The spot bitcoin ETFs did something quietly useful that has nothing to do with fees or flows. They dragged institutional bitcoin ownership into a disclosure regime that already existed. Any investment manager running more than roughly $100 million in US-listed securities has to file a Form 13F with the SEC within 45 days of each quarter end, listing every reportable position held on the last day of that quarter. ETF shares count as reportable securities, so every advisor platform, hedge fund, pension, and bank that held a spot bitcoin ETF through a quarter end is now on the public record, with share counts attached. The institutional adoption story used to run on conference anecdotes and unnamed sources. Now it is a quarterly dataset you can pull for free, provided you know how to read it.
Where the filings live
Everything starts at EDGAR, the SEC's filing system. There are two ways in. If you care about a specific manager, open their filer page and pull the latest 13F-HR, which contains an information table listing every holding. If you care about a specific ETF, which is the more useful direction for this job, grab the fund's CUSIP from the issuer's website and run it through EDGAR full-text search. That surfaces every information table containing the CUSIP, which is effectively a census of every large manager that held the fund at quarter end.
If you would rather not touch XML, free aggregators like WhaleWisdom and 13f.info index the same filings and let you screen by security. I still check the raw filings occasionally, because aggregators inherit every error in the source data and sometimes add their own. The information tables are structured XML, so a small script can pull every filing that mentions your CUSIPs and build a holder table in an afternoon. Each row gives you the filer, the share count, the market value, whether the position is shares or listed options, and who holds investment discretion. What the rows never give you is shorts, futures, or anything held by a manager under the reporting threshold. Keep that asymmetry in mind, because it drives most of the misreadings below.
Sticky money and fast money look identical in the table
The biggest mistake people make with ETF 13Fs is counting every holder as adoption. The filings mix at least three very different kinds of owner, and nothing in the table distinguishes them for you.
Advisory platforms and RIAs are the sticky end. A wealth manager files one 13F covering thousands of client accounts, so per-account sizes are small but filer counts are huge, and the money tends to stay. Advisors rebalance on a schedule, and an allocation that survived an investment committee once tends to survive drawdowns too. Pensions, endowments, and insurers sit in the same bucket and show up rarely, but each one is worth noting, because the diligence bar for getting a bitcoin ETF through that kind of committee is high. One new state pension tells you more than ten new hedge funds.
Hedge funds and market makers are the other end. They are often the largest dollar positions in the table and often the least informative. The classic trade is cash and carry: buy the ETF, short CME bitcoin futures, collect the spread. Futures are not 13F-reportable, so a fully hedged arbitrage book shows up in the filing as a giant bullish bet. Market makers are noisier still, since their holdings are inventory from creation and redemption flow, large one quarter and gone the next, with no view attached.
You can usually classify a filer in under a minute. The name helps, and the rest of their book helps more. A filer whose other holdings are index funds, munis, and hundreds of small equity lines is an advisor. A filer with concentrated positions and a stack of put and call lines is fast money. For the arbitrage question specifically, cross-check the CFTC Commitments of Traders report. When leveraged funds are running a large net short in CME bitcoin futures, assume a meaningful slice of the hedge fund ETF longs you see on 13Fs is basis trade rather than directional exposure. When the basis compresses, those positions unwind, and the resulting ETF outflows get reliably misread as institutions losing faith. It is the same trade in both directions, and it carries no opinion about bitcoin either way.
The lag, and the other ways this data misleads
The 45-day window is the trap everyone knows about and still falls into. Positions dated to the end of March arrive in the middle of May. A hedge fund could have exited its whole book weeks before you read the filing. My rule is to use 13Fs for structure and never for timing. They tell you who owns the thing and how the ownership base is shifting, and they tell you nothing about what anyone did last week.
The subtler traps are worth listing too. Filings are a quarter-end snapshot, so a position opened in week two and closed in week eleven never existed as far as EDGAR is concerned, and quarter-end window dressing is a real behavior. Parent and subsidiary managers can both report the same shares, so check the other-managers section of the cover page before you double count. The reporting threshold cuts off the long tail, which means smaller RIAs, arguably the purest adoption signal, are invisible. Initial filings also contain errors often enough that any number that looks insane deserves a check against later amendments before you repeat it. And the shares reported on 13Fs are typically a minority of an ETF's total, with the rest held by retail and by managers under the threshold. Treat the 13F universe as a large sample of institutional behavior, useful for direction and unreliable for totals.
The quarterly routine
Here is the version I run four times a year, in the week after each deadline.
- Calendar the deadlines, roughly mid-February, mid-May, mid-August, and mid-November, and wait a few days past each one so laggards and amendments trickle in.
- Pull every information table containing the CUSIPs of the ETFs you track, from EDGAR directly or through an aggregator.
- Count distinct filers and compare against last quarter, splitting the change into new holders and exits. A flat headline count can hide heavy churn underneath.
- Bucket filers into advisors, institutions, and fast money, then compute each bucket's share of total reported value. The advisor share trending up is the most durable adoption signal this dataset produces.
- Measure concentration, meaning the top ten holders as a share of reported value. A broadening filer count with falling concentration is the healthy pattern. Rising value on a shrinking filer base means a few whales, which is a more fragile story.
- Cross-check the Commitments of Traders report to estimate how much of the fast-money bucket is basis trade before crediting it as conviction.
- Write your read down in one paragraph before looking at price, and compare it against last quarter's paragraph. That comparison is where the signal actually lives.
The failure mode this routine protects against is narrative whiplash. One quarter the headlines celebrate a famous fund's big new position, the next quarter the position is gone and the same outlets announce that institutions are dumping, and both headlines described a single arbitrage trade opening and closing. With the buckets and churn numbers in front of you, you get to skip that cycle entirely.
We built the disclosure-tracking side of Blockcircle on a similar premise, that lagged filings, whether congressional trades or 13Fs, stay useful as long as you respect what the lag allows them to say. A 13F cannot tell you what a fund did yesterday. Four of them in a row can tell you whether the ownership base of the bitcoin ETFs is getting broader, stickier, and more boring, and boring ownership is roughly what maturing adoption looks like. That trend moves slowly enough that reading it 45 days late costs you almost nothing.