FactSet Research Systems Inc.

The full register, from 13F filings

The Register tab above is hand research: a dozen holders, each traced to the filing it came from. This asks the same question of the whole market for the quarter ended 31-mar-2026, where 707 managers reported a position. Broad where the research is deep, and shallow in the ways a 13F always is.

Built from 13F filings, which report US listed, long, US custodied positions only. A register from this source is a floor rather than a level: a holder's true position can be far larger, as BlackRock's ASML stake shows at 1.28m shares in a 13F against 26.3m in its 13G/A. Holdings held through derivatives, outside US custody, or below a manager's reporting obligation do not appear. Where a holder is marked as a family, its total sums several separate filing entities and describes something no single filing describes; the constituent entities and their share counts are listed so the total can be checked. That roll-up can legitimately exceed the family's own Schedule 13G, because a 13G is filed by one legal entity: Vanguard reports Alphabet through ten advisory entities, and The Vanguard Group Inc itself reported zero after its January 2026 disaggregation.

What a reader needs to know to read these numbers

• FactSet's fiscal year ends 31 August. As of today (20 Aug 2026) the newest FactSet SEC filing by date is the fiscal Q3 2026 10-Q (period ended 31 May 2026, filed 1 Jul 2026), which postdates the FY2025 10-K (period ended 31 Aug 2025, filed 22 Oct 2025). The FY2026 10-K will not be filed until around Oct 2026. shares_outstanding (35,565,543) is therefore taken from the fiscal Q3 2026 10-Q's cover page, as of the latest practicable date stated there, 25 Jun 2026, not from the FY2025 10-K.
• Self-gate computed on the FINAL file using the project's own chart arithmetic (raw shares summed, aggregate container reduced only where a member rolls into it; here the one aggregate row, the insider group, has no members so it counts at full value). Register array has 11 rows. Total shares = 4,491,845 + 4,200,856 + 2,858,389 + 1,873,867 + 1,439,274 + 1,380,700 + 1,100,714 + 954,411 + 350,319 + 54,704 + 439,643 = 19,144,722. Divided by shares_outstanding 35,565,543 = 53.83%.
• 53.83% is above the project's usual 25-45% band but below the 55% ceiling that requires investigation, and it is corroborated rather than merely asserted. This project's own independently built register snapshot (data/registers/FDS.json, a 31 Mar 2026 13F-derived snapshot) reports total_shares_reported of 37,810,441 across 707 filers, which is itself 106.3% of the 25 Jun 2026 shares_outstanding figure used here (37,810,441 / 35,565,543). That is consistent with FactSet being a heavily institutionally held mid-cap whose share count has been shrinking through continued buybacks (35,777,453 shares outstanding on the 31 May 2026 balance sheet date versus 37,645,870 on 31 Aug 2025, per the fiscal Q3 2026 10-Q balance sheet) while institutional share counts held roughly steady, mechanically pushing any percentage-of-company figure higher when divided by the smaller, more current denominator. Per holder, every one of the 9 distinct institutional filers checked here that also appears in the 31 Mar 2026 snapshot lands within about 1% to 21% of the snapshot's own figure for that SAME filer (Vanguard family -1.4%, BAMCO +11.8%, BlackRock +1.2%, State Street -1.3%, Loomis Sayles -2.6%, Geode +20.9%, Morgan Stanley +4.8%, Invesco -10.9%), which is the expected range of quarter-to-quarter drift for a genuine, non-duplicated filer rather than evidence of double counting. The two rows with no snapshot counterpart, Capital Group and T. Rowe Price Associates, are independently sourced small positions (0.99% and 0.15%) that the snapshot itself is documented to have omitted, per this project's brief.
• Implied-price check run on every row: dividing value_usd by shares gives approximately $230.03 to $230.11 per share on every row except T. Rowe Price Associates as filed. T. Rowe Price Associates' raw XML reports a value of $12,587 (i.e. thousands) against 54,704 shares, which implies about $0.23/sh if taken literally; multiplying by 1000 to $12,587,000 gives approximately $230.09/sh, consistent with every other row, confirming this is the known values-in-thousands filer quirk and not a units error. The value_usd field for that row (12,587,000) is the corrected figure, not the raw XML value.
