Capital Flow
Documentation
What institutional managers are collectively funding, read across a hand-picked watchlist of 96 13F filers at once rather than one manager at a time. Every position change is split into the part caused by trading and the part caused by price — the distinction a value screen hides, and the reason a name can look accumulated when it was actually trimmed into a rally.
What is Capital Flow?
Any institutional manager running more than $100m in US equities must file SEC Form 13Fevery quarter, listing what it held at quarter end. About 5,500 firms do. Reading one of them is easy and mostly useless: a single manager’s book tells you what one person thinks. What is hard to see, and far more interesting, is what a whole population of managers is funding together — and whether they funded it, or simply watched it appreciate.
Capital Flow aggregates a curated watchlist of 96 filers into one view and does two things no free 13F screen does:
Every screen ranks by change in market value, and most of that change is price. Capital Flow ranks on the dollars somebody actually moved and prints the price effect beside it, so an accumulation and a rally never look the same.
Vanguard buying a stock and Baupost buying a stock are not the same event. Discretionary managers, index trackers and dealer inventory are scored separately, and the index cohort doubles as a free market-cap benchmark.
Themes, conviction and hedging are computed across the whole watchlist — how many managers hold a name top-five, how the AI supply chain was funded quarter by quarter, and who is carrying index puts against their book.
The watchlist is hand-maintained, which is a bias. Rather than hide it, the dashboard names every one of the 96 managers, groups them by cohort, and lets you request an addition.
Capital Flow vs. Fund Tracker. They read the same filing and answer opposite questions. Fund Tracker is one manager at a time — open Berkshire, see its book, diff it against last quarter. Capital Flow is all managers at once — open a security or a theme and see what the whole watchlist did to it. If you want to know what Buffett owns, use Fund Tracker. If you want to know who else owns it and whether they were buying, use this.
Where does the data come from?
All data is sourced from SEC Form 13Ffilings pulled from EDGAR, aggregated by a separate pipeline, and published into a dedicated database that only this site’s server reads. Your browser never talks to the database directly. The dashboard reads pre-computed aggregates — nothing is summed in the page.
Editorially, by hand. There is no screen behind the watchlist — it is a judgement about which managers are worth aggregating, grouped into five buckets:
The dashboard never hides this. The header subtitle says the list is hand-picked, the KPI strip links straight to it, and the Coveragetab opens with a section titled “How these managers were chosen”. The roster drawer lists all 96 by cohort, searchable by name or CIK, with a form to request one that is missing — requests are reviewed before the next quarterly ingest.
The data is always in the past, by law.Form 13F is due 45 days after quarter end, so the freshest possible figure describes a portfolio that is already at least six weeks old — and in practice older. The header carries an age badge (“as of 30 Jun 2026 · 85 days old”) for exactly this reason. Nothing here is a real-time position, and no amount of engineering can make it one.
Some managers stop filing, file late, or fall below the threshold. Their positions are carried at the last quarter they reported — stale, not zero, and not current — and every one of them is named on the Coverage tab with how many quarters behind it is. At the time of writing three of the 96 are behind, including one index giant two quarters late, which matters because the index cohort is the benchmark.
Flow versus price effect
This is the product’s whole thesis, so it is worth stating precisely. A position’s market value can change for two unrelated reasons: the manager traded it, or the price moved. Every free 13F screen ranks by the change in value, which conflates the two — and since price usually dominates, those screens are largely ranking price moves and calling them positioning.
└── flow ──┘ └── price effect ──┘
Flow— labelled “traded” in the UI — is the change in share count valued at the current price. It is dollars somebody chose to move. Price effectis the residual: what last quarter’s shares did while nobody touched them. Capital Flow ranks on flow and prints the price effect beside it, as a second segment on the same bar.
Two rows from the active cohort in Q2 2026, both real:
The Rotation tab makes this filterable directly. A segmented control offers All moves, ≥50% traded and ≥80% traded, and when rows are removed the count is printed rather than silently dropped: “n of mrows hidden — their move was mostly price, not trading.”
New listings are not accumulation.When a security becomes 13F-reportable, every holder books it in the same quarter and the flow calculation reads the whole position as a purchase. In Q2 2026 that put SpaceX at the top of the buy list — $99.1bn across 19 funds, all of it “new”. Those rows carry a new listing chip and are excluded from the bar scale, so one listing event cannot flatten every genuine trade on the chart.
