The fee index

Where it stands

A and B are ratios of period sums, so the 365-day figure is the year's fees over the year's estimated volume rather than an average of 365 daily ratios. C is a median of medians, because it is already a daily median and averaging it would mean something else.

Read the two columns together or not at all. A fee is the clearing price of a fixed supply of block space, so a high index means demand pressing against a full block and a low one means slack, and neither says anything on its own about how much was moved. Over these 365 days the chain settled $2.98 trillion across 195,458,529 transactions. The chart below puts the quantity under the price for the same reason.

What these numbers are, and what they are not

A measures the settlement layer and nothing else. A Bitcoin transaction moves value from one set of keys to another, and the fee it pays for that is what the numerator counts. Everything a remittance company does after the wire lands, the agent who counts out cash in a shop, the compliance desk, the float, the branch, is outside this measurement. Western Union's agent commissions alone are 45% of its expenses. When Banca d'Italia went mystery shopping across ten corridors in 2026 they found the on-chain leg a marginal share of an all-in cost that ran from 0.3% to 9%. So a comparison of A against a remittance price is a comparison of a wire against a whole service, and the honest reading of the gap is that the expensive part of sending money home was never the wire.

C is the number to quote when someone asks what a user pays. An aggregate of 0.003% is dominated by a handful of enormous transfers, and a base-layer transaction averages around $15,000, so the aggregate says very little about a $200 payment. The median fee against the median output value says quite a lot. It reads 0.095% now and 0.724% across 2024 as a whole, which is card territory rather than bank-wire territory, and it crossed 0.3% in August 2017, the month Bitspark gave up on Bitcoin for remittances and said the fees were why.

B is the one that has somewhere to go. Users pay fees and holders pay dilution, and B counts both: miner revenue, subsidy included, over the same estimated volume. It was 21.3% in 2010 and 4.15% in 2011, and it is 0.479% for 2025, which is the level of a regulated US debit card. Every halving cuts the subsidy and B falls again. The interesting question is where it lands, because theory says a chain securing itself from fees alone needs far more than A currently collects. Auer, writing for the BIS in 2019, put one-hour finality at mining income of 8.3% of volume. Budish, in the Quarterly Journal of Economics in 2024, put the cost of securing against attack at 3% to 30% of honest volume depending on how long you wait. A is three orders of magnitude below that. Which of the two, the theory or the measurement, is describing Bitcoin wrongly is the open question this series exists to track.

The same percentage on other rails

Read as a group, the base layer sits with the interbank rails, SEPA and ACH and Fedwire, which is where a transaction averaging $15,000 belongs. A typical user sits with the cards. The full cost including dilution sits with a debit card and is falling. None of the three is the 0.5% the 2013 note that started this guessed at.

The series

The price is on top and the quantity it was charged on is underneath, sharing one time axis and one crosshair. A fee is the clearing price of a fixed supply of block space, so the index rises when demand presses against a full block and falls when there is slack, and the line alone cannot tell you which. The bars say how much was actually moved at that price.

The bitcoin spot price sits at the foot, on a log axis of its own because it runs from six cents to six figures. It is there to be ruled out. A is fees divided by the value those fees were charged on, so both halves are denominated in the same money and an exchange rate that doubles moves them together and leaves the index where it was. Read the panels against each other and the index does not follow the price, which is the point of showing them together.

A log axis turns exponential growth into something close to a straight line, and sixteen years of bitcoin spans about six and a half decades, so over the full history that panel flattens the cycles into a rising slope. Narrow the period and it refits its axis to whatever is on screen, so five years drops it from six and a half decades to about one and the peaks and troughs separate. Use the period buttons above, or drag across any panel to pick an exact window. Double-click to go back to the whole series. All three panels move together whichever you use.

C is exactly zero from August 2010 to April 2012, when fees were effectively zero, and a log axis has no place to put a zero, so that stretch is drawn as a gap rather than as a line along the floor. The three dashed verticals mark December 2017, when A reached 0.250% in the December fee auction, December 2023, when inscription traffic pushed C to 26.2% and the month before it to 72.0%, and April 2024, when the halving and the Runes launch put A at 4.38% on the 20th, the highest single day in the series.

Read it as a table

Yearly, from the same files the chart draws, with the quantities beside the ratios. Each ratio is the year's numerator over the year's denominator, and the year is however many days the series actually holds rather than however many the calendar has.

How it is computed

Five numbers, all of them percentages of value moved over a UTC day, summed into months and years. A is total fees in BTC over estimated transfer volume in BTC, so the exchange rate cancels and the number says nothing about the price of bitcoin. A prime is the same ratio in dollars, which differs from A when fees cluster in a price spike, as they did in December 2017: 0.102% against 0.147% for that year. B is miner revenue in dollars over estimated volume in dollars, which reproduces a series Blockchain.com has published since 2011. C is the median fee over the median output value, one day at a time. R is A with the unadjusted denominator.

The numerator is exact. Total fees per day are a sum over the chain, and two sources that built it independently agree: Blockchain.com and Bitcoin Visuals sit within 3.5% of each other at the fifth and ninety-fifth percentile of daily ratios, and the residue is how each treats the day boundary. Every weekly run compares the two over the latest 30 days and refuses to write a new store if they have drifted more than 5% apart.

The denominator is an estimate, and no estimate of it can be verified. A Bitcoin transaction's outputs mix the payment with the change returned to the sender, so the economic volume is not observable. Blockchain.com strips change by the address-reuse rule and documents the whole method in one sentence:

The total estimated value in BTC of transactions on the blockchain. This does not include coins returned to sender as change.

