Key Takeaways
- What it is: How a token's balance is distributed across addresses at a given block, where the counted unit is the address and the address is not the owner.
- How it is built: Enumerate balances at a block, choose a denominator, apply a dust floor, apply a label set, then compute a statistic. The last four are choices, not observations.
- Core expression: Top-N share = sum of the N largest balances divided by the chosen denominator. Everything difficult is in the words denominator and largest.
- Best use: Comparing one token against itself over time with the label set, dust floor and denominator all held constant.
- Main limitation: Custodial and contract addresses dominate the head, so an unlabeled figure measures custody structure, and address splitting flatters both common statistics.
- Practical rule: Publish the label set version, the dust floor and the denominator beside the number, or the number is not reproducible by anybody including you.
Who This Guide Is For
Read this page if you have looked at a top-holders chart and drawn a conclusion about who owns a token. The chart is usually describing custody arrangements, and the difference matters most for exactly the entries that look most alarming.
This page covers the address-level statistics: top-N share, Gini, Nakamoto and Herfindahl, plus what dust and labeling do to each. Balance-band cohorts and how holder groups behave over time are covered on whale accumulation and distribution. How exchange addresses get identified in the first place is covered on exchange inflows and outflows.
Educational content. Not individualized financial advice.
What Does Token Holder Concentration Measure?
It measures how a token's balance is spread across addresses at a block. The counted unit is the address, and the gap between address and owner runs in both directions at once, which is why a raw figure is so easy to misread.
What sits at the top of the list
On almost any token with real market presence, the largest balances belong to contracts and custodians rather than to individuals. The recurring classes are exchange hot and cold wallets, staking and restaking contracts, lending-protocol pools, automated market maker pools, bridge escrows, vesting and lockup contracts, treasury multisigs, and wrapped-token escrows. A top-ten chart is normally a list of programs and institutions with an individual or two mixed in.
The inversion that matters
A custodial address is the least concentrated entry in the whole list once you ask who owns the balance. One exchange wallet holding nine percent of supply can represent hundreds of thousands of separate customer claims, so treating it as a single large holder inverts the truth about that balance. The reverse mistake exists too: one operator can spread a position across dozens of addresses and produce a distribution that looks healthy. Neither effect is measurable from the balance list alone.
Plain-language definition
How lopsided the list of balances is, before anybody has established what those balances represent.
Technical definition
At block h the ledger yields a vector of balances indexed by address. Every concentration statistic is a functional of that vector. What gets published is that functional applied after a label set has been used to drop or keep certain addresses, after a dust floor has removed the smallest ones, and against a denominator that may be total supply, circulating supply or free float. The vector is the only observation in that list.
How Is Token Holder Concentration Constructed?
- Choose the block. Balances are state, so there is no current without a height. On a token with heavy activity, two heights a few minutes apart give visibly different tails.
- Enumerate holders. On an account chain the balance is a field in contract storage, which makes enumeration mechanical, and contract accounts share the address space with externally owned ones, so the list arrives with programs already in it.
- Handle the UTXO case separately. A UTXO chain has no balance field anywhere. A holder balance has to be assembled by aggregating unspent outputs, and the grouping rule is a wallet heuristic rather than a ledger fact. Change outputs, described in the Bitcoin developer guide on transactions, continually create outputs that look like fresh small holders and are nothing of the sort.
- Choose the denominator. Total supply, circulating supply and free float are three different questions, and the choice is usually invisible in the headline.
- Apply a dust floor and state it. This parameter has no natural value and it moves some statistics a great deal.
- Apply the label set, class by class. For each class of institutional address the options are to exclude it, keep it, or expand it into the holders behind it. Expansion requires off-chain data and is generally unavailable, which is why the honest choice is between the first two.
- Publish the parameters with the number. Label set version, dust floor, denominator and block height. Without all four the figure cannot be reproduced or compared to itself later.
Formula and Measurement Logic
Top-N share = (sum of the N largest balances) / (chosen denominator)
Three other statistics are in common use, and each responds to a different part of the same balance list.
- Gini coefficient. With balances sorted ascending as x_1 through x_n,
G = (2 * sum of i * x_i) / (n * sum of x_i) - (n + 1) / n. It summarizes the entire curve. - Nakamoto coefficient. The smallest k such that the k largest balances together exceed half the denominator. An integer, and it asks only where the halfway line falls.
- Herfindahl index.
