Corporate ownership data · release 2026.10-1 · first versioned release · published 29 October 2026 · quarters 2000 Q4 – 2020 Q4
What changed in release 2026.10-1
Release 2026.10-1 is the first versioned release of the data. It replaces the previous data (August 2026) in the all-firms download and in the Data Explorer. All releases · This release on GitHub, with the issues it closes and opens.
This release removes 13F holdings that were repeated into quarters the institution never reported, and fixes a dozen specific errors. It also introduces a small number of new errors, and its stricter totals drop some thinly held companies. This page counts both.
If you used data downloaded before 29 October 2026
- Company totals are lower: the average total stake per company-quarter falls from 88% to 81%.
- 350 companies join the data and 285 leave.
- The data has 46.3 million rows instead of 58.4 million.
- Re-run your analyses on release 2026.10-1 rather than mixing the two, and cite the release name.
Better Corrected, removed or restored
- 13,900,540 → 24,66513F holdings repeated into quarters whose 13F did not list them
- 52,186 → 22,732company-quarters whose recorded owners add up to more than 110%, which signals an error (-56%)
- 14,321,216holdings changed by the fixes listed below
- 18,839company-quarters now in the data that were not (350 companies new to the data)
Worse New errors and lost coverage
- at most 3,600holdings this release made wrong (known problem I-0045)
- 12,267company-quarters no longer in the data (285 companies, mostly thinly held)
- 533company-quarters in both the previous data and this release that now exceed 100% and did not before
- 22,732company-quarters still above 110% (9% of all): known problems remain
On balance: about 14.3 million holdings were corrected, and at most 3,600 were made wrong. We also checked a sample of changed holdings against the SEC filings; the problems this found are listed as known problem I-0045.
Every fix, with its counts · Known problems in this release · New: firm identifiers file
New in this release
A link file from companies to permno and CUSIP
The all-firms data is at the company level (permco), with each owner’s share classes added together, so it carries no permno and no CUSIP. Release 2026.10-1 comes with a link file, firm_identifiers (CSV or Parquet), that maps each company and quarter to its share classes’ identifiers, so you can merge the data with security-level data such as our S&P 500 and profit-weight datasets. Its columns:
rdate: the quarter;permco: the company, as in the all-firms data;permno: the share class; empty for share classes that do not trade;cusip: the share class’s 8-character historical CUSIP as of the quarter end.
One row per company, quarter and traded share class, plus one row with an empty permno when the company also has untraded shares: 266,985 rows, covering every company-quarter of the data. To get it, request the all-firms data: once the request is approved, its download page offers the link file beside the data. Download the link file first: the page closes once the data download starts.
The main fix · I-0001
13F holdings now appear only in the quarters that report them
Institutions file Form 13F after each quarter. The previous data carried a holding forward into later quarters when the institution's next 13F did not list it, as if it still held the shares. Those copies were 13,900,540 of 58,365,351 rows. They are gone; the 24,665 13F rows left with a filing date inside their own quarter are genuine filings dated early.
Data table
| Year | Previous data | 2026.10-1 |
|---|---|---|
| 2000 | 12.8% | 0.0% |
| 2001 | 18.0% | 0.0% |
| 2002 | 22.5% | 0.0% |
| 2003 | 24.8% | 0.2% |
| 2004 | 24.7% | 0.0% |
| 2005 | 26.3% | 0.0% |
| 2006 | 27.0% | 0.4% |
| 2007 | 28.3% | 0.2% |
| 2008 | 31.2% | 0.1% |
| 2009 | 33.7% | 0.1% |
| 2010 | 32.7% | 0.0% |
| 2011 | 32.3% | 0.0% |
| 2012 | 31.4% | 0.1% |
| 2013 | 28.8% | 0.0% |
| 2014 | 26.6% | 0.0% |
| 2015 | 26.2% | 0.0% |
| 2016 | 25.6% | 0.0% |
| 2017 | 23.0% | 0.0% |
| 2018 | 20.1% | 0.0% |
| 2019 | 16.7% | 0.0% |
| 2020 | 8.4% | 0.0% |
What it does to company totals
Fewer companies whose owners add up to more than they could own
For each company and quarter, the stakes of all recorded owners add up to the share of the company the data accounts for. Slightly above 100%, up to about 110%, is possible: shares lent to short sellers are reported both by the lender and by the buyer. Above 110% points to an error. With the repeated copies gone, the average falls from 88% to 81%; company-quarters above 110% fall from 52,186 to 22,732 (-56%), and those above 100% from 85,562 to 51,297 (-40%). The 100–110% band itself goes from 33,376 to 28,565. Of the company-quarters in both the previous data and this release, 43,415 came back to 100% or below and 533 went above it. The same holds for voting power: 82,995 → 49,915 company-quarters above 100%.
