F is a structured data set with a long history and a set of discontinuities that will silently corrupt any aggregate built on it. Each is stated here with the count it produced in this sample. The build script refuses to write its output when any of its checks fail.

The VALUE column changed units

Until the January 2023 amendments, the VALUE column of the holdings table was reported in thousands of dollars; from January 3, 2023 it is reported in whole dollars. The break is verified from the data rather than taken on trust. Some managers report a position for the same period, the same security and the same share count both before and after the cutoff, which happens whenever an amendment is filed late; there are 206,953 such pairs from 81 managers here, and the median ratio of the later VALUE to the earlier one is 999.93. 152,616 of the pairs sit within 10% of 1000 and 52,707 within 10% of 1, the second group being filers who did not change units when the rule did.

Read off periods instead of filing dates, the raw median VALUE per position jumps by a factor of 396.5 at the period ended December 31, 2022, which is the first quarter whose filings were mostly made after the cutoff. That is the wrong place to cut, and the difference between the two readings is the whole point: the unit belongs to the FILING DATE and not to the period, so a normalisation by period would misstate every late amendment. This build normalises by filing date and then repairs the filers who ignored the change, described below. The chart shows the raw implied price per share and the same figure after normalisation.

Source: SEC, Form 13F Data Sets | SEC, Frequently Asked Questions About Form 13F The normalised series is the gate: the build refuses to write if any fully covered quarter has a median implied price outside $1 to $400 per share. Across the 52 quarters here it runs $37.24 to $72.62. Methodology

Filers who did not change units, and filers who never did

Normalising by filing date is necessary and not sufficient, because a minority of filers report the wrong unit for their own filing date. Two independent tests find them. The first compares each filing's position table to the total the SAME filer declared on the summary page of the SAME filing. Across the fully covered quarters, of 297,195 filings carrying a usable declared total, 3,854 sit an exact power of ten from their own declaration and 1,550 disagree by more than 20% without being a power of ten apart. That test measures but cannot correct, because a ratio of 1000 says the two numbers disagree and not which of them is wrong.

The second test supplies the direction. Many managers report the same security in the same quarter, so the cross-manager median implied price of a security is a consensus that no single filer can move. Where a strict majority of a manager's testable positions sit the same whole number of decimal places from that consensus, the whole filing is divided by that power of ten: 10,776 manager-quarters of the fully covered quarters were corrected this way. The test assumes the share count is right and the value column wrong, which is not always which way round it is, so a correction that would carry a filer past the largest manager the test does not touch is read as pointing the wrong way and dropped: 2 corrections were dropped on that rule. Finally, 140 manager-quarters whose implied prices remain more than an order of magnitude from the consensus after all of that are dropped from the panel entirely, share counts included, because no power of ten reconciles them. The build refuses to write if the excluded managers exceed half a percent of any quarter's managers, or if any single filer ends up with more than 35% of a quarter's reported value.

The second of those tests has a blind spot and it should be stated plainly. It works by comparing what different managers report for the same security, so it cannot say anything about a security only one manager reports: in 2026 Q1 that is 6,236 of the quarter's 19,859 securities, carrying 0.04% of reported value. Those positions are carried as filed, and the crowding section above shows the one quarter in which they matter. The first test, against the filer's own declared total, reaches them, but a disagreement there says only that the two numbers cannot both be right.

Amendments, notices and double counting

An amendment on Form 13F is one of two different things and the form says which. A RESTATEMENT replaces the filing before it; NEW HOLDINGS adds to it. There are 10,295 restatements and 4,980 new-holdings amendments in the sample, plus 53 amendments carrying no type at all, which are treated as restatements because that is the conservative reading. Resolving them leaves 10,403 superseded filings out, 24 of which are base reports superseded by a later base report for the same manager and quarter.

A 13F-NT is a notice: it reports that the manager's holdings appear in another manager's filing, and it carries no positions. In 2026 Q1, 1,907 managers filed a notice against 8,741 filing holdings. Notices never enter a denominator here, and the 549 position rows found sitting on 11 notice filings across the whole sample, which should not exist, are excluded and counted.

