Key Takeaways
- What it is: The supply held in addresses attributed to exchanges at a point in time.
- How it is built: Reserve metrics sum balances across known hot, warm, cold, deposit, and operational addresses assigned to exchange entities.
- Core expression: Attributed exchange reserve = sum of eligible balances controlled by the exchange cluster, adjusted for known liabilities only when separate evidence exists.
- Best use: Reserve trends can show custody migration and available on-chain inventory but do not prove solvency because liabilities are not visible from assets alone.
- Main limitation: Unknown addresses create undercounting, omnibus custody creates ambiguity, borrowed assets inflate balances, and off-chain liabilities remain hidden.
- Practical rule: Do not act on a headline reserve figure until unlabeled addresses, omnibus custody, and potentially borrowed (non-owned) balances have been accounted for.
Who This Guide Is For
Exchange reserves is the stock measure: how much of an asset sits at addresses attributed to exchanges right now. Its companion metric, inflows and outflows, is the flow measure covering what moved in and out. The two are related but answer different questions and fail in different ways.
Read this page if you use reserve levels as a supply-side indicator. The construction section covers the label dependency, which is the single largest source of error, and the pitfalls section covers why a falling reserve does not mean what it is usually said to mean.
Educational content. Not individualized financial advice.
What Does Exchange Reserves Measure?
Exchange reserves measures the total balance held at addresses that a data provider has attributed to exchanges. It is a stock: a level at a point in time, not a rate.
The reasoning behind its popularity is that coins on an exchange are readily sellable and coins in self-custody are not, so the reserve is sometimes described as available sell-side supply. The logic is reasonable and the measurement has two structural weaknesses. The attribution is an inference rather than an observation, and the mapping from custody location to intent is weak.
Plain-language definition
Exchange reserves is how many coins sit in wallets somebody believes belong to exchanges.
Technical definition
Given a label set E of addresses attributed to exchanges, reserves at block h is the sum of balances of all addresses in E as of h. Both terms carry weight: the balance is chain state and exact, and E is a private dataset maintained by whoever published the series.
How Is Exchange Reserves Constructed?
The balance half of the calculation is trivial and exact. The attribution half is the entire methodology.
How exchange addresses get identified
- Disclosure. Some exchanges publish addresses, particularly for proof-of-reserve attestations. This is the most reliable source and covers a minority of the total.
- Deposit-address clustering. An analyst deposits to an exchange, observes where the funds are swept, and identifies the consolidation wallet. Repeating this maps deposit addresses to the exchange.
- Common-input heuristics. On UTXO chains, addresses spent together in one transaction are usually controlled by one entity, which propagates a known label across a cluster.
- Behavioral patterns. Withdrawal batching signatures and characteristic sweep timing suggest exchange operation without proving it.
Why the label set is always incomplete and always changing
Exchanges create new addresses continuously, rotate cold storage, and change custody arrangements without announcement. A label set therefore lags reality, and the lag is not uniform across exchanges. When a provider discovers a previously unknown cold wallet holding a large balance, reserves jump on the day of discovery, and that jump is a labeling event rather than a deposit.
Some providers apply the correction retroactively, rewriting history so the balance was always included. Others apply it going forward, leaving a step in the series. Both are defensible; the two produce visibly different charts from identical chain data, and neither usually annotates the change.
Formula and Measurement Logic
Exchange reserves(h) = Σ balance(a, h) for all a ∈ E
whereEis the provider's set of addresses attributed to exchanges.
Every property worth knowing is a property of E rather than of the arithmetic. Two things follow directly.
First, reserves is only comparable to itself when E is held constant. A change in the label set changes the series independently of any coin movement, which means a reserve chart is a joint measurement of exchange balances and of how well the provider has mapped them.
Second, the derivative of reserves is not the same as netflow. Reserves can change through deposits, withdrawals, internal transfers between labeled and unlabeled addresses of the same exchange, and label revisions. Only the first two are flows in any economic sense.
| Cause of a reserve change | Economic content | Detectable how |
|---|---|---|
| User deposit or withdrawal | Yes | Matching flow on the counterparty side |
| Internal wallet reorganization | No | Offsetting move within the same cluster |
| Newly labeled address | No | Step change with no matching transfer |
| Custody provider change | No | Large single move to an unlabeled address |
| Retroactive label revision | No | Only by comparing vintages |
How Should Exchange Reserves Be Interpreted?
The standard reading is that falling reserves are bullish because sellable supply is leaving exchanges. Before accepting it, three alternatives have to be excluded, and at least one of them is usually the actual cause of a large move.
