Live prices are currently unavailable — the exchange feed could not be reached and no recent cached reading is held.

Bitcoin Data GuideBitcoin Reference

On-Chain Analytics

Exchange Flows

Exchange flows rest on a simple claim: coins moving to an exchange are being prepared for sale, and coins leaving one are being withdrawn for holding. The claim is plausible and sometimes useful, but it depends on address attribution that is never certain.

Last reviewed 2026-09-21Source: Coin Metrics exchange flow methodology; address attribution researchNo figure is quoted here. The metric depends on address attribution, which is an inference rather than a fact.

The claim

The measure begins by identifying a set of addresses that belong to exchanges. It then tracks the coins moving into and out of those addresses, and reports the net flow. A net inflow is read as supply arriving on venues where it can be sold, and a net outflow is read as supply leaving for storage. The series is often presented as a leading indicator of selling pressure.

The reasoning is not unreasonable. An exchange is where a sale happens, so coins that are about to be sold do have to reach one. The difficulty is that the reverse is not true: coins reaching an exchange are not necessarily about to be sold. They may be moved for custody reasons, for a transfer between a user's own accounts, or as part of the exchange's internal operations.

The measure is therefore a statement about the movement of coins to and from a set of addresses, with an interpretation layered on top. The movement is a fact about the ledger. The interpretation is an inference, and it is where the measure is most often overread.

How the addresses are identified

Exchange addresses are identified by a combination of methods. Some are disclosed by the exchange itself, either deliberately or through a proof-of-reserves procedure. Some are inferred from the pattern of transactions: an address that receives from many unrelated sources and consolidates into a small number of outputs looks like a deposit address. Some are identified from public sources such as block explorers that label addresses.

Each method has a failure mode. A disclosed address may be retired and replaced without notice. An inferred address may belong to a payment processor or a custodial service rather than an exchange. A labelled address may be mislabelled, and the error propagates into every metric built on it. The attribution is a best effort, and it is the part of the measure that is least verifiable.

The clustering heuristics that support the inference are covered on the tracing page. They rest on assumptions about how wallets construct transactions, and those assumptions are exactly what a careful wallet design can defeat.

Where the inference breaks down

The most common distortion is internal shuffling. When an exchange moves coins between its own wallets, the coins appear to leave one address and arrive at another, and a metric that treats the two as distinct will record a flow that did not happen. A large reshuffle can dominate the series for a period and produce a reading that has nothing to do with holder behaviour.

A second distortion is a change in custody practice. If an exchange moves its holdings to a new custody arrangement, the old addresses stop receiving and the new ones start, and the series shows a discontinuity that reflects an operational decision rather than a change in the market. A reader who does not know about the change will interpret the discontinuity as a flow.

A third is the growth of off-chain settlement. Coins that move between users on an exchange's internal ledger never touch the blockchain, so a sale can occur without any on-chain movement at all. The measure sees only the on-chain portion of exchange activity, which is a shrinking share of the total as internal settlement grows.

Sources and references

The description of the measure and its attribution methods is taken from the published methodology of the providers that compute it, and from the academic work on address clustering.

  • The flow metric and its methodology. Glassnode, Exchange Flow Metrics: documents the address sets, the inflow and outflow definitions and the net-flow calculation.
  • The clustering heuristics behind address attribution. Meiklejohn et al., A Fistful of Bitcoins: characterises the address clustering that exchange attribution depends on.
  • The change-output heuristic and its limits. Bitcoin, Bitcoin Developer Guide — Transactions: describes how change outputs are created, which is the basis for one of the most common clustering assumptions.

Related reading