Tony Wang11 min readSEC 13F Data Has a Hidden 2,000-Position Cap. State Street's Real Portfolio Has 4,177.
A suspicious pattern — dozens of managers capped at exactly 2,000 positions — turned out to be real. Here's the proof, and what's still trustworthy.
If you've ever pulled a "top institutional managers" ranking from SEC 13F data and seen a suspicious number of them tied at the same round figure, you've seen this bug. We went looking for it deliberately, because a flat cap in aggregated data usually means one thing: the underlying crawl or index stopped counting past some fixed limit, and every "ranking" built on top of it is quietly wrong.
This one was real. Here's how we proved it, what it actually looks like from the inside, and what's still safe to cite once you know where the floor is.
The tell: dozens of managers, one suspicious number
Pulling SEC institutional positions aggregated by manager and sorted by position count should return a real distribution — some managers run concentrated portfolios, some run thousands of small positions, and the numbers should look like it. Instead, the top of the list looks like this:
Pension funds, banks, quant shops, and asset managers with nothing in common except one number. That's the signature of a fixed limit somewhere in the pipeline, not a genuine finding that dozens of unrelated institutions each hold precisely 2,000 positions. The question is where the limit actually lives, and whether the one manager that doesn't fit the pattern (Mizuho, at 2,042) is a clue — or a second problem on top of the first.
Proving it: State Street's real portfolio is more than double what either endpoint shows
A round number by itself is circumstantial. To confirm it, we needed a source that doesn't go through the same index — something that reads a number straight from the filing. SEC 13F cover pages report their own total holdings count, and a direct per-filer lookup can surface it.
We resolved State Street Corp (a global custodian bank, not a boutique shop — a large real 13F portfolio is exactly what you'd expect) to its SEC CIK and checked three numbers for the same manager, the same quarter:
Round number — the tell
The documented 'reliable' path — still capped
Read directly from the 2026-06-30 13F
That's the part that makes this worse than a display-only bug. search() filtered to an exact manager CIK is the documented reliable path in this dataset — the one built specifically to sidestep the facet's aggregation quirks. It still returns exactly 2,000. The cap lives in the underlying search index itself, not just in how the facet presents it. Querying around the facet doesn't get you around the cap.
A second large manager confirms the same shape. AllianceBernstein L.P.'s facet count and exact-CIK search total both read 2,000; its real 2026-06-30 filing holds 3,339 positions — read directly off the filing via the same raw lookup.
Show the full three-manager cross-check
| Manager | Facet count | Search (exact CIK) | Raw filing total |
|---|---|---|---|
| State Street Corp | 2,000 | 2,000 | 4,177 |
| AllianceBernstein L.P. | 2,000 | 2,000 | 3,339 |
| Mizuho Securities USA LLC | 2,042 | 1,379 | 690 |
The outlier isn't a clue — it's a second, separate bug
Mizuho Securities USA is the one manager that broke the round-2,000 pattern in the original facet pull, at 2,042. A reasonable first guess is that Mizuho is the one honestly-reported large manager — maybe its real count happens to sit just past the cap boundary.
It's the opposite. Mizuho's real single-quarter portfolio, read directly from its 2026-06-30 filing, is 690 positions — nowhere close to 2,000, let alone 2,042. Its exact-CIK search total is 1,379, which lines up neatly with two quarters of real filings (690 × 2 ≈ 1,380) — that number, at least, is internally consistent and trustworthy. But the facet's 2,042 doesn't match either of the other two numbers for the same manager. It's higher than Mizuho's own search total and unrelated to its true holdings.
That means the facet's per-manager count comes from a different, apparently buggier computation than the search endpoint — not simply "the same undercount, worse." The single number that looked like a hint about how the cap works is itself unreliable, and shouldn't be read as evidence of anything.
A second, unrelated bug: pagination doesn't work
While cross-checking the facet, we also tested whether paging past the first results would surface managers below the visible top block. It doesn't. Requesting page 1, page 10, page 30, and page 99 of the same facet — each with the same page size — returned byte-identical results every time, for both the manager facet and the issuer facet.
There is currently no way to retrieve more than the first ~50-100 buckets of either facet. Any "top 200 managers" or "top 500 issuers" pull is silently impossible today, not just capped in magnitude — the tool will happily return the same first page over and over without any error telling you it never advanced.
The issuer side has a quieter, harder-to-catch problem
The manager-position cap has a giveaway: a suspicious round number. The issuer-level facet (aggregating by company held, not by manager) doesn't show that tell — its top values look like a real, irregular distribution: iShares Trust at 173,660, Vanguard Index Funds at 40,139, down through individual companies like Alphabet at 16,795 and Microsoft at 10,042.
But irregular isn't the same as accurate. We checked six single-share-class mega-cap stocks — companies simple enough that one CUSIP fully identifies the security, so an exact-CUSIP search total is directly comparable to the issuer facet's bucket for the same company:
There's no clean cap boundary here, no obvious "this is a limit" signal. A reader pulling the issuer facet directly and citing "Apple: 9,919 institutional positions" would silently under-report the real number (11,394, per an exact CUSIP filter) by about one-sixth, with nothing in the response telling them it's wrong. That's a harder failure mode to catch than the manager cap, precisely because it doesn't announce itself.