The duplication that cannot be removed is stated rather than hidden. When a manager is included in another manager's report, both may appear: 73,562 manager-quarters are named by CIK in another filer's list of other included managers, and 10,873 of those also file a holdings report of their own for the same quarter. In 2026 Q1 that is 417 managers whose own reports carry $23.54T, 37.3% of the quarter's reported value. Whatever part of that also sits inside the including filer's report is counted twice in the totals on this page, and the filings do not say what part, so that figure is an upper bound on the double count and the totals are reported value rather than assets. Separately, 47,333 manager-quarters named in the other direction, as managers on whose behalf a report is filed, also file their own report, which is the notice mechanism working rather than duplication. Beyond what the filings themselves declare, 13F cannot say whether two managers report the same underlying position, and this build does not guess.

What the panel is

Every security-keyed number on this page reads one panel: share positions only. The amount-type column distinguishes SH, a number of shares, from PRN, a principal amount of debt, and the two are never added; in 2026 Q1 there are 15,254 principal-amount rows carrying $198.3B, all outside the panel. Option positions carry a put or call marker and their share count is a notional, so they are excluded too: 116,152 rows carrying $4.99T in 2026 Q1. Within one filing a manager may report the same security on several lines, because the form asks for investment discretion per line, and those lines are all real and are summed. Across the effective filings of one manager-quarter, where the same security appears in more than one, only the latest filing's lines count, which is what an amendment means.

Periods are assigned from the period of report and never from the filing date. 98 of the 98 distinct periods filed on are calendar quarter ends, leaving 0 that are not. The median filing arrives 43 days after the period it reports on and the largest gap in the sample is 12,831 days, so a filing-date assignment would misplace years of holdings. A quarter is treated as fully covered only when its whole 45-day filing window sits inside the sample: 52 of 98 quarters qualify, and every chart above uses only those.

The archives themselves changed shape in 2024, from one file per quarter to rolling three-month windows of filing dates. The 53 archives are checked for gaps and overlaps in their filing-date windows before anything is read: there are 0 gaps and 0 overlaps. Filings are then deduplicated on the accession number, so an archive that did overlap another could not double count. 120,182,194 raw position rows were read in total.

CUSIP hygiene

The CUSIP column is filer-typed and shows it. 1,701 distinct strings were filed shorter than nine characters with a zero-padded form that passes the check digit, so the leading zeros are restored rather than the row dropped: the most common is 37833100 with 2,511 rows, and the repair is verified against the check digit rather than assumed. Padding stops being needed after 2016 Q3, and 14 of the 52 fully covered quarters are affected. A further 10,671 strings are still not nine characters after the repair (131,257 rows, most commonly COM). CUSIP allows the digits, the upper-case letters and the characters * @ #, and no filed string carries anything outside that set.

The ninth character of a CUSIP is a MOD 10 double-add-double check digit over the first eight, so a filed identifier can be tested without any external security master. 56,674 distinct strings fail it, on 1,561,188 rows, the most common being 0000000NA with 7,661 rows. The split by row class is the finding: across the fully covered quarters 0.17% of share-position rows fail the check digit against 20.4% of option rows, because filers construct their own identifiers for derivative positions. Rows whose CUSIP fails are outside the panel, which removes the placeholders and the constructed option identifiers by one rule rather than a hand-written blocklist. Case is normalised before any of this: 6,648 rows in 2026 Q1 were filed in something other than upper case and would otherwise have keyed as securities of their own.

What this page does not say

No sentence here states or implies a fraction of a company. Form 13F reports positions, not ownership, and turning a position into an ownership share needs shares outstanding, which is not in these filings and not on this platform. The form also covers long positions only and only in section 13(f) securities, so the reported portfolios are not the managers' portfolios: an equity long book is what a 13F is. No individual position of any named manager is published anywhere on this page, and no holdings table is downloadable from it. What is published is aggregates: counts of managers, shares of reported value, similarity scores and quantiles.