Sector Rotation Explained: What 13F Flows Really Show

What sector rotation actually means

Two different things travel under the same name. The first is return rotation: leadership in price performance shifting from one sector to another, the thing financial TV means when it says "money is rotating into energy." The second, more useful one is positioning rotation: institutions actively selling holdings in one sector and buying another, a deliberate reallocation of capital.

The distinction matters because price leadership is routinely presented as evidence of flows when often nothing has moved except prices. A sector can outperform for a quarter while large holders are net sellers into the strength. Returns are visible every day; positioning is only visible in disclosures.

For US equities the broadest positioning disclosure is the quarterly Form 13F. Any institutional investment manager with discretion over $100 million or more in covered US equities must report its long positions within 45 days of each quarter end. The SEC's own 13F FAQ covers the mechanics, and how to read a 13F filing walks through a real one field by field.

Aggregate enough filers and the positioning map gets genuinely broad. Arkolith's Q1 2026 13F dataset covers 1,824 institutional filers reporting 1.87 million long positions with $53.7 trillion in reported value. That is a large fraction of professionally managed US equity exposure, snapshotted every quarter. Whether the map shows real rotation, though, depends entirely on how you difference it between quarters. Difference it naively and you measure the market's price action, not anyone's decisions.

Why raw net flows mislead: the drift problem

Here is the failure mode, with a hypothetical. A fund holds $10 billion of semiconductor stocks at the start of a quarter. The sector rallies 20 percent. The fund does not trade a single share. Its semiconductor book is now worth $12 billion, so a naive quarter-over-quarter value delta reports a $2 billion "inflow" into semiconductors from a holder that did precisely nothing. Multiply that across hundreds of filers and a strong quarter, and the raw flow table prints enormous inflows into every sector that went up.

The tell is unmistakable: in a rising market, raw net flows show positive "rotation into" nearly every sector at once, which is arithmetically guaranteed and analytically meaningless. We know this failure firsthand. An early version of Arkolith's rotation engine served raw value deltas, every sector printed green, and the output was confidently wrong. The production pipeline now serves drift-adjusted flows only.

The fix is to decompose each position's change in value into two parts: the drift component (what the position would be worth if the holder had done nothing, given the price move) and the active component (the residual that reflects actual buying or selling). In practice the cleanest anchor is share count. 13F information tables report shares held (the sshPrnamt field) alongside dollar value, so you can ask the question that matters: did the holder's share count rise or fall? Then value that share change at a consistent reference so accumulation in a falling sector is not understated.

The third row is the dangerous one: raw flows label accumulation into a drawdown, a classic institutional pattern, as an exodus.

How filings reveal rotation, and on what delay

13F positioning arrives on a fixed cadence: a snapshot of quarter-end holdings, due 45 days later. The 2026 deadlines are:

So a 13F rotation signal describes where capital stood at quarter end, seen roughly a month and a half later, and the comparison baseline is another snapshot 90 days before that. The full calendar, with what tends to drop early versus at the deadline, is in 13F filing deadlines for 2026. Academic work on 13F disclosures has generally found that positioning information retains value despite the lag, particularly for slow-moving questions like crowding and multi-quarter sector tilts, while short-horizon trading edges decay quickly.

Faster forms partially fill the gap. Corporate insiders must report trades on Form 4 within 2 business days, and anyone crossing a 5 percent stake with activist intent files Schedule 13D within 5 business days (the differences are mapped in 13F vs 13D vs 13G vs Form 4). Insider buying clustered in a sector that the latest 13Fs already showed institutions accumulating is a timelier confirmation than either signal alone; Arkolith tracks 51,000 plus insider transactions alongside the 13F spine for exactly this kind of join. The official primer on what 13F reports contain lives at investor.gov.

