A Bank for International Settlements working paper has found that estimates of Bitcoin onchain transfer value can differ by as much as sixfold depending on how researchers treat transaction outputs, a gap that undermines many commonly quoted blockchain metrics. The paper, published in September 2026, is titled “Hidden by complexity? Measuring stablecoin, crypto and decentralised finance ecosystems” and was written by Timothy Aerts, Ronald Heijmans, Jan Paulick and Violeta Vuletic.
How the study was built
The research uses data from Bitcoin, Ethereum and Tron collected through Mercurius, a project operated by De Nederlandsche Bank and developed with the BIS Innovation Hub and Deutsche Bundesbank. Across the three networks the dataset holds 100 billion records, though the authors caution that this counts data points stored through their processing system and should not be read as 100 billion distinct blockchain events, since records can overlap between base-layer and smart-contract data and appear at several stages of the pipeline.
For Bitcoin, the study examined data covering 2009 through 2026, roughly 1.3 billion transactions and 3.6 billion transaction outputs. The sheer scale is part of the point: any metric quoted for the network is a processed estimate, not a raw observation.
Why the numbers diverge
The problem comes from Bitcoin’s UTXO model, which returns unused funds to the sender as change. The researchers tested three ways of calculating transfer value. The upper estimate counts the value of all outputs. An adjusted measure removes outputs sent back to the sending address, treated as likely change. A conservative lower estimate removes identified self-transfers or, when none can be identified, subtracts the transaction’s largest output on the assumption that smaller outputs represent the economic transfer.
Applying the different methods produced gaps of up to sixfold in monthly estimates of Bitcoin transfer value, with the estimates diverging most during periods of high activity and rising prices. The authors caution that their lowest estimate is a conservative heuristic, not a mathematically certain measure. CoinJoin transactions can contain several change outputs, and mixers, spam activity and intermediary transfers create classification problems the model does not attempt to remove. Self-transfer exclusions became more pronounced from March 2016 as address reuse increased.
The authors said commonly quoted metrics can convey a degree of accuracy that is not supported by the nature of the underlying data.
Market cap has the same problem
The measurement problem extends to valuing Bitcoin’s supply. The study compared conventional market capitalization, which applies the current price to all outstanding supply, with alternatives that account for dormant coins and the price at which outputs last moved. Removing UTXOs untouched for more than 15 years excluded just over 1.8 million BTC, though the researchers stress that inactivity cannot prove keys are lost. About 3.5 million BTC had sat dormant for more than 10 years, but nearly 24,000 of those eventually moved, including close to 3,000 after 15 years, evidence that age-based lost-coin estimates stay uncertain.
Realized capitalization, which values each output at the price when it was created, told a different story across cycles. Conventional market cap reached as much as four times realized cap during rapid price appreciation, and during Bitcoin’s sharp 2022 decline realized cap temporarily stood above market cap because older outputs kept their earlier valuations.
Ethereum and stablecoins fare no better
The researchers examined 67.5 million deployed and active Ethereum contracts and could not place more than 54 million into their technical categories. Among those they could classify, close to 12 million were proxies, 1.4 million were fungible-token contracts and roughly 100,000 were NFT contracts. The authors used bytecode and standards such as ERC-20 and ERC-721 for classification but warned that following a technical standard does not establish a contract’s economic purpose.
On stablecoins, the study found USDT held in smart contracts exceeded 20% of supply on Ethereum, while on Tron the figure stayed near 1% for most of the period studied. The authors associated Ethereum’s smart-contract balances with liquidity provision, lending and other DeFi uses, and described Tron’s USDT as appearing more in transactional and store-of-value activity, while warning that account type remains an imperfect proxy for actual use, since externally owned addresses can represent payments, exchange custody, remittances or holdings.
Visa already filters, BIS wants ranges
The gap between raw and adjusted numbers is visible in commercial dashboards. Visa’s Onchain Analytics dashboard, built with Allium Labs, separates total stablecoin activity from an adjusted measure that filters high-frequency trading, bots, bridge routing, exchange activity, minting and burning. A September 16 snapshot showed $6.4 trillion in total stablecoin transaction volume over the preceding 30 days and $313.1 billion after Visa’s adjustments, a reduction of more than 95%. Visa’s methodology counts only the largest stablecoin transfer within one transaction, cutting duplicate internal movements generated by complex smart-contract calls.
The BIS authors recommend supplementing single-number estimates with bounded ranges that disclose protocol-specific uncertainty, explicit assumptions, and separate treatment of an asset’s identity from the blockchain it runs on. They describe onchain indicators as noisy approximations rather than direct measures of economic activity.
Why it matters
Headline onchain metrics move markets and inform policy. ETF issuers cite network activity in filings, regulators cite stablecoin volumes in rulemaking, and traders build strategies on transfer data. If the same underlying blockchain can support estimates six times apart depending on methodology, the precision those numbers imply is an illusion. The BIS paper does not say the data is useless, only that anyone quoting a single figure for onchain activity should say how it was counted. For a market asking regulators to treat crypto as measurable financial infrastructure, that is not a small caveat.