Data table
| Year | Previous data | 2026.10-1 |
|---|---|---|
| 2000 | 6.4% (200) | 4.4% (140) |
| 2001 | 10.6% (1,534) | 6.0% (889) |
| 2002 | 18.1% (2,481) | 8.7% (1,217) |
| 2003 | 24.7% (3,246) | 12.0% (1,614) |
| 2004 | 32.7% (4,226) | 16.7% (2,220) |
| 2005 | 35.0% (4,517) | 19.1% (2,533) |
| 2006 | 40.5% (5,182) | 24.3% (3,202) |
| 2007 | 46.8% (5,950) | 30.5% (4,018) |
| 2008 | 46.9% (5,797) | 27.9% (3,562) |
| 2009 | 43.8% (5,084) | 21.0% (2,527) |
| 2010 | 44.8% (5,011) | 22.4% (2,605) |
| 2011 | 44.0% (4,781) | 21.7% (2,456) |
| 2012 | 43.0% (4,594) | 21.8% (2,418) |
| 2013 | 42.8% (4,568) | 23.1% (2,558) |
| 2014 | 41.8% (4,656) | 23.6% (2,708) |
| 2015 | 42.5% (4,806) | 25.7% (2,980) |
| 2016 | 40.4% (4,455) | 25.6% (2,886) |
| 2017 | 38.9% (4,213) | 25.0% (2,775) |
| 2018 | 36.8% (4,031) | 25.7% (2,868) |
| 2019 | 32.8% (3,600) | 24.4% (2,729) |
| 2020 | 23.7% (2,630) | 21.2% (2,392) |
Data table
| Year | Previous data | 2026.10-1 |
|---|---|---|
| 2000 | 63.7% | 61.6% |
| 2001 | 65.6% | 62.0% |
| 2002 | 74.2% | 68.2% |
| 2003 | 79.9% | 72.7% |
| 2004 | 85.8% | 78.4% |
| 2005 | 87.9% | 80.1% |
| 2006 | 92.0% | 84.0% |
| 2007 | 96.5% | 87.6% |
| 2008 | 96.5% | 86.4% |
| 2009 | 93.8% | 83.0% |
| 2010 | 94.1% | 84.2% |
| 2011 | 93.4% | 83.9% |
| 2012 | 93.0% | 84.3% |
| 2013 | 92.7% | 85.0% |
| 2014 | 92.2% | 85.5% |
| 2015 | 92.4% | 86.3% |
| 2016 | 91.2% | 86.3% |
| 2017 | 90.4% | 86.4% |
| 2018 | 89.5% | 86.5% |
| 2019 | 87.9% | 85.9% |
| 2020 | 84.4% | 84.3% |
Which companies are in the data
350 companies join, 285 leave
The data leaves out a company whose recorded ownership looks implausible too often. That check now sees lower, more accurate totals. Companies that the previous data left out only because the repeated copies inflated them are back. Others now fall below the check's floor: 261 of the 285 companies that leave do so because their recorded owners add up to less than 30% in more than 8 quarters or more than a fifth of their quarters; 24 leave for totals that are too high. Some of the holdings that leave this way are correct.
Everything this release fixes
Fixes, and how many holdings each changed
A holding is one owner's stake in one company in one quarter. Titles are those of the issues in the public issue tracker; each issue there describes the problem and its examples. Companies and quarters are those of the holdings the fix changed.