- Custody migration. An exchange moving coins to a third-party custodian, or a large client moving to segregated custody, removes coins from labeled addresses with no change in who owns them or what they intend.
- Label lag. Coins moved to a new exchange wallet that is not yet in the label set read as an outflow.
- Product structure change. Coins moving into a fund or trust structure leave exchange addresses and enter a custodian's, which is a wrapper change rather than a supply withdrawal.
The diagnostic in all three cases is the same: find where the coins went. A genuine withdrawal to self-custody disperses into many addresses. A custody migration goes to one or a few large addresses and sits there. That distinction is visible on chain and is far more informative than the reserve level itself.
The deeper caveat is that the sellable-supply premise is weak in both directions. Coins on an exchange may be collateral for a derivatives position rather than inventory awaiting a sale, and coins in self-custody can be sold in minutes by depositing them. Custody location constrains timing slightly and intent almost not at all.
Step-by-Step Workflow
- Identify the label provider and when the set was last revised.
- Check whether large moves are step changes without matching transfers, which indicates labeling rather than flow.
- Trace where coins went on any significant decline: dispersed to many addresses, or concentrated in a few.
- Separate the aggregate into per-exchange series; one venue can drive an entire aggregate move.
- Check for known custody or product-structure changes over the window.
- Compare against inflow and outflow series, which measure the flow rather than the level.
- Express reserves as a share of circulating supply for any comparison across years.
Worked Hypothetical Scenario
Aggregate exchange reserves fall 180,000 units over one week and commentary describes it as a supply squeeze. Decompose by venue.
| Venue | Reserve change | Destination pattern |
|---|---|---|
| Exchange A | −164,000 | Three transactions to two previously unseen addresses |
| Exchange B | −9,000 | Approximately 4,100 destination addresses |
| Exchange C | −5,000 | Approximately 2,600 destination addresses |
| Exchange D | −2,000 | Approximately 900 destination addresses |
Exchange A accounts for 91 percent of the decline and did it in three transactions to two addresses that then held the balance without further movement. That is the signature of a custody migration or a newly separated cold wallet, not of users withdrawing.
The remaining 16,000 units left exchanges B, C, and D across roughly 7,600 destination addresses, averaging about 2 units each. That is the signature of genuine retail withdrawal to self-custody.
The defensible statement is that around 16,000 units moved to self-custody, roughly one eleventh of the headline figure, while 164,000 units changed custody arrangement without changing hands. Whether the two large destination addresses should be relabeled as exchange-controlled is exactly the question the provider will resolve later, possibly retroactively, at which point this week's chart will change shape.
What Can Make the Interpretation Wrong?
- Label changes read as flows. Discovering or losing an address moves the series with no coin movement, and retroactive corrections change historical values.
- Custody migration read as withdrawal. The largest reserve moves are frequently one entity changing custodian. The destination pattern distinguishes them.
- Aggregate hides venue concentration. One exchange routinely drives an entire aggregate move.
- The sellable-supply premise is weak. Exchange-held coins may be derivatives collateral, and self-custodied coins can be deposited and sold within minutes.
- Derivatives collateral is not spot inventory. On venues offering margin, a share of the balance backs open positions rather than awaiting sale.
- Reserves is not netflow. The level changes for reasons that are not flows at all, so differencing it is not a substitute for measuring flows.
- Unnormalized long-history comparison. Reserves in absolute units across years ignores supply growth; the share of circulating supply is the comparable form.
Cross-Network and Provider Comparison
Label quality, not chain mechanics, determines how reliable this metric is, and label quality varies more by asset than by ledger model.
Assets with a long history and a concentrated set of major venues have relatively mature label sets, because researchers have had years to map them. Newer assets, and assets traded predominantly on venues that do not disclose addresses, have much weaker coverage, and reserve series for them should be treated as indicative at best.
The ledger model does affect the clustering technique available. UTXO chains permit common-input heuristics, which propagate a known label across many addresses efficiently. Account chains lack that heuristic entirely, so exchange identification depends more heavily on deposit-address mapping and disclosure, and coverage is correspondingly patchier for a given research effort.
Assets that exist on several chains introduce a further ambiguity: an exchange may hold the same asset natively on one chain and in bridged form on another, and whether a reserve series aggregates across representations or counts one is a decision that materially changes the total and is often undocumented.
Advanced Analytical Methods
Per-venue decomposition
The aggregate is the least informative form of this metric. Per-exchange series reveal whether a move is broad or driven by one venue, and the latter is far more common than the former.
