A new Bank for International Settlements study argues that one of crypto's most quoted numbers — how much value moves on the Bitcoin network — depends less on the blockchain than on the analyst reading it. Change the counting rule, and the same ledger yields a figure up to six times smaller.
Why can Bitcoin transfer volume estimates differ by 6x?
Bitcoin on-chain transfer estimates can diverge by a factor of six because the measurement methodology, not the ledger, decides what counts as an economic transfer. BIS Working Paper No. 1377, published on 15 September 2026, finds that excluding self-transfers and change outputs can cut measured on-chain flows by up to six times with no change to the underlying data . The unadjusted figure is the one most widely reported.
The root cause is Bitcoin's UTXO (unspent transaction output) model, which consumes whole outputs rather than debiting account balances. The authors' worked example: spending a 4 BTC output to send 1.5 BTC creates a 1.5 BTC recipient output plus a 2.5 BTC change output returning to the sender . Sum every output and the transaction reads as 4 BTC moved, though only 1.5 BTC changed hands.
The scope is not a sample. The Bitcoin dataset spans 2009 through 2026, covering roughly 1.3 billion transactions and about 3.6 billion outputs, drawn from some 100 billion records across Bitcoin, Ethereum and Tron via Mercurius — a platform operated by De Nederlandsche Bank and developed with the BIS Innovation Hub and Deutsche Bundesbank .
"Decentralised finance data presents a distinctive paradox: while every data point is publicly recorded and accessible, deriving meaningful insights is obscured by the scale, fragmentation and complexity of the ecosystem," write Timothy Aerts, Ronald Heijmans, Jan Paulick and Violeta Vuletic (source: BIS Working Paper 1377, 2026-09).
One distinction matters before reading further: the finding concerns on-chain transfer value, not exchange trading volume .
What changed: BIS's three estimators, from upper bound to conservative floor
The BIS builds three nested measures of Bitcoin on-chain transfer value from the same ledger, and the spread between them is where the sixfold gap comes from . Each estimator makes a different assumption about change outputs — the remainder a wallet returns to itself when it spends a UTXO. None of them changes a single byte of blockchain data.
| Estimator | Method | Assumption about change |
|---|---|---|
| 1. Upper bound | Sums the value of all transaction outputs | Assumes zero change; this is the figure most widely reported as "transfer value" |
| 2. Adjusted | Removes self-transfers — outputs returned to a sending address | Treats identifiable self-transfers as change |
| 3. Conservative | Removes identified self-transfers; where none is visible, subtracts the transaction's largest output | Assumes the smaller output is the genuine economic transfer |
Excluding self-transfers alone can cut measured on-chain flows by up to a factor of six . That matters because estimator 1 is the version that circulates most freely in dashboards and headlines (source: BIS, 2026-09).
Two timing details deserve attention:
- March 2016 onward: the three series began diverging markedly, which the paper links to rising address reuse and shifting user and service-provider behaviour .
- High-activity windows: the estimators separate most sharply during periods of elevated Bitcoin activity and rising prices — the exact stretches analysts quote most often .
The authors are explicit that estimator 3 is a conservative heuristic, not a mathematically proven floor. CoinJoin transactions can carry several legitimate change outputs, and mixers, spam, intermediary transfers and exchange-internal movements all introduce classification error the model does not attempt to strip out (source: crypto.news, 2026-09). The claim is methodological: the spread is specific to these three estimators, not a verdict that any single published number is wrong by six times.
Why it matters: market cap, dormancy and the metrics traders actually watch
The same measurement problem sits inside Bitcoin's valuation metrics. Conventional market capitalisation multiplies total supply by the current price, while realised capitalisation values each coin at the price it last moved at — and BIS calculates that simple market cap has reached up to four times realised cap during price surges . The gap widens exactly when traders are quoting the number most.
Dormant supply compounds the distortion. Excluding outputs untouched for more than 15 years removes roughly 1.8 million BTC from the market-cap calculation altogether . The paper's dormancy tiers:
- ~3.5 million BTC unmoved for more than 10 years
- ~1.8 million BTC unmoved for more than 15 years
- ~24,000 BTC later spent after more than 10 years dormant; almost 3,000 BTC after more than 15
That last line is the caution. BIS treats dormancy as a proxy for loss, not proof of it — coins written off as gone do occasionally move, so any "adjusted supply" figure is a probabilistic estimate wearing a precise-looking decimal.
The March 2026 snapshot shows how concentrated recent activity actually is: more than 8 million BTC changed hands within the prior year, while about 6 million BTC had last moved more than five years earlier .
"Decentralised finance data presents a distinctive paradox: while every data point is publicly recorded and accessible, deriving meaningful insights is obscured by the scale, fragmentation and complexity of the ecosystem," — Timothy Aerts, Ronald Heijmans, Jan Paulick and Violeta Vuletic, BIS Working Paper No. 1377 (source: Bank for International Settlements)
For an active trader, the practical read is narrower than the headline. A supply-adjusted valuation model, a dormancy-based "lost coins" thesis and an on-chain volume spike all inherit the same classification assumptions — and none of them survives being quoted to three significant figures .
