Measuring Portfolio Turnover From 13F Filings

Turnover from snapshots is a different animal

True portfolio turnover is a flow measure: dollars traded over a period divided by average portfolio value. Registered mutual funds disclose a version of it in their prospectuses. Hedge funds and other institutional managers do not. What a manager above the $100 million threshold files, on Form 13F, is a quarter-end inventory of long US equity positions, due 45 days after quarter end (February 17, May 15, August 14, and November 16 in 2026; see the full 2026 calendar). The SEC's own 13F FAQ covers the mechanics.

So the honest question is never "what is this fund's turnover." It is "how different is snapshot N from snapshot N minus 1." That difference is still worth measuring carefully. It tells you how long ideas live in a manager's book, how seriously to take a filing that is already 45 days old when you read it, and whether quarterly data is even the right instrument. Applied consistently across filers, even a blunt snapshot diff sorts the investing world into slow capital and fast capital. If you have not worked with the raw documents before, start with how to read a 13F filing; everything below assumes you know what an information table row is.

Three metrics you can compute from two filings

Take two consecutive quarters of holdings for one filer, joined on security identifier. Three useful measures fall out.

Name turnover. Count positions initiated plus positions exited, then divide by the average number of positions across the two quarters. It is cheap, robust to pricing problems, and a good first sort across thousands of filers. Its weakness: trimming a giant core position counts for nothing, while churning ten tiny tail positions counts ten times.

Share-delta turnover. For each security, take the absolute change in share count, value the change at a single reference point, sum across the book, and divide by average book value. This version is size-aware, so a halved flagship position finally registers. It demands clean share data: splits and corporate actions must be adjusted first.

Implied holding period. Annualize the share-delta figure (a quarterly number times four is a serviceable approximation) and invert it. A book that replaces roughly a quarter of itself each quarter implies ideas live about a year.

Whichever you choose, compute it on shares, not on reported values. Reported value moves with price even when the manager does nothing.

What high and low turnover imply

Low turnover is what makes 13F data usable at all. If a manager replaces only a small fraction of the book each quarter, the filing you read 45 days late is still a decent map of current exposure, and quarter-over-quarter position changes carry real information. Berkshire Hathaway is the canonical case: a handful of changes per quarter against a very large book. Academic work on 13F-based replication has generally found that the disclosure lag hurts least for concentrated, low-turnover managers, which matches common sense: slow ideas survive slow disclosure.

High turnover inverts every one of those statements. The quarter-end position may have been closed before the filing went public. Worse, a quarter-end snapshot of a fast trader may not represent their typical behavior; quarter-end window dressing is a long-discussed concern in the academic literature, and a snapshot cannot rule it out. For these filers, read the 13F as information about universe, style, and sizing habits rather than as a list of current positions.

For anyone building screens or agents on this data, turnover is the first conditioning variable, not an afterthought. The same "fund X initiated a position in Y" datapoint means something entirely different coming from a five-year holder than from a manager who recycles the whole book every quarter. Rank filers by turnover first, and let that ranking set how much confidence the rest of the filing deserves.

Method caveats that bite

Every one of these has corrupted a real analysis at some point.

• Intra-quarter round trips are invisible. A position opened in April and closed in June appears in no snapshot. All snapshot-based turnover is a floor, never the true figure.
• The first tracked quarter is not a buying spree. The first filing in your dataset, whether from a genuinely new filer or simply the start of your coverage, makes every position look initiated. Turnover math must be filing-history aware.
• Amendments can double or erase a quarter. Restating and superseding amendments must be collapsed to one canonical snapshot per quarter, or your diff compares a filing against its own amendment. The full failure catalog is in how accurate is 13F data.
• Identifier changes masquerade as trades. A CUSIP change from a merger, spinoff, or re-incorporation looks like a full exit plus a full initiation. CUSIPs are not forever.
• Option legs are not stock. 13F filings include listed puts and calls reported at the notional value of the underlying shares. Include them in the diff and a routinely rolled option position reads as violent trading in the equity.
• Splits corrupt share deltas. A share count that quadruples on a 4-for-1 split is not buying. Adjust or exclude affected rows.
• Shared reporting inflates books. Multiple managers can report the same position when investment discretion is shared. Deduplicate before computing book size.