H = sum of s_i^2over all holders, where s_i is holder i's share. Squaring makes it dominated by the very largest balances and almost immune to the tail.
| Statistic | Responds most to | Effectively ignores |
|---|---|---|
| Top-N share | The head, and the arbitrary choice of N | The entire tail |
| Gini | The count of near-empty addresses | Nothing, which is the problem on token data |
| Nakamoto | Whether the head crosses half the supply | How the head is split below that point |
| Herfindahl | The very largest balances, squared | The dust tail |
What dust and address splitting do to the numbers
Take a hypothetical five-address distribution A holding 60, 20, 10, 6 and 4 units. Build B by adding 45 addresses holding 0.1 units each, a total of 4.5 units or 4.3 percent of the new supply, without touching any of the original five. Build C by having the holder of the 60 split it across two addresses of 30, which changes nothing about control.
| Distribution | Addresses | Gini | Nakamoto coefficient |
|---|---|---|---|
| A: 60, 20, 10, 6, 4 | 5 | 0.504 | 1 |
| B: A plus 45 addresses of 0.1 | 50 | 0.905 | 1 |
| C: A with the 60 split into two 30s | 6 | 0.353 | 2 |
Dust took Gini from 0.504 to 0.905 while the five real holders kept their exact relative shares. One address split took Gini down to 0.353 and doubled the Nakamoto coefficient, again with nothing changed. Airdrop farming and dust attacks produce the first effect continuously; a single transaction produces the second.
How Should Token Holder Concentration Be Interpreted?
An unlabeled concentration figure is a statement about custody structure, and it becomes a statement about ownership only to the extent that the label set converts it into one.
Concentration is not a risk measure on its own
Forty percent of supply in a staking contract with a withdrawal delay, forty percent in one exchange hot wallet, and forty percent in one unlabeled address are three completely different situations that produce an identical statistic. The first is committed and slow to release, the second is thousands of unrelated claims, and only the third is what people picture when they read the number. A risk statement has to name the class, not the percentage.
Change is more informative than level, but only under a frozen label set
A concentration series moving over time is genuinely useful, but a label set update rewrites it. When a provider discovers that a previously unlabeled address is a custodian, an exclusion-based series steps discontinuously on the day of discovery, and no coin moved. Comparing this month's labels against last month's mixes label drift with real change, indistinguishably.
What would be needed to make it mean ownership
Two things, and only one is achievable. Collapsing an operator's many addresses into one entity is possible with clustering heuristics of varying reliability. Expanding a custodial address into the holders behind it requires records the custodian does not publish. So an ownership-level concentration figure for any token with meaningful custodial holdings does not exist, and a number presented as one rests on an assumption nobody stated.
Step-by-Step Workflow
- Fix the block height and record it.
- Pull the full balance vector, not a truncated top-N list: the tail is needed for anything beyond top-N share.
- Set and disclose a dust floor.
- Apply a label set and record its version and provider.
- Classify every address in the top fifty by hand: custodial, protocol contract, vesting, treasury, market-making pool, or apparently individual.
- Choose the denominator deliberately, and compute the figure under at least two denominators.
- Compute top-N share for several values of N rather than one.
- Compute Gini and Nakamoto together, since either alone can be read the wrong way.
- Recompute Gini at a second dust floor to see how much of it is tail.
- State in the caption which of the top ten are contracts or custodians.
Worked Hypothetical Scenario
This example is hypothetical. Every balance is invented so the arithmetic can be checked.
The raw top ten
A token has 1,000,000,000 units in total supply. Its fifteen largest addresses, with the classification an analyst assigned after inspecting each one, look like this.
| Rank | Balance | Share of supply | Classification |
|---|---|---|---|
| 1 | 120,000,000 | 12.0% | Staking contract |
| 2 | 95,000,000 | 9.5% | Exchange hot wallet |
| 3 | 74,000,000 | 7.4% | Unlabeled |
| 4 | 58,000,000 | 5.8% | Bridge escrow |
| 5 | 41,000,000 | 4.1% | Unlabeled |
| 6 | 33,000,000 | 3.3% | Exchange cold wallet |
| 7 | 27,000,000 | 2.7% | Unlabeled |
| 8 | 22,000,000 | 2.2% | Vesting contract |
| 9 | 18,000,000 | 1.8% | Unlabeled |
| 10 | 15,000,000 | 1.5% | Unlabeled |
| 11 | 13,000,000 | 1.3% | Unlabeled |
| 12 | 11,500,000 | 1.15% | Unlabeled |
| 13 | 10,200,000 | 1.02% | Unlabeled |
| 14 | 9,400,000 | 0.94% | Unlabeled |
| 15 | 8,800,000 | 0.88% | Unlabeled |
The ten largest sum to 503,000,000 units, so the headline is that the top ten addresses hold 50.3 percent of supply. Cumulative balance crosses 500,000,000 only at rank ten, so the Nakamoto coefficient is 10.
