13F Amendments Explained: Restatement vs New Holdings

One form type, two opposite instructions

Any institutional manager holding at least $100 million in covered US equities must file a 13F-HR within 45 days of quarter end (the 2026 deadlines are February 17, May 15, August 14, and November 16; the SEC's investor.gov primer on Form 13F covers the basics). Amendments arrive afterward as 13F-HR/A documents tagged to the same reporting period, and they keep arriving for weeks, sometimes months.

The critical detail sits on the cover page. EDGAR's 13F schema requires every amendment to declare an amendmentType of either RESTATEMENT or NEW HOLDINGS, a distinction the SEC's own Form 13F FAQ walks through. A restatement says: discard my original report, this document is the complete and correct holdings table for the quarter. A new-holdings amendment says the opposite: my original report stands, and these rows are positions that were omitted from it.

So the same form type carries two opposite instructions. A restatement is a full replacement; a new-holdings amendment is a patch. If you parse filings yourself (our guide to reading a 13F covers the document layout), this one field decides whether the information table you just downloaded is a whole portfolio or a fragment of one. Plenty of new-holdings amendments contain a handful of rows. Treat that fragment as the fund's entire quarter and a giant book quietly shrinks to a sliver of itself in your dataset, and nothing inside the rows themselves warns you.

Why managers amend in the first place

Three reasons dominate. The first is ordinary error correction: wrong share counts, a value column reported in the wrong units, a bad CUSIP. The manager, or more often its filing agent, catches the mistake and refiles. These typically arrive as restatements, since the cleanest fix for a broken table is a complete replacement.

The second is omission. A position, an account, or a sub-manager's book was left out of the original report, and the filer adds it once discovered. These arrive as new-holdings amendments, because everything in the original was correct as far as it went.

The third, and the most interesting one if you track institutional money, is confidential treatment. Section 13(f) allows a manager to ask the SEC to withhold specific positions from public view while an accumulation program is underway, on the theory that immediate disclosure would reveal the program and move the price against it. When the confidential period lapses or the request is denied, the omitted positions must be disclosed, and that disclosure lands as a new-holdings amendment, sometimes a full quarter or more after the original deadline. The most famous practitioner is Berkshire Hathaway (live holdings on its fund page), which has repeatedly used the provision while building large stakes that only surfaced in later amendments.

The practical consequence: a quarter is not final at the deadline. The "complete" picture of what a major filer held can legitimately change months after everyone has finished writing about it.

The double-counting trap

Naive pipelines break on amendments in two mirror-image ways, and a small table makes the trap obvious:

Strategy one, keep everything, sums the original and its restatement. The fund's reported value roughly doubles, and every continuing position shows up twice, which delta logic downstream then misreads as a buying spree. Strategy two, latest-document-wins, handles restatements correctly but destroys the confidential-treatment cases: the fund appears to have liquidated everything except the patched rows, which generates a dramatic and entirely fictional "sold it all" narrative, followed by an equally fictional re-entry next quarter.

Both failure modes produce convincing, wrong stories, and they are not hypothetical. When we audited our own pipeline, we found collapsed quarters at some of the most-watched filers, Berkshire included, and rebuilt them with merge-forward logic that folds new-holdings rows into the original report instead of replacing it. Amendment handling is one of several repeatable 13F traps; the broader catalog is in how accurate is 13F data.

Supersession chains: the model that survives

The mental model that survives every edge case: a fund-quarter is not a document, it is the result of folding an ordered chain of documents. Build the chain per filer and reporting period, ordered by EDGAR submission sequence (later submissions receive later accession numbers; the cover page's amendment number is a useful cross-check). Then fold:

1. Start from the original 13F-HR as the running state.
2. A RESTATEMENT replaces the running state entirely.
3. A NEW HOLDINGS amendment merges its rows into the running state.
4. The final state of the fold is the effective view. Every earlier document is marked superseded, never deleted.

Two engineering notes from running this in production. First, compute supersession in exactly one place. The moment two code paths each decide "which filing wins," they will eventually disagree, and you get phantom portfolios that differ between your API and your screens. Second, keeping superseded documents is not hoarding: point-in-time questions ("what did the market actually know on May 20?") can only be answered from the as-filed history, since the effective view backfills knowledge that did not exist at the deadline.

Per-row provenance is what makes the whole structure auditable. Every figure should carry the accession number of the document it came from, so a human (or an agent) can pull the original filing on EDGAR full-text search and verify it. For AI agents this is also the grounding story: a number with a citation chain is checkable, a number without one is a hallucination risk, as we argue in stop AI hallucinating market data.

Querying amendment-resolved data

Arkolith serves the effective view, not the raw document pile. Our Q1 2026 dataset covers 1,824 institutional filers and 1.87 million long positions representing $53.7 trillion in reported value, every row resolved through the supersession logic above and carrying the SEC accession number it came from. The same data backs the REST API, the MCP tools, and the human screens at /investors.

# List covered institutional filers
curl -H "Authorization: Bearer YOUR_KEY" "https://arkolith.com/api/v1/funds"

# Resolve a fund name to its CIK
curl -H "Authorization: Bearer YOUR_KEY" "https://arkolith.com/api/v1/search?q=berkshire"

# Amendment-resolved holdings for one filer (Berkshire Hathaway)
curl -H "Authorization: Bearer YOUR_KEY" "https://arkolith.com/api/v1/funds/1067983/holdings"

Each holding row includes its provenance, so an agent can cite the exact filing behind any position instead of asserting it from memory. If you want the figures as they stood before an amendment landed, the superseded documents remain queryable rather than overwritten. Setup takes a few minutes: mint a key, point your agent at the MCP endpoint or the REST surface, and start with the quickstart; the full tool and endpoint reference lives in the docs.

One honest caveat: no provider can resolve an amendment before it exists. If a manager held a position under confidential treatment, every dataset on earth shows the same gap until the disclosure is filed. What a careful pipeline guarantees is that the gap closes correctly the moment the amendment hits EDGAR, instead of corrupting the quarter it patches.

Frequently asked questions about 13F amendments

How late can a 13F amendment be filed?

There is no fixed outer deadline. Amendments routinely arrive weeks or months after the 45-day filing deadline, and confidential-treatment disclosures can land a quarter or more later. Treat any recent quarter as potentially still in motion.

Does a 13F/A always replace the original filing?

No. Only a restatement replaces the original in full. A new-holdings amendment supplements it, so the effective portfolio is the original merged with the amendment, and discarding the original destroys the quarter.

How do I tell which type of 13F amendment I am looking at?

The cover page declares it: the amendmentType field is either RESTATEMENT or NEW HOLDINGS. Never infer the type from the size of the information table, because small restatements and large new-holdings patches both exist.

Why would a huge position appear only in an amendment?

Usually confidential treatment: the SEC permitted the manager to omit the position while an accumulation program was underway, and the disclosure obligation landed later as a new-holdings amendment. Plain omission errors, corrected after the fact, are the other common cause.

This article explains public filings and data concepts. It is not investment advice.