Actionable insights — reading 13F season (Q2 2026)

1. Count the funds, not the conviction — diffusion is the signal a 13F can actually carry

1. Fix a constant fund universe first (App Economy's is a list of 20 alpha-screened funds curated in early 2020 and left unchanged since — the sample must not move, or every quarter-on-quarter comparison is noise).
2. For each name, produce two counts: how many funds hold it in their top five, and how many bought it in their top five buys. Rank by the counts, not by dollar size — dollar size just re-ranks by market cap.
3. Treat a name bought independently by three or more unrelated funds as a stronger observation than any single fund's #1 position: independent convergence is much harder to explain as one manager's idiosyncrasy.
4. Then read the gap between the two counts. Widely held but rarely bought = a consensus position at maturity. Rarely held but widely bought = a theme in the act of forming.

• A name that is top-five-held by many funds and top-five-bought by one. That is the profile of a finished trade — no thesis break, just no marginal buyer left inside the smart-money sample.

2. Follow a theme outward through its supply chain, one layer at a time

1. Name the theme's core position (here: the AI accelerator). Then ask what the core has to buy to deliver: fabrication, equipment, memory, storage, interconnect, power, cooling, land, freight.
2. Check each layer against the filings — is new money already there, or still only in the core? The layer where buying is currently arriving marks the frontier of the trade.
3. Distinguish a genuine broadening (spending diffuses to suppliers because the core is capacity-constrained) from a rotation (money leaving the core). Core exposure being kept while new money moves outward is the former.
4. Follow the chain until it leaves the sector entirely — power generation, materials, rail, real estate — and check whether the same capex dollar is being counted twice in your own portfolio.

• The frontier reaching the least-obvious layer (landlords, freight, aggregates). Late-stage diffusion into the most tenuously connected names is where a broadening starts to look like a crowded theme.

3. Discount a newly-listed company's first 13F — disclosure is not demand

1. For any name that IPO'd during the quarter, check whether the funds now reporting it were private-market holders beforehand. If they were, the "new position" is an accounting event, not a purchase.
2. Ask what actually changed: the holding, or only the requirement to report it? A 13F states a position at quarter end; it does not state the trades that produced it.
3. Wait one or two more quarters before treating a recent IPO's fund count as a demand signal — the second filing is the first that can show real accumulation or distribution.

• A newly public company topping the "new positions" table in its first reportable quarter. That is the default outcome of the rules, and it will read as a stampede in every headline that quarter.

4. Read the absences — a top-five-holdings table reports rankings, not sentiment

1. Before reading who is on the list, list the names you expected to be on it and are not. The missing mega-caps are the quarter's most information-dense observation.
2. For each absence, separate the two mechanisms: (a) managers sold, or (b) the position simply fell in value and dropped out of a top-five-by-value ranking without a share being traded. A large drawdown makes (b) the null hypothesis.
3. Correct for the sample's own bias. A momentum-tilted, alpha-seeking fund list will systematically under-represent slow compounders — absence there is not absence from the market.
4. Only treat an absence as bearish when it survives all three checks and is corroborated outside the filings.

• A name leaving a top-five list in the same quarter its price fell sharply. The ranking mechanically produces the "exodus" headline; check share counts before believing it.

5. Skip the top sells — an add is a decision about the world, a trim is a decision about the portfolio

1. Default to ignoring "top sells" tables. As App Economy puts it: "hedge funds often trim high-conviction positions to manage risk or rebalance. So a 'top sell' doesn't always mean a bearish turn."
2. When a sell must be read, require it to be a complete exit, and check whether any other fund in the sample bought the same name that quarter. A full exit with offsetting buys elsewhere is disagreement, not a verdict.
3. Never infer a bearish view from a partial trim after a large gain — position sizing and tax management produce the same filing as a thesis change.

• A well-known manager's exit being reported as a sector call when other funds in the same sample were buying the identical name in the identical quarter.

6. Measure how fast a manager can reverse — it prices how much a snapshot is worth

1. Compare each notable fund's current filing to its previous one and look specifically for reversals: positions eliminated last quarter and rebuilt this quarter, or vice versa.
2. Use the reversal rate as a discount factor on that manager's filings. A manager who can fully exit and fully re-enter a mega-cap inside two quarters is telling you their 45-day-old snapshot has near-zero predictive value.
3. Apply the opposite treatment to a low-turnover filer: a change there is a policy shift, and worth reading closely.
4. Never size a position off a filing from a manager whose turnover exceeds the reporting lag.

• A complete round trip in a single mega-cap across two filings. That manager's 13F is a historical document, not a lead.

7. Separate a portfolio trade from a capital-allocation policy change

1. For the low-turnover filers, ignore individual names first and ask a single question: did the direction of net flow change — net seller to net buyer, or the reverse?
2. Date the flip. A flip after a multi-year streak is a policy statement about the opportunity set, not a view on one stock.
3. Then look at where the capital landed and how many separate uses it took. Several simultaneous uses (equities, buybacks, an outright acquisition) confirm a genuine change of stance rather than an opportunistic single purchase.
4. Reset the question from timing to returns once capital is moving — the risk changes from "will they deploy?" to "at what return, and under whose judgement?"

• A multi-year net-selling streak ending. That single fact carries more information than the identity of any of the individual purchases.

8. Sort ideas into two piles: multi-fund clusters and single-fund conviction — they need different research

1. Cluster (three or more unrelated funds, same sub-sector, same quarter): research the sub-sector's supply-demand setup, because the funds are expressing a shared industry view. The name matters less than the cycle.
2. Single-fund #1 buy: research the company. One manager sized a position to the top of their book on a specific, falsifiable argument — reconstruct that argument and decide whether you believe it.
3. Write down the argument in one sentence before doing any work. If you can't state why the fund would own it, the idea isn't yet sourced — it's copied.
4. Set the disconfirming evidence in advance, since you will not have the fund's exit discipline.

• A cluster forming in a sub-sector you don't own any exposure to. That is the highest-value output of a 13F sweep — the theme, sourced early, before the individual name matters.

9. Before copying any strategy, find the version that already exists and check its record

1. Whenever a strategy sounds obviously good ("own what the best investors own"), search for a fund or ETF that already implements it mechanically, and pull its long-run record against the index.
2. Check what the benchmark comparison excludes. A published product's returns are usually net of its own fee but exclude the fee layer of the underlying strategy — so the honest comparison is worse than the printed one.
3. If the mechanical version has failed over a full cycle, identify which ingredient the mechanism cannot copy — conviction, sizing, exit discipline, the hedges and shorts that never appear in the filing — and either supply it yourself or abandon the strategy.
4. Then use the source material for what it is good at. Filings are excellent idea generators and poor portfolios.

• Any strategy whose mechanical implementation is publicly available and has trailed the index for a decade. The gap between the idea and the product is where the real assumption is hiding.

10. Apply the four structural blind spots before drawing any conclusion from a filing

1. No shorts, no cash. A fund shown as fully long may be hedged or net-short the same theme. Never infer net exposure from a 13F.
2. Partial universe. Only managers over $100M in US equities file, and only the sampled subset is being read — the absence of a name is weak evidence.
3. US long equity only. Non-US listings, bonds, commodities, options and private positions are invisible; a macro fund's real book may be almost entirely outside the filing.
4. 45-day lag. A June 30 snapshot is read in mid-August. Match the lag against the manager's turnover (insight 6) before treating anything as current.

• Any 13F conclusion phrased as a net exposure or a market call. If the claim requires knowing the shorts, the cash or the non-US book, the filing cannot support it.