Building a drift-adjusted rotation read

A defensible pipeline looks like this:

1. Aggregate by security per filer per quarter. Filers often report the same security across multiple rows (different discretion or manager codes), and option legs must be excluded from the long book. The join key is the CUSIP; see what a CUSIP is for why tickers alone fail here.
2. Resolve each CUSIP to an issuer and a sector. Filer-supplied issuer names are inconsistent, so resolution has to be done against a reference, not by trusting the filing text.
3. Compute position-versus-position share deltas between quarters. New positions, exits, and amendments need explicit handling: an "exit" might be a true sale, a filer dropping below the $100 million threshold, or a position under confidential treatment surfacing later.
4. Value the share change at a consistent reference to turn it into an active dollar flow that is comparable across rising and falling sectors.
5. Sum by sector and net it. Only now is a positive number evidence of rotation in, rather than evidence that prices went up.
6. Optionally weight by holder type. A concentrated active manager doubling a position carries different information than a broad passive complex mechanically rebalancing. The same fund-level deltas drive the manager leaderboard on /investors.

None of this is exotic. It is bookkeeping discipline applied consistently across 1.87 million positions, which is exactly the kind of work that is tedious for a human and trivial to get subtly wrong in a one-off script.

Pulling rotation inputs from an API

If an agent is doing this analysis, the failure to avoid is letting the model estimate flows from memory. Ground every number in filings. Each datapoint Arkolith serves carries provenance back to its SEC accession number, so a claim can be checked against the source document on EDGAR full-text search.

Anchor the universe first, then pull holdings per filer:

# Resolve an issuer to anchor the sector join
curl -H "Authorization: Bearer YOUR_KEY" "https://arkolith.com/api/v1/search?q=NVIDIA"

# List covered institutional filers, then pull one fund's holdings by CIK
curl -H "Authorization: Bearer YOUR_KEY" "https://arkolith.com/api/v1/funds"
curl -H "Authorization: Bearer YOUR_KEY" "https://arkolith.com/api/v1/funds/1067983/holdings"

Diff two quarters of holdings by share count, roll the active deltas up by sector, and you have the raw material for a rotation read. The same underlying data backs the human-readable ownership views, for example the institutional holders of NVDA. Setup for both the REST API and the MCP server (the agent-native interface) takes a few minutes via the quickstart.

What rotation analysis cannot tell you

Honest limits, because 13F-based rotation has real ones. The filings are long-only: a manager who appears to rotate out of a sector may simply be re-expressing the view through shorts or derivatives you cannot see. The snapshot is quarter-end only, so intra-quarter round trips are invisible, and practitioners have long debated how much quarter-end window dressing distorts the picture. Confidential treatment can hide positions for months. And the 45-day lag means you are always reading last quarter's map.

What survives those caveats is still valuable: multi-quarter accumulation and distribution trends, crowding measurement, and the question of which holders are on the other side of a move. Treat 13F rotation as a structural positioning instrument, not a trading trigger, and audit your inputs; how accurate is 13F data covers the error modes worth checking before you trust any aggregate.

Frequently asked questions about sector rotation

What is sector rotation in markets?

Sector rotation is the reallocation of capital between sectors as investors reposition, for example selling consumer staples to fund technology exposure. It is distinct from return rotation, where one sector merely outperforms another. True rotation is about positioning changes, which in US equities are observable mainly through SEC disclosures like Form 13F.

Why do raw 13F net flows overstate sector rotation?

Because position values include price drift. If a sector rallies 20 percent, every holder's book in that sector grows by roughly that much with zero buying, so summed value deltas show "inflows" into anything that went up. Drift-adjusted analysis works from share-count changes instead, isolating actual buying and selling.

How long is the lag before filings show institutional rotation?

Form 13F is due 45 days after quarter end, with 2026 deadlines on Feb 17, May 15, Aug 14, and Nov 16. So institutional positioning is always read with a lag of 45 days or more. Form 4 insider filings, due within 2 business days of a trade, provide a much timelier partial signal to cross-check.

Can an AI agent measure sector rotation from filings?

Yes, and it is a natural agent workload: pull holdings per filer via API or MCP, compute share-count deltas between quarters, resolve securities to sectors, and aggregate. The critical constraints are using drift-adjusted flows rather than raw value changes, and grounding every figure in filing-level provenance so nothing is hallucinated.

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