Key Takeaways
- What it is: Balance changes at blockchain addresses attributed to a regulated fund, its custodian, or a trust structure.
- How it is built: Attribute addresses to the custodian, read balance changes over time, then try to match them against the issuer's published creation and redemption activity.
- Core expression: Observed balance change = settled fund creations minus settled redemptions, plus movement by every other client sharing the address, plus custody reorganization, plus label revision.
- Best use: Custody questions the issuer does not publish, such as address concentration, sub-custody, staking, and whether the holding is native or a wrapped representation.
- Main limitation: Settlement lag and batching break the timing, omnibus custody breaks the attribution, and the issuer already publishes the flow figure more accurately than the chain can reconstruct it.
- Practical rule: Take the flow number from the issuer's own daily disclosure. Use the chain for the questions that disclosure does not answer.
Who This Guide Is For
Read this if you have tried to read institutional demand off a custodian's wallet, or have seen a large transfer into a known fund address reported as evidence that money moved that day. Both readings assume a directness that the settlement chain does not have.
Educational content. Not individualized financial advice.
What Does Crypto ETF On-Chain Flows Measure?
It measures the balance held at addresses somebody believes belong to a custodian, and how that balance changed. The economic event a reader usually has in mind, an investor buying into the product, sits three steps away from anything the chain records.
- An investor buys shares from another investor on the secondary market. No asset moves anywhere.
- Persistent buying pushes the share price away from net asset value, and an authorized participant closes the gap by creating new shares in the primary market.
- The creation is agreed on a trade date, priced against that day's valuation, and settles on a later date under the product's settlement cycle.
- At settlement, and only then, assets reach the custodian and the on-chain balance moves.
The first step is the one that dominates daily activity and it never touches a blockchain. The chain is blind to that entire market by design, not by measurement failure.
Plain-language definition
How many coins sit at wallets somebody believes hold a fund's assets, and how that number changed since last time.
Technical definition
For an attributed address set A and window W, the observed quantity is the net change in balances across A during W. That single number is a sum of several unrelated processes.
How Is Crypto ETF On-Chain Flows Constructed?
Construction means attributing a set of addresses to a custodian and then interpreting its balance changes against a settlement process that happens off chain.
Why the same economic flow leaves three different traces
The creation mechanism decides what the chain sees, and two products with identical inflows can look nothing alike. Take a single creation worth 1,200 asset units.
| Creation mechanism | What the authorized participant delivers | On-chain trace |
|---|---|---|
| Cash create | Cash, which the fund's agent uses to buy the asset | Several transfers arriving from trading-venue or dealing-desk addresses, spread over one or two days |
| In-kind create, participant custodies elsewhere | The asset itself | One transfer from the participant's address into the fund's address |
| In-kind create, participant custodies at the same firm | The asset itself | None. The custodian reallocates on its own books |
The third row is the one that breaks naive analysis. The economic flow occurred, the fund gained 1,200 units, and the omnibus address holding them did not change balance by a single satoshi or wei, because the assets never left the building.
The first row causes a different problem. A cash create's purchases are withdrawals from trading venues, so the same economic event also appears in an exchange outflow series. Reading both metrics without noticing the link counts one purchase twice.
Omnibus custody and sub-custody
A custodian may hold assets for several funds plus clients that are not funds at all in one omnibus address, or may segregate each client into its own addresses. Nothing in the chain data distinguishes the two arrangements. If the account is omnibus, attributing its entire balance to one product is simply wrong. If it is segregated, you still cannot be confident you have found every address belonging to that client.
Sub-custody adds another layer. A custodian may place part of a holding with a second firm, which moves assets to addresses attributed to a different entity or to no entity at all, and the transfer looks exactly like an outflow from the fund.