The three cohorts
Aggregating Vanguard together with Baupost destroys both signals. Vanguard’s trades are a mechanical consequence of fund inflows; Baupost’s are decisions. Every manager on the watchlist is therefore hand-assigned to one of three populations, and every panel is computed within a single cohort at a time.
Discretionary managers — Berkshire, Baupost, Pershing Square. 57 of the 96, carrying roughly $1.6tn. This is the only cohort whose changes are decisions, and it is the default selection on every tab.
Market-cap trackers — Vanguard, BlackRock, State Street. 16 managers and roughly $19.1tn. Used as the benchmark, never as a signal: because their aggregate weights ARE market-cap weights, active ÷ index gives a true overweight with no external market data.
Market makers and multi-strategy desks — Citadel, Jane Street, Millennium. 23 managers and roughly $6.2tn. Their holdings are largely warehoused customer-flow inventory, so a dealer 'buying' a name usually means a client sold it.
The assignment is by hand, and that is deliberate. Two automatic classification rules were tried and both mis-tagged managers who carry index puts as market makers, which put discretionary hedgers in the dealer bucket and quietly emptied the Hedging tab. A hand-maintained list is a bias you can read; a bad heuristic is a bias you cannot. Anything unlisted defaults to active, and the full assignment is visible in the roster drawer.
The five tabs
Loading the page costs exactly one request. Each panel fetches only when its tab is first opened, so a tab you never look at costs nothing.
The default tab. One sparkline per theme — 17 of them, from 'AI · Compute Silicon' to 'Financials' — over eight quarters, ordered largest first. Each chart is scaled to its OWN range, so compare shapes rather than heights. Every cell prints net dollars bought over the window, plus either a growth multiple, an 'emerged' chip where the base was near zero and a multiple would be meaningless, or the holder count. Themes are assigned by the build and recomputed each quarter; 'Other' is a residual and is always pinned last.
The quarter's accumulated and reduced names, side by side on one shared scale. Every row carries a split bar — coloured for flow, grey for price effect — plus the holder count, the price effect in dollars and the traded share as a percentage. This is where the purity filter and the new-listing chips live.
A leaderboard of securities held as a top-five position by two or more concentrated active managers, ranked breadth × depth. Re-sortable by conviction, overweight, breadth or average weight. Each row shows how many managers hold it top-five, their average and peak weight, and its overweight versus the index cohort.
Managers ranked by index-put notional as a share of their long book. Dealers are excluded by default — 99% of all options notional in the dataset sits with Jane Street, Susquehanna and Citadel, where it is warehoused customer flow — and a checkbox brings them back with a warning attached. In Q2 2026 Elliott's index puts read 58% of a $14.3bn long book.
Never hidden, and carries a badge with the stale-filer count. Explains how the managers were chosen, names every filer that has not reported for the latest quarter and how far behind it is, and breaks the tracked value down across the three cohorts. If you read one tab before trusting the others, read this one.
Above the tabs sits a five-tile KPI strip — managers on the watchlist, how many are active, how many are concentrated, securities tracked and quarters of comparable history — and the cohort selector, which applies to every panel below it.
Every panel prints its own formula. Rotation says flow = Δshares × price_now; Conviction says score = funds holding top-5 × average weight; Hedging says index puts ÷ long-book value. If a figure ever looks wrong, the arithmetic that produced it is on screen beside it.
What the numbers mean
Every figure below is pre-computed in the warehouse, not derived in your browser — which is why totals do not shift as you filter or page.
Δshares × price_now. Dollars somebody chose to move. Always printed signed, so +$1.2bn and −$340m are never confused.shares_prev × Δprice, equivalently Δvalue − flow. The part that happened while nobody did anything.|traded| ÷ (|traded| + |price|), 0–100%. Rendered as “64% of move was trading”. Prints an em dash rather than a fabricated 0% when there is nothing to attribute.funds holding top-5 × average top-5 weight — breadth multiplied by depth, not a holder count. A security qualifies only if it is a top-five position in two or more concentrated active funds, which is why the leaderboard is short: 24 securities clear the bar.active cohort weight ÷ index cohort weight. Above 2× earns an overweight chip, below 0.5× an underweight one. Where the index cohort holds none of it the value is null, displayed as no benchmark — never infinity — and sorted last.A worked example of why overweight is worth reading. Nvidia is the most famous name in the AI trade. Its conviction row shows five concentrated managers holding it top-five at an average 6.4% weight — and an overweight of 0.22×. Relative to market-cap weight, the concentrated active cohort is materially underweight it. That is not visible in any list of who owns Nvidia.