| Issue | What changed | Holdings | Companies | Quarters |
|---|---|---|---|---|
| I-0001 | Institutions' 13F holdings appear in quarters in which their 13F did not include themRemoved: a 13F holding is now shown only in the quarters whose 13F lists it. | 14,063,774 | 6,699 | 2000 Q4–2020 Q3 |
| I-0002 | Some companies appear as large holders of their own sharesSchedule 13D/13G stakes booked to the company itself now go to the investor who filed. | 141,943 | 3,989 | 2000 Q4–2020 Q4 |
| I-0027 | Some holdings are too large or too small around share-count changesPartly fixed: carried holdings of exchange-traded share classes now change only at real splits and stock dividends. | 91,391 | 1,900 | 2000 Q4–2020 Q4 |
| I-0003 | Insiders' holdings of some share classes are missingRestored: insider holdings in share classes whose name lacks the word "common". | 10,590 | 204 | 2000 Q4–2020 Q4 |
| I-0026 | At some dual-class companies, insiders' holdings are in the wrong share class or missingInsiders' holdings now match the right class at dual-class companies. | 5,220 | 11 | 2003 Q3–2020 Q4 |
| I-0006 | Dover Motorsports: share count and votes per share are wrongFixed. | 3,658 | 1 | 2000 Q4–2020 Q4 |
| I-0033 | Kimball International: Class B votes are missing from 2014Fixed. | 3,392 | 1 | 2014 Q4–2020 Q4 |
| I-0005 | Shares outstanding are wrong at some companies in some quartersFixed at the reported companies. | 893 | 2 | 2008 Q3–2011 Q2 |
| I-0007 | Marcus Corp: 2020Q4 Class B share count is from a later dateFixed. | 158 | 1 | 2020 Q4 |
| I-0004 | Jeff Bezos's Amazon holding is shown as zero in every quarterFixed: restored in all 81 quarters. | 81 | 1 | 2000 Q4–2020 Q4 |
| I-0009 | Alphabet: Larry Page's and Sergey Brin's voting power is understatedFixed. | 42 | 1 | 2015 Q4–2020 Q4 |
| I-0013 | Bank of America: an implausible 41–45% stake is shown for one Form 4 filerRemoved. | 27 | 2 | 2016 Q3–2020 Q1 |
| I-0043 | Some fund managers' stakes are counted twice, once from Form 13F and once from Schedule 13GPartly fixed: Legg Mason Capital Management's 2005Q4-2007Q3 13Gs, already inside Legg Mason Inc's 13F. | 21 | 4 | 2006 Q1–2007 Q3 |
| I-0028 | Some Schedule 13D/13G stakes in convertible share classes are counted twiceFixed for the reported cases; the general problem stays open. | 9 | 1 | 2018 Q4–2020 Q4 |
| I-0029 | Some insiders' stakes are still shown after a later filing reported their sale or exitFixed for the reported cases; the general problem stays open. | 9 | 2 | 2019 Q1–2020 Q4 |
| I-0031 | Some insiders' holdings through several trusts are understatedFixed for the reported cases; the general problem stays open. | 6 | 1 | 2019 Q3–2020 Q4 |
| I-0032 | Some holdings are missing where two owners hold exactly the same number of sharesFixed for the reported cases; the general problem stays open. The same holdings as I-0031. | 6 | 1 | 2019 Q3–2020 Q4 |
| I-0008 | Some Schedule 13D/13G stakes shown for 2020Q4 are out of dateFixed for the reported cases; the general problem stays open. | 1 | 1 | 2020 Q4 |
| I-0030 | Laredo Petroleum: Warburg Pincus funds still appear as holders after selling their stakeFixed for the reported cases; the general problem stays open. | 1 | 1 | 2020 Q4 |
What got worse
New errors this release introduces
Some fixes expose an older error that the removed or moved rows used to hide. Checking a sample of changed holdings against the SEC filings found four kinds, together at most about 3,600 holdings (known problem I-0045):
| What goes wrong | Example |
|---|---|
| A blockholder group's old Schedule 13D/13G stake is shown as held by the company itself after the group reported it had sold | Asbury Automotive 2007Q2: 54%, though the May 2007 13G/A reports 0; Metromedia Fiber Network 2002Q1: 55% |
| An insider's Form 4 and its amendment are both counted | American Software 2011Q2: James C. Edenfield's Class B shown as 3.84 million shares instead of 1.86 million |
| A stake is shown in the wrong share class | Loews 2007Q1, Carolina Group stock |
| A split is applied to a share count already adjusted for it | Life Partners 2010Q4: Pardo Family Holdings shown with 1.25 times its 7.49 million shares |
Companies that are back in the data bring their own older errors with them, now visible again: for example overstated Schedule 13D/13G stakes (I-0044), and at Scholastic one insider shown with more than 100% of the votes. Insider rows grow from 2,625,757 to 2,771,789, mostly from restored share classes and returning companies.
Known problems in this release
Still open, and planned for the next release
The Data Explorer marks these problems at the companies and investors where they are known, in red.
How these numbers were computed
The full data files of the previous data (August 2026) and of release 2026.10-1 compared row by row, both as published. A company-quarter's total is the sum of its recorded owners' stakes (shares over shares outstanding); voting power the same with votes. "Repeated" 13F rows are 13F rows whose filing date falls on or before the end of their own quarter: a 13F reports a quarter after it ends, so such a row is a copy of an earlier report. Holdings per fix are the rows each fix was found to change. The count of new errors is an upper bound from the groups of changes that the check against the SEC filings found wrong.
To get the data, request a download; it is free of charge for academic research.