Labeling three addresses as custodial
Now mark the exchange hot wallet at rank 2, the bridge escrow at rank 4 and the exchange cold wallet at rank 6 as custodial. Those three hold 95,000,000 plus 58,000,000 plus 33,000,000, or 186,000,000 units. None of them owns anything: the first two hold customer and counterparty balances, and the escrow's economic owners are whoever holds the bridged representation on another chain. Removing them promotes ranks 11, 12 and 13 into the top ten, giving a new top ten of 120.0, 74.0, 41.0, 27.0, 22.0, 18.0, 15.0, 13.0, 11.5 and 10.2 million units, summing to 351,700,000.
| View | Top-ten total | Denominator | Top-ten share |
|---|---|---|---|
| Raw, everything counted | 503,000,000 | 1,000,000,000 | 50.3% |
| Custodial excluded from the numerator only | 351,700,000 | 1,000,000,000 | 35.2% |
| Custodial excluded from both sides | 351,700,000 | 814,000,000 | 43.2% |
| Free float, also excluding staking and vesting | 227,900,000 | 672,000,000 | 33.9% |
Check the third row: 1,000,000,000 minus 186,000,000 leaves 814,000,000, and 351,700,000 divided by 814,000,000 is 43.2 percent. Check the fourth: also removing the 120,000,000 staking contract and the 22,000,000 vesting contract takes the excluded total to 328,000,000 and the denominator to 672,000,000, and the ten largest remaining balances are 74.0, 41.0, 27.0, 18.0, 15.0, 13.0, 11.5, 10.2, 9.4 and 8.8 million, summing to 227,900,000, which is 33.9 percent of 672,000,000.
Four figures, from one token at one block with every balance agreed: 50.3, 35.2, 43.2 and 33.9 percent. The spread is 16.4 percentage points, and every one of the four is a correct answer to the sentence the top ten holders hold X percent of supply. The Nakamoto coefficient behaves the same way. It is 10 on the raw view; under the third view the halfway line is 407,000,000 and the ten largest come to only 351,700,000, so it exceeds 10, and under the free-float view the line is 336,000,000 against 227,900,000 at rank ten, so it exceeds 10 again. Nothing moved on chain in any of this. Somebody made three labeling decisions and one denominator decision.
What Can Make the Interpretation Wrong?
- Address is not owner, in both directions. One custodial address hides many owners, and one owner can occupy many addresses. Neither is visible in the balance vector.
- Contracts dominate the head. Staking, lending, market-making and bridge contracts hold large balances that belong to their depositors, not to the contract.
- Dust inflates Gini. Near-empty addresses drive the statistic upward while telling you nothing, and they accumulate faster on chains with cheap transactions.
- Address splitting flatters everything. One transaction lowers Gini and raises the Nakamoto coefficient with no change in control.
- The denominator is unstated. Total supply, circulating supply and free float give materially different answers from identical balances.
- Label vintage. Comparing across time without freezing the label set mixes label drift with real change.
- Cross-chain double counting. One holder with balances on three chains appears as three holders unless the addresses are consolidated.
- UTXO change outputs. On a UTXO chain, ordinary spending manufactures outputs that read as new small holders.
- Snapshot timing. Two blocks minutes apart give visibly different tails on an active token.
Cross-Network and Provider Comparison
The object being counted is not the same across ledger designs, which makes cross-chain comparison the weakest common use of this metric.
On an account chain a balance is a stored field and enumeration is mechanical, but contract accounts occupy the same namespace as user accounts, so the head of the list arrives full of programs. On a chain where smart accounts are the norm, excluding contracts would remove the users themselves, so the usual exclusion rule inverts. On a UTXO chain there is no balance field at all, holders are heuristic clusters of unspent outputs, and address reuse depends on wallet software, so identical real distributions produce different measured ones.
Multi-chain tokens add a consolidation problem. A holder with balances on three chains appears three times. Where the chains share an address format and a key derivation, consolidating by address string is usually reasonable; where they do not, consolidation is impossible for ordinary accounts. Bridging also mechanically raises measured concentration on the home chain, because the escrow appears there as one enormous holder while the many owners of the representation appear on a chain the statistic is not looking at, a structure covered on bridge flows.
Provider disagreement therefore has very little to do with the chain data. It comes from the label set, the dust floor and the denominator, in that order.
Advanced Analytical Methods
Entity resolution before computation
Collapsing addresses controlled by one operator into a single entity before computing anything removes the address-splitting vulnerability. The heuristics are imperfect and some wallet software defeats them deliberately, so the figure carries its own uncertainty, but it answers a better question.
Publishing the full top-N curve
Reporting top-1, top-10, top-50, top-100 and top-1000 shares instead of one arbitrary N shows the shape of the head. A token where top-10 and top-100 are close has a short head; one where they diverge sharply has a long one, and the two behave differently under stress.
Dust floor sensitivity sweep
Recomputing Gini at several floors and publishing the range converts a fragile scalar into a statement with error bars. A Gini that moves substantially across floors was mostly measuring the tail.
Label vintage comparison
Recomputing last period's figure using this period's label set isolates label drift from real change. The difference between the two is the part of the series that was never about coins moving.