What's actually trustworthy
Following the same rule this data forced on us in our SEC insider-trading piece — test the suspicious number, report the negative plainly, then find what's real — three things survive:
The population total is real and uncapped. An unfiltered search call — a straightforward count, not a per-bucket aggregation — returns 3,219,825 manager-issuer position rows in the live dataset. That number isn't subject to either bug above, because it isn't grouped by anything.
Two managers' true single-quarter portfolio sizes, read straight from their own filings. State Street Corp: 4,177 positions as of 2026-06-30. AllianceBernstein L.P.: 3,339 positions, same quarter. Both are more than double what the dataset's own aggregation layer will ever show you for the same manager, same quarter — a directly reproducible "the tool you're using understates reality by 2x+" finding.
Corrected institutional-holder counts for six household-name stocks, reconciled through an exact CUSIP filter instead of the undercounting issuer facet:
Show all six corrected holder counts
| Company | CUSIP | Exact-CUSIP total |
|---|---|---|
| Microsoft | 594918104 | 11,602 |
| Amazon | 023135106 | 11,354 |
| Apple | 037833100 | 11,394 |
| Nvidia | 67066G104 | 10,992 |
| Visa | 92826C839 | 8,369 |
| Broadcom | 11135F101 | 8,812 |
None of this is the "top institutional managers" ranking the original data seemed to promise — that ranking is dead, on both the manager and issuer sides, for exactly the reasons above. What's left is narrower but real: a documented, reproducible cap; two managers' genuine filing-level portfolio sizes; and six stocks' corrected holder counts, each one checkable by anyone with the same three calls.
What it's for
13F data is public and free from SEC EDGAR, and it's also one of the most inconsistently aggregated corners of financial data — commercial and free tools alike are notorious for silent caps, stale snapshots, and issuer-name collisions that merge share classes together. The three numbers here aren't a market insight; they're a map of exactly where this particular dataset's floor is, so anyone building on it knows which calls to trust and which to double-check.
Institutional 13F holdings, company financials, insider transactions, and full-text filing search are queryable at /datasets/sec over one REST API. If you're citing a "top managers" or "top holders" number from any 13F source — this one included — ask what happens when you filter by an exact CIK or CUSIP instead of trusting the aggregate.
Frequently asked questions
Is SEC 13F institutional-holdings data capped at 2,000 positions per manager?
In some aggregation and search tools, yes — we found and proved a hard 2,000-position cap in a live 13F dataset. State Street Corp's real 2026-06-30 filing holds 4,177 positions, but both the manager-level facet and an exact-CIK-filtered search endpoint plateau at exactly 2,000 for the same manager, same quarter. AllianceBernstein L.P. shows the identical pattern: real filing total 3,339, both aggregation paths capped at 2,000.
How do you prove a data cap is real and not a coincidence?
Cross-reference three independent sources for the same entity. We resolved manager names to real SEC CIKs via EDGAR full-text search, then compared each manager's facet count, its exact-CIK search total, and a raw per-filer lookup that reads a total-holdings figure straight from the filing itself. When two aggregation paths agree with each other but both fall far short of the raw filing number, that's a cap, not noise.
Why doesn't filtering by an exact manager CIK avoid the cap?
Because the cap lives in the underlying search index itself, not just in how a facet aggregation displays results. State Street's exact-CIK-filtered search total reads exactly 2,000, identical to the facet — the documented 'reliable, exact filter' path is still truncated at the same limit.
Was Mizuho Securities USA's 2,042 position count a clue to how the cap works?
No — it's a separate, unrelated bug. Mizuho's real single-quarter 13F holds only 690 positions, nowhere near 2,000. Its exact-CIK search total (1,379) roughly matches two real quarters of filings (690 x 2 = 1,380), which is internally consistent. But the facet's 2,042 figure doesn't match either of those numbers — it's higher than Mizuho's own search total despite Mizuho never approaching the cap boundary. The facet's per-manager count comes from a different, buggier computation path than search.
Does the issuer-level (by company) side of 13F data have the same problem?
A related but quieter one. The issuer facet isn't capped at a round number the way the manager facet is, but checked against an exact CUSIP filter, it silently undercounts real institutional-holder counts by 13-18% for six single-share-class mega-cap stocks (Apple, Microsoft, Amazon, Nvidia, Broadcom, Visa) — with no round-number tell to signal anything is wrong.
What SEC 13F numbers can you actually trust from this kind of dataset?
Numbers built from an exact filter, not a facet aggregation's count field. In this dataset: an unfiltered unaggregated row-count total (3,219,825 manager-issuer positions, not subject to either bug), a manager's true single-quarter holdings read directly off its own filing, and a specific security's holder count via an exact CUSIP filter rather than the issuer-name facet.