None of these are exotic edge cases. They are the default state of raw EDGAR data.

Computing turnover with the Arkolith API

Arkolith's Q1 2026 13F dataset covers 1,824 institutional filers and 1.87 million long positions representing $53.7 trillion in reported value, with amendments collapsed to canonical quarters, option legs flagged separately, and every row traceable to its EDGAR accession number. The flow for a single filer takes three calls. First, resolve the manager to a CIK:

curl -H "Authorization: Bearer YOUR_KEY" "https://arkolith.com/api/v1/search?q=berkshire"

Then pull the holdings, which carry share counts, reported values, and quarter-over-quarter change classification:

curl -H "Authorization: Bearer YOUR_KEY" "https://arkolith.com/api/v1/funds/1067983/holdings"

Diff consecutive quarters on the security identifier, exclude option legs, adjust for splits, and compute name turnover and share-delta turnover as defined above. To screen across the whole filer universe rather than one fund, start from the listing endpoint:

curl -H "Authorization: Bearer YOUR_KEY" "https://arkolith.com/api/v1/funds"

The same surface is exposed as an MCP server, so a Claude agent can run this loop conversationally; the quickstart takes a few minutes. Provenance is the part that matters for agents: because each datapoint cites its source filing, a claim like "this fund exited the position" can be checked against the actual document. If a number looks wrong, pull the cited accession on EDGAR full-text search and read the information table yourself. The point of provenance is that you never take an aggregator's diff on faith.

Cross-check fast traders with faster filings

The 13F is the slowest disclosure in the ownership family. When turnover analysis tells you a manager trades faster than quarterly snapshots can track, the partial fix is the faster forms. Form 4 insider filings land within 2 business days of the trade. Schedule 13D activist stakes arrive within 5 business days. Form 3 is due within 10 days of someone becoming an insider, and Form 5 annual catch-ups within 45 days of fiscal year end. These forms cover only specific actors, insiders and large beneficial owners, but where they overlap a position they provide intra-quarter timestamps that 13F filings structurally cannot. The differences between the forms are mapped in 13F vs 13D vs 13G vs Form 4.

Arkolith tracks 51,000+ insider transactions alongside the 13F spine, joined on issuer, so the question "did anyone with faster disclosure obligations trade this name inside the quarter" is a single query (see the insider view for a ticker like TSM). For high-turnover managers this combination is often the only honest read: the 13F gives you the universe and the sizing habits, the fast forms give you timing for the slice they cover, and the turnover metric itself tells you how much weight each source deserves.

Frequently asked questions about portfolio turnover from 13F filings

Can you calculate exact portfolio turnover from a 13F?

No. 13F filings are quarter-end snapshots of long US equity positions, so any position opened and closed within the quarter is invisible, and shorts, foreign listings, and most other instruments are excluded entirely. Snapshot-based turnover is a lower bound on true trading activity and should always be presented as one.

What counts as high turnover in 13F data?

There is no regulatory threshold; the reading is relative to strategy peers. A useful qualitative line: if most of the book carries over from quarter to quarter, the filing remains a workable map of exposure, while if a majority of names are new each quarter, the snapshot is mostly stale by publication day. Compare a filer against managers of similar style and book size rather than against an absolute number.

Why use share counts instead of position values?

Reported value changes whenever price changes, even if the manager never trades, so value-based deltas conflate market moves with actual decisions. Share counts isolate the trading decision, provided you adjust for splits and other corporate actions first. Value re-enters only as a weight, to make the metric size-aware.

Does low 13F turnover make a manager safe to copy?

Low turnover means the 45-day lag costs less, because positions are likelier to still exist when you read about them. It says nothing about whether the ideas are good, whether the visible long book is hedged through instruments the form does not cover, or whether the position sizes suit anyone else. Treat turnover as a data-quality signal about the filing, not a quality signal about the fund.

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