Analysts watch the total held in wallets identified as belonging to exchanges, treating a rise as supply moving toward sale. The logic is reasonable and the measurement is harder than it looks.
The underlying reasoning is straightforward
Assets can only be sold on an exchange once they are on the exchange. Moving coins there is therefore a necessary step for anyone intending to sell.
Conversely, withdrawing to self-custody signals an intention to hold, since it costs a fee and adds friction to any later sale.
Aggregated across many holders, the direction of that flow is treated as a rough measure of whether the market is preparing to sell or to sit still.
Wallet attribution is inference, not fact
Exchanges do not publish their address lists. Analysts identify them from transaction patterns, clustering heuristics and the occasional confirmed address.
When an exchange reorganises its wallet structure, previously tracked addresses go quiet and new ones appear unattributed. The measured reserve falls without a single coin leaving.
These artefacts are common enough that large single-day moves in the metric usually deserve suspicion before interpretation.
Custody and trading balances are mixed together
A modern exchange also runs a custody business, holds collateral for derivatives, and keeps balances for institutional clients who never trade spot.
All of it sits in addresses that look identical from outside. Growth in custody assets registers as growth in reserves, implying selling pressure that does not exist.
The separation cannot be made from chain data alone, so the metric increasingly measures something broader than it did when exchanges were only trading venues.
Derivatives break the link to spot supply
Exposure can now be gained or shed without moving any coins, through perpetual contracts and options that settle in stablecoins.
A holder who wants to reduce risk can short a contract while leaving their coins untouched in cold storage. Reserves do not move and selling pressure exists anyway.
As derivatives volume grew relative to spot, the share of positioning visible in reserve data shrank correspondingly.
The measure works better over long horizons
Day-to-day changes are dominated by operational transfers, wallet rotations and the occasional large mover, none of which carry a market view.
Sustained multi-month trends survive that noise, because operational artefacts do not persist in one direction for long.
Treated as a slow-moving indicator of custody preference rather than a short-term trading signal, it still says something. Treated as a daily input it mostly reports its own measurement error.