Bitcoin Transfer Totals Differ Sixfold by Method, a BIS Paper Says

A Bank for International Settlements working paper released in mid-September argues that the headline figure for how much value moves across Bitcoin in a month depends mostly on how it is counted. Count every transaction output and one number comes out. Strip the coins a wallet sends back to itself and a much smaller one does. The paper puts the distance between those two at a factor of six, and builds a wider claim on it: on-chain metrics are quoted with a precision the underlying data does not support.

Change looks exactly like a payment

Bitcoin wallets spend in discrete pieces called unspent transaction outputs. Spending part of one means spending the whole thing and taking the remainder back, and the chain puts no label on which output is the remainder. Analysts infer it from transaction shape and address patterns. The authors, Timothy Aerts, Ronald Heijmans, Jan Paulick and Violeta Vuletic, set three methods against each other: an upper estimate counting every output value, an adjusted measure that removes outputs returned to the sending address, and a conservative floor that strips identified self-transfers or subtracts the largest output. Run across roughly 1.3 billion transactions and 3.6 billion outputs from 2009 to 2026, the top and the bottom sit six times apart.

"The resulting estimates of transfer volume can differ by a factor of six, depending on the methodology used," the paper said, in the wording The Crypto Times carried.

A title two outlets spell differently

The BIS page for the paper could not be retrieved for this article. Both the summary page and the PDF for Working Paper 1377 returned a 404 on Wednesday, so everything here comes from outlet accounts, with each figure checked against at least two of them. The title is one casualty. crypto.news, The Crypto Times and Coindoo all give it as Hidden by complexity? Measuring stablecoin, crypto and decentralised finance ecosystems, while Cointelegraph renders it Hidden Complexity: Measuring Stablecoin, Crypto and Decentralised Finance Ecosystems. Three accounts against one, and no primary copy to settle it. The date splits the same way, with two outlets putting the paper on 15 September and crypto.news giving only the month.

Tickers, contracts and one dashboard read twice

Bitcoin is the clearest case, not the only one. The authors examined about 67.5 million deployed and active Ethereum contracts and could not categorise more than 54 million of them. Ticker collisions are part of the reason. crypto.news says roughly 7,000 contracts carry a USDT label and exactly one is the real thing, while Coindoo counts 6,867 contracts using the ticker. Those two counts do not reconcile. Where the tokens actually sit varies by chain as well, with smart contracts holding above 20% of USDT supply on Ethereum in 2022 against about 1% on Tron.

The stablecoin figures come from Visa's Onchain Analytics dashboard, which coin-turk says runs on Allium Labs data. Cointelegraph and coin-turk both report a 30-day total of $6.4 trillion against $313.1 billion once adjusted. The Crypto Times gives $6.1 trillion and $306.1 billion for the same pair. A rolling window read on two different days would explain the gap, and none of the three accounts says when it was read. We could not establish which reading came later. Firms that sell crypto market data publish single figures of this kind daily.

What a bounded range would look like

The prescription is modest. Publish ranges instead of point estimates, the authors say, with the uncertainty specific to each protocol stated openly. Market capitalisation gets the same treatment, having run at four times realised capitalisation during fast rallies; realised cap values each coin at the price it last moved at. Dormancy sits in the supply figure every ratio divides by, with about 1.8 million BTC untouched for more than fifteen years, and decade-dormant wallets emptying now and then without notice. Whether any index provider or vendor changes what it prints is not something a working paper decides. Nor does the paper name one of its three methods as the one to use.