Formula and Measurement Logic
Observed balance change = settled fund flow + other clients in the same address + custody reorganization + label-set revision
Only the left-hand side is observed. The right-hand side has four terms, three of which are usually unknown, so the identity is a decomposition problem rather than a calculation.
| Term | What makes it non-zero | Can it be isolated? |
|---|---|---|
| Settled fund flow | Creations and redemptions whose settlement date falls inside the window | No, only jointly with the rest |
| Other clients | Any non-fund client of an omnibus address moving assets | No |
| Custody reorganization | Consolidation, address rotation, sub-custody placement | Sometimes, by tracing the counterparty |
| Label-set revision | Addresses added to or removed from the attributed set | Only by storing dated vintages |
Why even a perfect decomposition is the wrong number
Settled fund flow is not reported fund flow. A creation agreed on Monday and settled on Wednesday belongs to Monday in the issuer's disclosure and to Wednesday on the chain. Over a long enough window the two converge, but at daily frequency they are different series with a shifting offset.
How Should Crypto ETF On-Chain Flows Be Interpreted?
Interpret it as custody evidence, not as flow evidence. For the flow question the issuer publishes shares outstanding daily, computed from its own books, with no attribution problem and no settlement lag, which makes the chain the strictly worse source for the one thing it is most often used for.
What the chain answers better than the disclosure
- Concentration. Whether a product's assets sit in one address or many, and how large the largest single point of failure is.
- Custody changes. Whether assets moved to a different firm, which a monthly report may not surface for weeks.
- Representation. Whether the fund holds the native asset or a bridged or wrapped version of it, which changes the risk profile materially.
- Staking posture. Whether assets sit idle, are delegated, or are locked in a validator deposit contract.
- Address hygiene. How often the custodian rotates addresses and whether balances are consolidated or fragmented.
Why daily correlation between the two series is a trap
Batching and settlement lag together resample the flow series onto a different, irregular clock. Correlating the daily on-chain change against the daily reported flow therefore measures the settlement calendar and the custodian's batching habits far more than it measures demand. Compare cumulative totals over windows several times longer than the settlement cycle.
Step-by-Step Workflow
- Take the issuer's own creation and redemption or shares-outstanding series as the flow benchmark.
- Establish the settlement cycle and whether the product creates in cash or in kind.
- Identify the attributed address set and whether the custodian is known to use omnibus accounts.
- Shift the reported series onto settlement dates before comparing anything.
- Compare cumulative totals over a window several times the settlement lag, never day against day.
- Trace the counterparty of any large unexplained move before treating it as fund flow.
- Check whether the holding is the native asset or a wrapped representation.
- Report the residual explicitly instead of assigning it to the fund.
Worked Hypothetical Scenario
This example is hypothetical and every figure is invented for the arithmetic. Fund F creates in blocks of 400 asset units, settles two business days after trade date, and its custodian holds the assets in an omnibus address shared with other clients.
| Day | Reported primary-market flow | Settles on | On-chain change at the omnibus address |
|---|---|---|---|
| Monday | +1,200 | Wednesday | 0 |
| Tuesday | +400 | Thursday | 0 |
| Wednesday | −800 | Friday | +1,200 |
| Thursday | +800 | Following Monday | +1,300 |
| Friday | 0 | Not applicable | −800 |
| Week total | +1,600 | +1,700 |
The weekly totals nearly agree. Reported flow is 1,600 units and the observed balance change is 1,700, a gap of 100 or 6.25 percent, which is close enough that a reconciliation done at weekly granularity would be signed off.
The agreement is a coincidence between two errors pointing opposite ways. Only 800 units of the observed change came from this fund, because Thursday's 800 unit creation had not settled by Friday: 1,200 plus 400 minus 800 is 800, exactly half the reported figure. The remaining 900 units arrived on Thursday and belong to an unrelated client of the same omnibus address, which is why Thursday shows 1,300 rather than 400.
Daily correspondence is worse than the weekly picture suggests. Thursday's 1,300 matches no reported figure at all: Tuesday's flow was 400 and Thursday's was 800, and neither equals 1,300. A correlation run on these two daily series would be measuring the settlement calendar plus a stranger's deposit.
