How the Bank Pulse score works.

Bank Pulse assigns every bank in our coverage universe a daily composite score from 0 (worst) to 100 (best), plus five component sub-scores. The composite is a weighted average of the components; the components are deterministic functions of publicly disclosed metrics. There is no LLM in the pipeline.

Algorithm version

Every score row stored in the database carries the version of the algorithm that produced it (v0.2.0-2026-05-09at time of writing). When we tune weights or normalisation curves, we bump the version and re-compute historicals. That makes the comparison “was the SVB pre-failure score under the new algorithm worse than the new algorithm would have flagged today?” mechanically answerable.

Weights

The five components and their weights in the composite:

Sub-score curves

Each input metric is mapped to 0–100 by a piecewise-linear curve calibrated against historical bank failures. The two curves that matter most:

• Tier 1 ratio. 4% maps to 0 (well below the 6% regulatory floor, signalling severe distress); 16% maps to 100 (substantial loss-absorbing buffer). Linear in between.
• HTM unrealised loss / Tier 1. 0% maps to 100; 67% (a higher concentration than SVB carried) maps to 0. Linear penalty in between. This is the variable that distinguishes our score from a CAMELS-style regulatory rating, because it captures duration mismatch that does not show up on the balance sheet under standard accounting.

Bands

The composite score is bucketed into six bands purely for editorial display. The thresholds are not used in any downstream calculation; they exist so a reader can glance at a profile and know how concerned to be:

• 0–29 — On the Brink. Capital and liquidity both flagging.
• 30–44 — Stressed. Multiple risk factors compounding.
• 45–59 — Watch. At least two components below peer median.
• 60–74 — Adequate. No individual component in the warning band.
• 75–89 — Healthy. Comfortably above peer medians on most components.
• 90–100 — Robust. Top decile of the universe.

Inputs and sources

• Capital ratios, AOCI, loan composition, brokered deposits, NPL. Federal Financial Institutions Examination Council (FFIEC) Central Data Repository for US banks; FR Y-9C for US bank holding companies; European Banking Authority (EBA) Pillar 3 disclosures and ECB Supervisory Banking Statistics for EU/EEA banks.
• HTM unrealised losses. Derived from schedule-RC-B holdings × current Treasury yield curve from FRED. Where banks publish a marked-to-market figure directly we use the disclosed figure.
• Equity price, market cap, P/B. Yahoo Finance unauthenticated query endpoint, daily close. For European tickers we use the local-exchange listing (Xetra, Borsa Italiana, LSE).
• 5Y CDS proxy.Senior unsecured bond YTM minus matched-tenor Treasury or Bund YTM. Real CDS spreads from a paid feed will replace the proxy in a later release; until then, the proxy is approximately 80% correlated with traded CDS for the largest banks and unavailable for smaller banks (which surfaces as “missing” in the score breakdown).

Update cadence

Fundamentals refresh on the regulator filing cycle — quarterly, with a 60–75 day lag for FFIEC and ~80 days for EBA disclosures. Market signal refreshes daily. Score is re-computed daily so that a sudden equity decline can move the composite even when the underlying call-report data has not yet refreshed.

Every metric carries its “as-of” date on the per-bank profile. We never silently roll a stale call-report number forward.

Coverage

Top 250 US banks by total assets, all G-SIBs and D-SIBs globally, the largest EU/EEA banks by total assets, and the ten largest non-Western international banks. Privately held banks within scope are included with fundamentals only — no market signal column, and the market sub-score is omitted from their composite (the remaining four components reweight to sum to 1).

Cross-jurisdictional comparability

Banks across the US, EU/EEA, UK, Switzerland, and Asia file under different accounting and disclosure regimes. The score should be read as comparable but not identical across jurisdictions. The largest specific differences:

• IFRS 9 vs US GAAP. EU/EEA and UK banks classify expected credit losses under IFRS 9 stages; US banks under CECL. The two are conceptually similar enough that the NPL-driven Earnings sub-score is comparable, but absolute provision levels differ by jurisdiction.
• Held-to-maturity treatment. US GAAP permits HTM classification with no recurring fair-value disclosure. IFRS 9 categorises the equivalent positions as amortised cost with similar effect. Where IFRS- reporting banks publish a fair-value-of-securities footnote, we use it directly; otherwise the FRED-derived estimate applies the same way as for US banks.
• Brokered deposits. A US-specific concept. For non-US banks we substitute the closest disclosed equivalent (intermediated time deposits, wholesale-funding ratio); where neither exists the metric is omitted and the Liquidity sub-score reweights to brokered=0.
• Capital ratios. Tier 1 and CET1 are standardised under Basel III and broadly comparable. Leverage-ratio numerators differ slightly between US and EU (G-SIB surcharges); we use the disclosed figure as published.

We disclose the source for every metric on the per-bank profile so readers can drill into the underlying filing.

Corrections

We commit to publishing a correction within one business day of any factual error in our inputs or score. Send corrections to corrections@regulator.watch with the bank name, the metric, and the source we should have used.

What Bank Pulse is not

• Not a regulatory rating. Banks are supervised by their primary regulator using CAMELS. The Bank Pulse score is an independent editorial signal, not a substitute.
• Not investment advice. Use it to inform questions; don’t use it to decide trades.
• Not a prediction. A low Bank Pulse score is not a prediction of failure; it is a description of the inputs at a point in time.

Back to the leaderboard · the museum · general methodology.