Drill-through drawers
Almost everything on the dashboard is clickable, and clicks open a drawer rather than navigating away — a right-edge panel on desktop, a bottom sheet under 700px. Drawers stack: open a theme, click one of its constituents, click that security’s theme chip, and a breadcrumb records the path. Esc pops one level rather than throwing the whole trail away.
Opened from Rotation, Conviction or a theme's constituent list. The cohort strip across the top is the point of it: all three cohorts side by side, each showing position value and dollars traded, and doubling as the chart selector. Below that, position value over time, a traded-versus-price bar chart, every quarter on file as a table, and the concentrated-conviction block where one exists.
The sparkline redrawn full size, with theme value, net traded, managers holding at least one name and securities above the size floor. 'What is in it' lists the constituents for the latest quarter, rankable by position value or dollars traded, each row clickable. If the server cap was hit, the drawer says so rather than implying the list is complete.
Opened from Hedging and from the stale rows on Coverage. Long book, position count, top-5 concentration, index puts — then the same figures charted over every quarter on file, including an index-put hedge chart that only renders when there is a hedge to show.
Opened from the KPI strip or the Coverage tab. All 96 managers grouped Active → Index → Dealer, largest book first within a cohort — never alphabetical, never by size alone. Searchable by name or CIK, with a form at the bottom to request a manager that is missing.
What the cohort strip is for. In Q3 2024 the active cohort sold $63.7m of Micron while the index cohort bought $10.5bn of it. Aggregated together those two facts cancel into noise. Side by side they are a disagreement between discretion and mechanics, which is the only reason the cohort split exists.
Limitations & known gaps
None of these are bugs. Each is a property of Form 13F or of a deliberate design decision, and each is stated on the dashboard as well as here.
96 managers out of roughly 5,500 filers, chosen by hand. A theme 'the market is funding' is really a theme this watchlist is funding. Widening the list would change every aggregate on the page.
Form 13F is due 45 days after quarter end. The dashboard describes positions as of the quarter-end date, not today, and the header prints how old that is. A manager may have exited a position months before you read about it.
13F does not report short stock. A manager who is net flat through a short book reads here as fully long. The Hedging tab is the closest available proxy and only covers listed index puts.
13F carries no strike and no expiry, so a far out-of-the-money put and an at-the-money one are indistinguishable. Index-put notional at 58% of a long book is not a 58% hedge.
13F reports US-listed equities and certain options above a threshold. Bonds, cash, private holdings and foreign listings are absent, so the 'long book' is smaller than the manager's real book.
Theme assignment is algorithmic and recomputed each quarter; cohort assignment is by hand. Both are printed rather than hidden, but neither is an industry standard, and 'Other' is a residual rather than a category.
A manager can amend a past filing, which changes figures you have already read. The build re-aggregates on each sync, so a quarter's numbers are not frozen once published.
Three managers are currently behind. Their last-reported positions still count toward every aggregate, because dropping them would be a worse lie than carrying them — but they are not current, and Coverage names each one.
Original filings are public. Look any manager up on SEC EDGAR by name or CIK and compare — the roster drawer prints the CIK for every one of the 96.
Disclaimer
Informational purposes only. Capital Flow is a research tool built on public regulatory filings. Nothing on it is investment advice, a recommendation, or an offer to buy or sell any security. Institutional positioning is not a trading signal, and following a manager into a position you learn about three months late is not a strategy.
Themes, cohort labels, conviction scores and flow attributions are algorithmic outputs whose formulas are printed beside them on the dashboard. They are our interpretation of the filings, not facts asserted by the filers. Verify anything that matters against the original filing on SEC EDGAR.
Rock Group · Capital Flow Documentation · Last updated September 2026
Questions? Contact us