The average bank is not the sector
Three honest numbers for the capital strength of EU banks, all from the same supervisory returns: 16.2%, 18.0% and 22.7%. Only one of them is the sector. The largest is the one you get by default.
The European Banking Authority publishes the capital position of every large bank it supervises, drawn from the COREP and FINREP returns those banks file each quarter. I rebuilt three years of it as a Power BI model: 129 banks, twelve quarters, every line mapped across three exercises that renumber their items every year.
The first thing anyone asks of data like that is how well capitalised the sector is. The file will give you three different answers, depending on which obvious thing you do next.
Three questions, three answers
The CET1 ratio is capital divided by risk-weighted assets. Each bank reports its own. There are three ways to turn 119 of them into one number, and each answers a different question.
- Sum the capital, sum the risk-weighted assets, divide once: 16.2%. This is how much loss-absorbing capital the system holds against the risk it carries — the question a supervisor asks, and the only one of the three that describes the sector.
- Take the middle bank: 18.0%. The typical institution. The right number when the question is how one bank compares with its peers.
- Average the reported ratios: 22.7%. The smallest bank in the file, with €5bn of assets, gets the same vote as the largest, with €2.6tn. It describes no bank, and no banking system.
All three are arithmetically correct. The problem is that the third is the easiest to produce. The ratio is sitting in the file as a column; averaging a column is one click, and in most BI tools it is close to what happens if you do nothing at all.
Never average a ratio. Sum the numerator, sum the denominator, and divide once.
Where the six and a half points come from
Some of it is three banks. Two municipal lenders and a custody bank carry almost no risk-weighted assets — lending to local government and holding client assets attract very low risk weights — so their ratios are enormous: 74%, 89% and 353%. In an average each one counts as much as the largest bank in Europe.
Leave those three out and the average of the rest is 18.8% — still 2.6 points above the sector. The remainder is structural. Smaller banks hold more capital for every euro of risk, and there are a lot more of them than there are large ones.
| Total assets | Banks | Sector ratio | Average of banks | Share of RWAs | RWA density |
|---|---|---|---|---|---|
| Over €500bn | 16 | 14.9% | 15.1% | 60% | 32% |
| €150bn to €500bn | 20 | 17.2% | 17.4% | 18% | 33% |
| €50bn to €150bn | 42 | 18.3% | 29.1% | 15% | 38% |
| Under €50bn | 41 | 20.5% | 21.7% | 8% | 84% |
The density column is the other half of the story. The largest banks carry 32 cents of risk-weighted assets per euro of balance sheet; the smallest carry 84. Internal models are what make a large bank’s balance sheet light, and the output floor that arrived in 2025 exists to limit exactly that — which is why density, not the ratio, is the column worth watching as the floor phases in.
The gap does not close
It is tempting to treat a gap like this as noise that averages out over time. It does not. Over the 101 banks present in all twelve quarters — a fixed panel, so a move is the banks moving rather than the membership — the sector ratio rose from 14.8% to 16.0% between September 2022 and June 2025. The average of banks sat between 5.2 and 8.1 points above it in every quarter.
A trend chart built on the reported ratio column would show the higher line and label it the sector, every quarter, for three years. Nobody reviewing it would have a reason to object, because nothing about it looks wrong.
How the model makes the wrong answer hard
In 2025 the EBA withdrew the template that published the headline ratios directly. It turned out not to matter, because the model had never used them. It stores capital and risk-weighted assets as amounts and computes every ratio as a division of sums, at whatever grain the visual asks for — sector, country, size band, single bank. Recomputed that way, the ratios match the ones the banks reported to six decimal places wherever both exist.
The average of banks is in the model too, as a measure with a name that says exactly what it is. It sits beside the sector ratio on the first page of the report, because the honest way to deal with a tempting wrong number is to show it next to the right one and explain the difference, not to hope nobody computes it.
Store the numerator and the denominator. Let the ratio be something the model computes, never something it sums.
None of this is specific to bank capital. Margin by branch, arrears by portfolio, cost-to-income by business unit: any table of ratios invites the same average, and it overstates or understates the whole in exactly the proportion that the small units differ from the large ones.
Figures from the European Banking Authority’s EU-wide transparency exercise, 2023 to 2025, reproduced with acknowledgement of the source. The full build — the item mapping across three renumbered exercises, the de-cumulated profit and loss, and the pandas cross-check behind every number here — is in the EU bank capital case study and its repository.