• Holdings side: FactSet's newest filings disclose only that the company holds and has been trimming some minority equity investments, with NO name, ownership percentage, or standalone balance disclosed anywhere searched. The fiscal Q3 2026 10-Q's non-GAAP reconciliation footnotes mention 'the impairment of an equity investment' (three months ended 31 May 2026) and, separately, 'the impairment of certain equity investments' (nine months ended 31 May 2026); its MD&A Liquidity section separately states $36.1 million in proceeds from 'the sale of certain equity investments' during the nine months ended 31 May 2026, roughly matching the $36.4 million drop in the balance sheet's 'Other assets' line (from $105.5 million at 31 Aug 2025 to $69.1 million at 31 May 2026), which is where these investments appear to sit unbroken out. No dedicated note (the 10-Q's 15 numbered notes run Description of Business through Segment Information, with no 'Investments' or 'Equity Investments' note) discloses a name, count, ownership percentage, or aggregate carrying balance for these investments. Searched in full for: equity method, equity investment(s), non-marketable, joint venture, unrealized gain, investments in affiliates, cost method, variable interest entity, minority interest, strategic investment, privately held, readily determinable, observable price, measurement alternative, ASC 321. All were either not found or matched only the generic mentions described above.
• The FY2025 10-K (fiscal year ended 31 Aug 2025, filed 22 Oct 2025) was also searched in full for the same terms and contains ZERO mentions of equity investments, joint ventures, or equity-method accounting of any kind; the equity-investment activity described above appears to be new to fiscal 2026. Exhibit 21 of the FY2025 10-K (the subsidiaries list) was checked and is a bare two-column table of subsidiary name and state/country of incorporation, with NO ownership-percentage column and no joint-venture entities, a genuine negative rather than evidence of looking in the wrong place.
• Per the project's rule against assigning a combined balance to one investee: because no named investee and no disclosed split (or even a disclosed combined balance) exists anywhere in the documents searched, no holdings row is created and no figure is modelled. The holdings array is empty; this is a sourced finding (equity investments exist and are being wound down, but are wholly undisclosed by name or size) rather than a gap left unexplained.
• Insider ownership is small and unremarkable: FactSet has no founder, no dual-class share structure, and 'all directors and executive officers as a group (21 persons)' held only 439,643 shares (1.2%) per the 2025 DEF 14A, with no individual director or NEO reaching even 1% individually, so no member rows were broken out under it.
• CUSIP 303075105 was used to match every 13F row, per the project's rule to match on CUSIP rather than issuer-name string; several filers abbreviate the issuer name differently ('FACTSET RESH SYS INC' vs 'FactSet Research Systems Inc' vs 'FACTSET RESEARCH SYSTEMS INC') but all share the same CUSIP.
• Ninety One (a UK-based asset manager) appeared in this project's 31 Mar 2026 register snapshot as 'Ninety One UK Ltd' holding 1,194,996 shares, which would rank a top-6 holder if confirmed. The only Ninety One CIK independently located and checked for Q2 2026 (0001633343, 'Ninety One North America, Inc.') reports a much smaller position (84,770 shares), evidently a different legal entity than the UK one in the snapshot. The UK entity's own CIK could not be located within this task's time budget, so no row is included for Ninety One and the gap is recorded in unknowns rather than guessed at.
• Coordinator correction: the self-gate note above originally stated 10 register rows against 11 in the file. The SHARE TOTAL was always right, because the arithmetic it spells out sums 11 addends to 19,144,722, so the chart was never wrong; only the number a reader checks it against was. This is the recurring self-gate row-count defect already recorded in STATE.md as mechanically checkable, now seen in a fourth chunk.