Destination-pattern analysis
Counting how many distinct addresses received the outflow separates dispersal to self-custody from concentration into a new custodian. This is the highest-value diagnostic available on this metric and is computable from the same transfer data.
Reserves as a share of supply
Expressing the level as a fraction of circulating supply makes multi-year comparison meaningful, which absolute units do not.
Vintage comparison
Because label revisions are applied retroactively by some providers, storing dated retrievals is the only way to detect that a historical reserve series has been rewritten.
Cross-checking against proof of reserves
Where an exchange publishes an attestation, comparing it against the labeled balance measures how complete the label set is for that venue. Doing this for the largest venues bounds the error on the aggregate.
Practical Checklist
- I identified the label provider and its last revision date.
- I checked whether large moves had matching transfers or were step changes.
- I traced destination patterns on any significant decline.
- I decomposed the aggregate by venue.
- I checked for custody or product-structure changes in the window.
- I compared against flow series rather than differencing the level.
- I expressed reserves as a share of circulating supply for multi-year comparison.
- I did not equate exchange custody with intent to sell.
Frequently Asked Questions
How are exchange addresses identified?
Through voluntary disclosure, deposit-address clustering where a researcher deposits and observes where funds are swept, common-input heuristics on UTXO chains, and behavioral pattern matching such as withdrawal batching signatures. Only the first is direct evidence, and it covers a minority of the total.
Can exchange reserves change without any coins moving?
Yes. Discovering a previously unknown cold wallet adds its balance to the series on the day of discovery, which is a labeling event rather than a deposit. Some providers apply such corrections retroactively and rewrite history; others leave a step in the series. Both produce visibly different charts from identical chain data.
Does falling exchange reserves mean supply is leaving the market?
Not necessarily. Custody migration, label lag on new exchange wallets, and coins moving into fund or trust structures all reduce labeled balances without changing who owns the coins or what they intend. The destination pattern distinguishes them: genuine withdrawal disperses across many addresses, custody migration concentrates in a few.
Is exchange reserves the same as netflow?
No. Reserves is a stock and netflow is a flow, and the level changes for reasons that are not flows at all, including internal wallet reorganization and label revision. Differencing the reserve level is therefore not a substitute for measuring deposits and withdrawals directly.
Are coins on an exchange available sell-side supply?
The premise is weaker than it sounds in both directions. On venues offering margin, a share of the balance backs open derivatives positions rather than awaiting a sale. Meanwhile coins in self-custody can be deposited and sold within minutes. Custody location constrains timing slightly and intent almost not at all.
Why should exchange reserves be decomposed by venue?
Because one exchange routinely drives an entire aggregate move. A large aggregate decline is frequently a single venue reorganizing custody in a handful of transactions, while genuine user withdrawal across the other venues accounts for a small fraction of the headline figure.
Why is label coverage worse on account-based chains?
UTXO chains permit the common-input heuristic, where addresses spent together in one transaction are usually controlled by one entity, which propagates a known label across many addresses efficiently. Account chains lack that heuristic entirely, so identification depends more heavily on deposit-address mapping and disclosure.
How do exchange reserves relate to a proof of reserves attestation?
They measure different halves of the same question. An observed reserve balance shows what a labelled address set holds, which is an assets figure with incomplete coverage. A proof of reserves attestation attempts to demonstrate that assets cover customer liabilities, which requires the liability side that no external observer can see. A large observed reserve is therefore consistent with both a fully backed venue and an undercollateralised one, and the attestation's value depends on how the liability side was constructed and who verified it.
What happens to a reserves series when an exchange stops operating?
The series usually does not end cleanly. Balances often remain in labelled addresses for an extended period while proceedings run, so a defunct venue continues to contribute to an aggregate reserve figure as though it were an active market participant. Providers vary in whether and when they exclude it. An aggregate reserve series spanning a venue failure should be checked for whether the affected addresses were removed, kept, or reclassified, because each choice produces a visibly different history.
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.
- Dune: Address Labels: Address labeling and entity context.
- Dune: Curated Data Overview: Cross-chain normalized datasets.
- Etherscan API Documentation: Explorer and API transaction, transfer, balance, and label access.
- Coin Metrics API v4: Metric catalogs, coverage, and data access.
- Coin Metrics: Network Data Glossary: Cross-network address, account, ledger, and UTXO definitions.
- DefiLlama: Data Definitions: TVL, fees, revenue, and holder revenue distinctions.
- DefiLlama: Methodology: TVL, fees, revenue, and volume methods.
- Glassnode: Metric Catalog: Address, supply, valuation, and holder metrics.