Nothing here is a data error. The two series are answers to different questions, and the resemblance between their weekly totals is the most misleading thing in the table.
What Can Make the Interpretation Wrong?
- Settlement lag. The on-chain event dates from settlement, not from the trade that caused it, so every daily comparison is offset by a cycle that varies with weekends and holidays.
- Batching. A custodian may settle several days of activity in one movement, so an on-chain series is lumpier than the flow it represents and its zero days are meaningless.
- Omnibus commingling. A shared address mixes unrelated clients into one balance, and no chain evidence separates them.
- In-kind creates that never touch the chain. When the participant already custodies at the same firm, the transfer is a book entry and the observed balance does not move at all.
- Double counting against exchange metrics. A cash create's purchases register as exchange outflows, so the same purchase can be counted once as a fund inflow and once as coins leaving exchanges.
- Custody reorganization. Consolidation, address rotation, and sub-custody placement all move balances without any investor doing anything.
- The secondary market is invisible. Most share trading nets internally and produces no primary-market activity, so heavy volume can coincide with a flat chain.
- Staked products. Assets locked in a validator deposit contract are a claim subject to activation and exit queues, not a spendable balance, so a redemption cannot settle on demand.
Cross-Network and Provider Comparison
Reading a balance is a different job on each ledger model, and custody workloads sit exactly where the differences bite.
On a UTXO chain the balance is a sum over unspent outputs, and a custodian holding a large position holds it across many of them. Periodic consolidation produces very large transactions with change outputs returning to the same owner, which a parser that treats every output as a receipt will record as an inflow. The Bitcoin Developer Guide: Transactions describes the output structure behind the ambiguity.
On an account chain a native balance is a single field and easy to read, but a token holding lives in a contract's storage rather than in the account, and a bridged or wrapped representation is a different asset with different risk from the native one.
Staked products change the object entirely. Assets committed through a validator deposit contract are subject to activation and exit queues described in the Ethereum.org: Proof of Stake documentation, so the balance is illiquid for a period the custodian does not control.
Providers differ on whether they publish entity-level balance series for fund labels or leave you to assemble the address set. Either way its completeness is unknowable, and a newly attributed address produces a balance step that is a labeling event rather than a flow.
Advanced Analytical Methods
Lag-fitted reconciliation
Rather than assuming a settlement lag, fit it. Shift the reported flow series by a range of candidate lags, compare cumulative totals against the on-chain series at each, and keep the lag that minimizes the residual. The fitted lag is informative in itself, and the residual it leaves is the honest estimate of everything the fund did not cause.
Creation-unit consistency
Every creation settling on a given day converts at the same per-unit asset amount, so that day's genuine fund settlement should be close to an integer multiple of it. A movement that is not a plausible multiple is unlikely to be fund flow, which makes this a cheap filter for omnibus contamination.
Inbound counterparty typing
Classify what sent each inbound transfer. Arrivals from trading-venue addresses indicate a cash create being filled in the market, while a single arrival from one persistent counterparty indicates an in-kind delivery. The mix tells you which mechanism the product actually uses.
Zero-day analysis
Days on which the address does not move are as informative as days it does. A long run of zeros against reported creations is strong evidence of either batching or in-kind creates settling inside the custodian.
Practical Checklist
- I am using the issuer's disclosure for the flow question, not the chain.
- I know the settlement cycle and whether creates are cash or in kind.
- I checked whether the address is omnibus or segregated, and said so.
- I aligned the reported series to settlement dates before comparing.
- I compared cumulative totals, not daily values.
- I traced the counterparty of every large unexplained movement.
- I confirmed whether the holding is native or a wrapped representation.
- I reported the residual rather than attributing it to the fund.
Frequently Asked Questions
Why does a custodian balance lag the fund flow it is supposed to show?