That rule treats an output paying to an address that also appears among the inputs as change. Coin Metrics adds round-number and fast-spend tests. Glassnode clusters addresses into entities. The three disagree by four to eight times on what economic volume is, Glassnode's history revises itself as its clustering improves, and Coin Metrics says outright that without ground truth no heuristic's accuracy can be measured. Every percentage on this page is therefore a statement about that one sentence first and about Bitcoin second, which is why the raw lower bound is published beside it in every file.

Period ratios are sums over sums. A month's A is the month's fees over the month's estimated volume. Averaging daily ratios would let a quiet Sunday weigh as much as a busy Tuesday. C is the exception, since it is already a daily median, so a period takes the median of its days.

Missing days stay missing. A day is in the series when both the fee total and the estimated volume have a row for it. Between 2010-08-28 and 2026-09-11 that is 5,827 days out of 5,859, so 32 calendar days are absent, 14 of them in August 2011. An absent day is not in the files and not in the period sums, and nothing is interpolated to cover it.

Every fetch is kept. Upstream revises history quietly. The raw response of every Monday run is stored gzipped under its own dated directory with the URL, the HTTP status, the byte count and the SHA-256 of each file, the store holds the latest values, and a release is an immutable snapshot of that store. Any release can be rebuilt from its own raw directory with no network at all, which is what makes the series reproducible rather than merely downloadable.

Seven ways this can mislead you

1. The denominator, again. It is a heuristic with no ground truth, and it is the whole difficulty. Everything below is downstream of it.

2. Inscriptions and other non-payment use. Ordinals and BRC-20 put real fees in the numerator against almost no economic value in the denominator. The rise in A through 2023 and 2024 is largely this, 0.219% in December 2023 and 0.118% in April 2024. No vendor documents stripping it and this site does not strip it either, because there is no per-transaction source in this pipeline to strip it with. C is hit far harder than A: inscription dust drives the median output value toward zero, so C reads 72.0% in November 2023, 26.2% in December and 3.99% in January 2024. Those are artifacts of the denominator, not fees anybody paid, and they leave the trailing window only by the window moving past them.

3. Batching, CoinJoin and exchange shuffles. One economic transfer can be counted nine to seventeen times through a CoinJoin sequence. An exchange moving its own coins between its own wallets can print a quarter of a trillion dollars in a single day. Batched payments were 30% to 60% of raw output value in 2018. The address-reuse rule catches some of that and not the rest.

4. Secular drift. As activity moves onto exchange ledgers and Lightning, the on-chain denominator shrinks and A drifts upward as a measurement artifact, the way NVT did after 2018. A rising A is not evidence of a more expensive network until you have looked at B and C.

5. Fees are per byte, not per value. Two transactions of the same size pay the same fee whether they move $10 or a billion. A therefore tracks the size mix of transactions more closely than it tracks any pricing decision. That is the reason C exists on this page.

6. The subsidy. A describes what senders pay. B describes what the network costs, and its path from 21.3% in 2010 to 0.479% in 2025 is the security budget question in one line.

7. The remittance comparison is not like for like. Covered at the top of the page, and repeated here because it is the thing most likely to be quoted out of context. A measures the wire. A remittance price measures the wire plus the last mile, and the last mile is where the money goes.

Sources

Ten files, once a week, no key and no account anywhere in the pipeline.

• Blockchain.com charts API for total fees in BTC and USD, estimated transfer volume in BTC and USD, output volume, miner revenue, transaction count and market price. Nine daily series back to 2009, keyless, cached four hours upstream. They state no licence and forbid nothing, and these series are trivially re-derivable from public chain data, so the posture here is attribution and a link on every surface that shows a number.
• Bitcoin Visuals for the median fee and median output value that make C, and for the independent fee total the weekly cross-check runs against. One daily CSV, published because, in their words, they make all their website data public.
• World Bank Remittance Prices Worldwide for the benchmark, under CC BY 4.0. The quarterly global average cost of sending $200, recomputed from the raw dataset with the bank, money transfer operator, Western Union and MoneyGram columns. RPW is twice-yearly now and the latest issue is 2025 Q3.

Sources that were evaluated and rejected, so nobody spends a second afternoon on them: Coin Metrics (adjusted volume is paid and quote-only, the community licence is non-commercial), Glassnode (no free tier, and the terms forbid publishing derived data), Blockchair (the API has been frozen since 2022), Bitquery ($239 a month floor), Dune (free executions ended in September 2026), mempool.space (no volume series at all). BigQuery's public Bitcoin dataset is the one alternative worth keeping in mind: a change-adjusted backfill with published SQL fits inside its free tier, and that is the fallback if Blockchain.com ever stops serving.

Prior art

The ratio is not a new idea and the gap this fills is that nobody maintains it. Möser and Böhme computed the fees-only version weekly for 2011 to 2014 in Trends, Tips, Tolls and concluded fees ran below 0.1% of transmitted value, then stopped. Matteo Leibowitz named it the Fee Ratio in 2018, measured 0.41% for Bitcoin against Coin Metrics volume, and moved on to a different metric. Banca d'Italia published the full distribution of fee rate to transacted value for 2017 to 2019 and found more than 90% of payments paid less than the 2.4% a card cost. Blockchain.com has published the miner-revenue variant daily since 2011, which is the series reproduced here as B. What is missing from all of it is an ongoing fees-only series with a stated methodology, a benchmark line and a citable release.

Download and cite