Bank Run Monitor: evidence guide
The dashboard separates the existing five-family stress index from supplementary official banking data. An index is not a probability of a bank run, a bank failure assessment or a measurement of withdrawals.
New official evidence
| View | Source and scope | Interpretation |
|---|---|---|
| Global health | World Bank World Development Indicators, sourced from IMF Financial Soundness Indicators and International Financial Statistics; up to eight annual observations per economy/indicator. | NPL/gross loans, bank capital/assets and liquid reserves/assets. Reporting years and national definitions vary. Capital/assets is not CET1/risk-weighted assets; liquid reserves/assets is not LCR. |
| Bank explorer | FDIC financials: complete returned latest-quarter dataset, with previous-quarter records matched by FDIC certificate. | Assets, deposits, estimated uninsured deposits, equity, net loans, ROA and ROE. Monetary source fields are USD thousands; compact displayed amounts are converted to USD. Book equity/assets and net loans/deposits are derived ratios. |
| Deposit trends | Federal Reserve/FRED: all-commercial-bank and large/small domestic-bank deposits, primary credit, reserve balances, liquidity swaps and overnight reverse repo. | Up to 160 source observations, with actual dates and units. Series scopes differ. Weekly deposit changes can reflect transfers, mergers, reclassification and revisions. |
| Resolution history | FDIC failure/assistance registry, with published dates and resolution categories. | Historical records, not current allegations. A source index timestamp is shown separately from retrieval time and the most recent listed event. |
Refresh and failure handling
Public read-only API: bankrun-official-evidence. The server checks structural and FDIC datasets at most daily and FRED series at most every six hours. Browser refresh retrieves the cached dataset; it does not force providers. Sources update independently, persist last-good values and provider retry windows, and report failures or overdue retrieval. Provider 403/429 responses pause requests to that host. Missing values stay missing, including in ratios, changes and chart gaps.
Retrieval time is not observation time. Annual and quarterly reports must not be described as real-time bank health. Latest source records can lag the current date. Previous-quarter deposit change is 100 × (current/prior − 1), where matching valid deposits and a positive prior denominator exist; mergers and reporting changes affect interpretation. Thirteen-/52-week changes use source dates, with a maximum seven-day alignment allowance, not an arbitrary row count.
Coverage at initial acquisition
October 2, 2026: 217 economies in the World Bank country catalogue; NPL data for 153, capital/assets for 151, liquid reserves/assets for 148. FDIC financials: 4,313 institutions for June 30, 2026. Failure/assistance history: 4,117 returned records. These counts describe the source dataset, not a complete inventory of worldwide banks or live deposit flows.
Sources and reuse
- World Bank / IMF: nonperforming loans
- World Bank / IMF: capital/assets
- World Bank / IMF: liquid reserves/assets
- FDIC financial reporting
- FDIC failures and assistance
- Federal Reserve H.8 deposits via FRED
- Federal Reserve H.4.1 primary credit via FRED
World Bank indicators retain World Bank/IMF attribution and applicable data terms. Federal Reserve and FDIC observations retain their official source links. No commercial or owner-private bank datasets are redistributed through this endpoint.
Important gaps
Public sources do not provide a complete real-time inventory of institution-level withdrawals, uninsured deposit flight, intraday liquidity, private collateral haircuts, all cross-currency basis books or every country’s bank-level supervisory returns. Payment-rail incidents, market prices and news remain separately classified context. Annual country ratios are not used to populate missing live observations or open the existing score gate. Supplementary live data is paused in historical replay.