Paired statistics
Publishing Gini alongside Herfindahl exploits their opposite sensitivities: one is dominated by the tail and the other is nearly immune to it, so a large divergence between them localizes where the distribution is unusual.
Practical Checklist
- I recorded the block height the balances were read at.
- I classified every address in the top fifty by hand.
- I stated how many of the top ten are contracts or custodians.
- I set and disclosed a dust floor.
- I recorded the label set provider and version.
- I computed the figure under more than one denominator.
- I computed Gini and the Nakamoto coefficient together.
- I checked whether the same holder appears on more than one chain.
- I described the result as custody structure unless labeling supports an ownership claim.
Frequently Asked Questions
Are the largest token holders usually people?
Rarely. On almost any token with meaningful market presence the head of the distribution is exchange hot and cold wallets, staking or restaking contracts, lending and market-making pool contracts, bridge escrows, vesting contracts and treasury multisigs. A top-ten list is normally a list of programs and institutions. Individual holders who are large enough to appear there do exist, but they are a minority of the entries on most tokens.
Why does the Gini coefficient rise when nothing about ownership changed?
Because Gini responds to the entire distribution including the tail, and token distributions have enormous tails of near-empty addresses. Adding addresses that hold almost nothing raises Gini sharply while leaving every meaningful holder's relative share untouched. Airdrop farming, dust attacks and ordinary wallet churn all manufacture that tail continuously, so the Gini of a token drifts upward over time for reasons that have nothing to do with who owns it.
Why do two providers report different concentration for the same token?
Because the balances are the only part they agree on. The published number is a statistic plus a label set plus a dust floor plus a denominator, and the last three are all choices. Excluding custodial addresses from the numerator only, or from the numerator and the denominator, gives different answers from identical data, and choosing total supply against free float as the denominator changes it again.
Should staking contracts and bridge escrows count as holders?
It depends on the question, and the choice has to be published because it is not a small adjustment. A staking contract holds coins that are committed and cannot be sold until they are withdrawn, which is why excluding it produces a free-float view. A bridge escrow holds coins whose economic owners are the holders of the representation on another chain, so counting it as one holder attributes the whole balance to a contract that owns nothing.
How does the choice of holder threshold change a concentration reading?
Substantially, and it is the most consequential undocumented choice in most concentration figures. Measuring the top ten holders, the top one hundred, or the share held by addresses above a size cutoff produces different pictures of the same distribution, and a token can look concentrated on one measure and dispersed on another. A cutoff denominated in currency also shifts as the price moves, so the same set of holders crosses in and out of the measure without transacting. The threshold has to travel with the number.
Why does measured concentration usually fall as a token ages?
Two mechanical forces push it down regardless of whether ownership genuinely disperses. Early allocations vest and move out of a small number of addresses, and a growing share of holdings migrates into exchange and custodial addresses that are excluded from the calculation, which removes real concentration from view rather than dissolving it. A declining concentration series is therefore consistent with wider ownership and equally consistent with the same owners moving to custody. Distinguishing them requires tracking where the excluded balances went.
Can concentration be measured for a token that exists on several chains?
Only by combining the chains first, which is harder than it sounds. A holder with balances on three networks appears as three separate addresses, so a per-chain measure understates their position and a naive union of chains counts them as three holders. Linking the addresses requires evidence they belong to one party, which the chains do not provide. Bridge escrow contracts add a further complication, since they hold large balances representing many downstream holders. Single-chain concentration figures are the defensible ones.
What does a Lorenz curve show that a single concentration number does not?
The shape of the distribution rather than one summary of it. A Lorenz curve plots the cumulative share of supply against the cumulative share of holders, so it shows whether inequality is driven by a handful of enormous holders, by a broad upper tier, or by a very long tail of tiny balances. Two tokens can share an identical summary coefficient with completely different curves. The curve also makes visible where in the distribution a change occurred, which a single number cannot.
How do dust balances distort holder distribution statistics?
They inflate the holder count with addresses that hold negligible value. Tokens are routinely sent unsolicited to large numbers of addresses, and abandoned addresses retain trivial remainders indefinitely. Any statistic computed over all holders, including a count and any measure sensitive to the number of participants, is dominated by these. Applying a minimum balance filter changes the answer materially, which is why the filter, if any, belongs in the methodology note alongside the figure.
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 tagging, which determines what a concentration figure is measuring.
- Etherscan API Documentation: Token holder and balance lookups used to build the balance vector.
- Ethereum.org: Technical Introduction: Contract accounts sharing the address space with externally owned accounts.
- Bitcoin Developer Guide: Transactions: Outputs and change outputs, and why a UTXO chain has no balance field to read.
- Coin Metrics: Network Data Glossary: Cross-network address, account, ledger and UTXO definitions behind holder counts.