Because the flow is agreed in the primary market on a trade date and the assets move at settlement, which falls a set number of business days later. The issuer books the creation on the trade date and the chain records it on the settlement date, so the two series describe the same events on different clocks. Custodians also batch, which delays and lumps the on-chain half further.
What is omnibus custody and why does it break attribution?
An omnibus address holds assets for several clients at once, which may include multiple funds and clients that are not funds at all. The chain shows one balance and one set of movements with no client identifier attached. Attributing the whole balance, or any particular movement, to one product is therefore an assumption rather than an observation.
Can a fund creation happen with no blockchain transaction at all?
Yes. In an in-kind create the authorized participant delivers the asset rather than cash, and if that participant already custodies with the same firm the delivery is a reallocation on the custodian's internal books. The fund gains the assets and the omnibus address does not change balance, so the entire flow is invisible on chain.
Does heavy share trading show up on chain?
Usually not. Most secondary-market volume matches buyers against sellers without touching the fund, and only the residual imbalance is resolved through a creation or redemption. A day of very heavy trading can therefore produce no primary-market activity and no on-chain movement whatsoever.
Can the same purchase be counted twice across two metrics?
Yes, and it is a common error. A cash create is filled by buying in the market, so the assets leave trading venues and appear as exchange outflows before arriving at the custodian. Reading an exchange outflow series and a fund inflow series together without noticing the connection counts one purchase as two independent pieces of evidence.
What is the difference between an in-kind and a cash creation for an on-chain observer?
An in-kind creation delivers the asset itself to the fund, so a corresponding movement into custody can appear on chain. A cash creation delivers currency, and the fund or its agent buys the asset afterwards, possibly through a venue whose internal ledger records the purchase with no on-chain transaction until the coins are later moved into custody. The same economic event therefore produces a visible on-chain footprint in one case and a delayed or absent one in the other, which is a structural reason on-chain flows and reported fund flows diverge.
Why does a custodian move coins between its own addresses?
Routine operations account for most large custodial movements. Custodians rotate addresses, split balances across storage tiers, consolidate small holdings, migrate to new key management arrangements and rebalance between hot and cold storage. None of these represent a client depositing or withdrawing anything. An observer watching a labelled custodian address without knowing the rest of that custodian's address set will read internal housekeeping as a flow, and the error is largest exactly when the amounts are largest.
Can a fund's published holdings be reconciled against an on-chain address?
Only when the fund or its custodian publishes the addresses, and even then the reconciliation is partial. Some funds disclose addresses, which allows a direct comparison between a reported holding and an on-chain balance at a stated time. Many do not, and a custodian holding assets for several funds in shared addresses makes the split unobservable regardless. Where a reconciliation is possible, differences in the as-of timestamp between the fund's report and the block height chosen will produce a gap that is not an error.
What lag sits between a fund's reported flow and the matching on-chain movement?
Several steps sit between them, each adding time. A creation or redemption is agreed, settled in cash or in kind under the fund's own settlement cycle, and only then results in assets moving into or out of custody, which may itself be batched. The fund's published figure follows its own reporting schedule. The consequence is that an on-chain movement and a reported flow describing the same underlying event carry different dates, so aligning the two series day by day produces mismatches that are procedural rather than substantive.
References
These sources should be reviewed during editorial verification. They support data structures and methods, not the hypothetical conclusion. Provider formulas, chain rules, and APIs can change. Confirm current documentation before publication.
- Ethereum.org: Proof of Stake: validator deposits, activation and exit queues that make a staked custody balance illiquid.
- Bitcoin Developer Guide: Transactions: unspent-output balances, consolidation, and change outputs in custody wallets.
- Dune: Address Labels: how custodian and fund addresses are attributed to entities.
- Etherscan API Documentation: address balance and transfer lookups used to trace a custody counterparty.
- Coin Metrics: Network Data Glossary: cross-network address, account, and balance definitions.
- Glassnode: Metric Catalog: published entity-balance metric definitions. Editorial review should confirm that every provider definition remains current on the publication date.