What to watch next: dual reporting and methodological disclosure
The BIS recommendation is procedural, not predictive: on-chain indicators should be treated as "noisy approximations rather than direct measures of economic activity," and published as assumption-dependent ranges with explicit methodological disclosure . Working Paper No. 1377, published 15 September 2026, argues that transaction volumes, market capitalisation and total value locked "often suggest a degree of accuracy that is not supported by the nature of the underlying data" (source: BIS Working Paper 1377, 2026-09).
Parts of the analytics industry already report this way. Visa's Onchain Analytics dashboard, built with Allium Labs, publishes both total and adjusted stablecoin transaction figures, with the adjusted series designed to strip noise from bots, high-frequency trading, bridge routing, internal exchange operations and artificially inflationary activity . Allium markets the same filtering across more than 150 chains and 10,000 protocols — vendors now compete on classification assumptions, not data collection .
The gap between the two series is not marginal. One trailing 30-day reading cited in coverage of the paper showed roughly $6.4 trillion in total stablecoin transaction volume against $313.1 billion adjusted . These are third-party dashboard figures, not BIS outputs.
Three things worth tracking from here:
- Methodology tags on dashboards. Watch whether vendors label a series as upper-bound, self-transfer-adjusted or conservative — the three constructions BIS tested .
- Range reporting over point estimates. A volume figure quoted without a stated method carries an unspecified error band, not precision.
- Adjusted-series adoption beyond stablecoins. Visa and Allium apply dual reporting to stablecoin flows (Visa Onchain Analytics); Bitcoin transfer value has no equivalent standard yet.
The concrete takeaway: before acting on an on-chain volume or market-cap number, check which of the three methods produced it. If the dashboard does not say, treat the figure as an upper bound and size the position accordingly (CryptoSlate, 2026-09).
Frequently asked questions
Does this mean reported Bitcoin trading volume is fake?
No. The BIS paper measures on-chain transfer value — bitcoin moving between addresses on the ledger — which is a different metric from exchange trading volume, where buyers and sellers match orders on a venue. Working Paper No. 1377, published 15 September 2026, does not label any specific published figure as wrong . Its claim is narrower and methodological: the same ledger data yields estimates differing by up to a factor of six depending on how change outputs are handled, and the divergence is specific to the three estimators tested .
Why does Bitcoin's accounting model cause this problem but not, say, a bank account?
A bank account holds a single running balance, so a payment simply debits one number and credits another. Bitcoin uses UTXO accounting — unspent transaction outputs — where each spend must consume an entire referenced output rather than a slice of a balance. The BIS worked example: spending a 4 BTC output to send 1.5 BTC creates a 1.5 BTC output for the recipient and a 2.5 BTC "change" output returning to the sender . Summing raw outputs records 4 BTC transferred when only 1.5 BTC changed hands. Change outputs are not always identifiable from transaction data alone, especially where wallets avoid address reuse, so correcting for them requires behavioural heuristics .
What is realised capitalisation and how does it differ from market cap?
Realised capitalisation values each coin at the price it last moved on-chain, rather than marking the entire supply to the current spot price as conventional market capitalisation does. BIS finds simple market cap has reached up to four times realised cap during price surges . The paper also shows dormancy distorts the standard figure: excluding outputs untouched for more than 15 years removes roughly 1.8 million BTC from the calculation, and around 3.5 million BTC have not moved in more than a decade . Dormancy is a proxy for loss, not proof — about 24,000 BTC were later spent after ten-plus years idle.
Which on-chain metrics should traders trust less after this paper?
Any headline figure that does not disclose its methodology. The BIS recommendation is that on-chain indicators be treated as noisy approximations rather than direct measures of economic activity . The specific categories to discount:
- Raw transaction volume — the widely quoted upper-bound series that sums all outputs and assumes no change.
- Unadjusted market capitalisation — supply times spot price, with no dormancy or realised-cap cross-check.
- Total value locked figures presented without stated assumptions.
- Cross-chain comparisons where each chain's data was classified differently.
BIS notes these metrics "often suggest a degree of accuracy that is not supported by the nature of the underlying data" .
Is this BIS finding specific to Bitcoin?
No. The study covers Bitcoin, Ethereum and Tron using roughly 100 billion blockchain records drawn from Mercurius, the platform operated by De Nederlandsche Bank . On Ethereum, the classification problem is arguably larger: BIS classified 13 million active smart contracts, including about 1.4 million ERC-20 token contracts, while more than 54 million active contracts sit outside its technical classification entirely . Token symbols are unreliable identifiers — 6,867 ERC-20 contracts use the "USDT" symbol. Behaviour also diverges by chain: USDT held inside Ethereum smart contracts rose above 20% in 2022 against roughly 1% on Tron.
Enjoyed this article? Subscribe to get new stories by email whenever they're published.
