03 · Capital lab · Basel III · —
—
01 · The trade
Risk density against return on risk capital
Bubble area is RWA. Rings mark institutions below the 10.5% reference floor. Dashed lines are cohort medians. Click a bank to load it into the panels.
02 · Ladder
Who holds the most loss-absorbing capital
—03 · Composition
Where the RWA comes from
——
04 · Cross-section
Every institution, capital side
Δ Basel is in percentage points, quarter on quarter. Headroom is PR less 10.5% of RWA — a floor, not the bank's own requirement. RAROC is blank until five quarters of capital exist. NPL90 is the defaulted portfolio as filed, published by IF.Data only from 2025 Q1. Cover is implied provisions (reported coverage × credit book) ÷ that defaulted stock; below 1.00× the provision stock is smaller than the defaulted balance, and the footer prints the median institution, not the aggregate.
How this capital view is built — and what it deliberately does not claim
- Source
- BCB IF.Data, Informações de Capital (report 5), prudential conglomerate scope. Fetched from the portal's REST archive (
www3.bcb.gov.br/ifdata) because the OLinda endpoint was returning HTTP 500 for every period during this build. The two services carry the same account codes; 2024 Q4 and 2026 Q1 were reconciled bank by bank against the OLinda extract with zero differences. - Reported vs derived
- Basel, CET1, Tier I ratios and RWA are filed. PR is filed from 2024 Q4 onward; before that it is derived as Basel × RWA, and each point carries its basis in the data. Nothing is interpolated across a missing quarter.
- RAROC
- 12M net income ÷ average PR over the five quarter-ends in the window. Net income is after the provision charge, so expected loss is in the numerator; capital is risk-weighted, so risk is in the denominator. Provisions are not added back — that would mix a pre-tax charge with an after-tax profit. This is a return on regulatory risk capital, not a full economic-capital model. Payments institutions can run a very thin capital base, so RAROC can exceed 100%; the scatter caps its axis at 50% and marks the points above it rather than hiding them.
- Floor
- 10.5% = 8% minimum + 2.5% conservation buffer. Systemic surcharges (up to 2.5pp for S1/S2) and bank-specific add-ons are not included, so headroom is an upper bound. A bank above the line is not necessarily distributing freely.
- Defaulted book (NPL90)
- Carteira ativa inadimplida arrastada ÷ total da carteira ativa de crédito, per institution, from IF.Data report 130 (SCR, document 3040). It is the defaulted balance still on book — not the provision charge — and it is what makes RAROC readable: a 20% return on a book carrying a 12% default rate is not the same trade as the same return on a 1% book. Published from 2025 Q1 only; earlier quarters are blank rather than approximated from the older AA–H grading.
- Cover of NPL90
- Implied provisions — the reported coverage ratio (provisions ÷ credit) multiplied by the credit book — divided by the SCR defaulted stock. Both numerator and denominator are reconstructed from two filings that do not share a denominator, so it measures provisioning intensity, not a filed coverage ratio, and will not tie to a bank's own disclosure. Below 1.00× the provision stock is smaller than the defaulted balance; that is not proof of under-provisioning, because write-offs, collateral and guarantees sit outside the comparison. Needs a defaulted print, so it starts in 2025 Q1 and is blank before that.
RWA density is a map, not a score. A mortgage-heavy book carries low risk weights and a wholesale book carries moderate ones; neither is “better” without a return attached. The payments cohort also moves violently: eight institutions have no capital print in 2023 Q2, and Cielo's Basel ratio falls from 21.6% to 12.9% between 2024 Q4 and 2025 Q1 as its RWA base is re-scoped. Those are real prints, not gaps — read level comparisons across that seam with care.