How Accurate Is 13F Data? Seven Failure Modes to Know

Filings are accurate; datasets are built

The filing itself is a sworn regulatory document, and materially false ones carry legal consequences. The accuracy problem lives downstream: between EDGAR's raw XML and the "hedge fund holdings" table you actually query sit a dozen transformation decisions, and most aggregators get several wrong. We know because we have hit and fixed every failure mode below in our own pipeline, across 1.9 million holdings and 1,800+ filers.

The seven failure modes

1. The 45-day lag

Positions are quarter-end snapshots disclosed up to 45 days later (the 2026 calendar). By the time you read a Q1 filing, half of Q2 has passed. The data maps committed capital; it is not a trading feed.

2. Long-only blindness

13Fs disclose long U.S. equity positions. No shorts, no foreign listings, no futures, no cash. A fund that looks wildly bullish in its 13F may be hedged to neutral in instruments you cannot see.

3. Option legs counted as stock

Filings include listed put and call positions, reported with the notional value of the underlying shares. Sum the value column naively and a fund's bearish put position becomes its largest "holding." We have seen a manager's biggest put rendered by popular trackers as his top long. The tell: a large position value with dashes where shares should be. Long books must exclude option legs and surface them separately.

4. Missed amendments

Managers file 13F/A amendments for weeks after each deadline: restatements that replace the original, and additions that extend it. A pipeline that keeps originals alongside restatements double-counts entire portfolios; one that takes only the latest document can collapse a fund's quarter to a partial addendum. Correct handling is a supersession chain: every figure should come from the effective version of each filing.

5. Identifier changes masquerading as trades

When a company re-lists, re-domiciles, spins off, or gets acquired, its CUSIP changes or disappears, and every holder mechanically shows a "new" or "exited" position. In Q1 2026, the ten most-entered securities included exactly one case of genuine buying; the rest were spin-offs and listing changes. Position-change analytics need corporate-action awareness or they measure paperwork, not behavior.

6. Filer unit errors

Filers occasionally report values in the wrong units (a long-standing thousands-vs-dollars confusion), inflating or deflating a portfolio by a factor of a thousand. A $2M fund suddenly reports $1.5B. These errors are detectable: cross-check reported values against share counts and quarter-end prices, and flag scale jumps between adjacent quarters. We run both checks continuously.

7. Issuer names are free text

The same security appears as "APPLE INC", "Apple Inc.", "APPLE COMPUTER", and worse across filers, because the name field is whatever each filer typed. Naive grouping by name splits or garbles holdings. The CUSIP is the join key; the displayed name must be resolved from a reference source, never from an arbitrary filer's spelling.

What this means in practice

Ask three questions of any 13F dataset before trusting it: Does a long-book number exclude option legs? Is every figure from the effective post-amendment filing? Do position changes distinguish trades from identifier events? If a provider cannot answer all three, its top-ten tables will be wrong in exactly the ways above.

This is also why we attach provenance to every datapoint: each holding in our data links to the specific EDGAR accession it came from, so any number can be audited back to the sworn document, by a human or by an AI agent grounding its answers. Start with the consensus readout to see the corrected output, or read a raw filing yourself to appreciate the gap.

Frequently asked questions about 13F accuracy

Can funds legally hide positions from 13Fs?

Within limits, yes: confidential treatment lets a manager delay disclosing specific positions, and assets outside the covered-securities list never appear. The filing is accurate for what it covers; coverage is the limit.

Are 13F dollar values trustworthy?

They are quarter-end marks computed by the filer, with occasional unit mistakes (failure mode 6). Share counts are the more reliable field, and value-vs-price cross-checks catch most scale errors.

Why do different 13F websites show different numbers for the same fund?

Amendment handling, option treatment, share-class aggregation, and identifier resolution all differ. Two sites can both read the same EDGAR documents and publish materially different portfolios.

Is 13F data still worth using?

Yes, when handled correctly it is the only systematic public record of institutional ownership: 1,824 managers and $53.7 trillion in Q1 2026 alone. The failure modes are all fixable in the pipeline; the information has no substitute.

This article reports on public regulatory filings and data engineering. It is not investment advice.