How every module is built

How the FinObservatory crisis layer is built: five primary chronologies, each source's own crisis definitions, dedup rules, cross-source agreement, the two-panel run-up event study (advanced economies and emerging markets), and caveats.

The three competing haircut measures, which column is which and how the data proves it, the coverage of the 1823-2024 record by era, and what the haircut data cannot tell you.

How the /housing event study is built: the BIS unit and value_type filter that must be applied to both columns, the annual aggregation rule, the -5/+5 completeness rule that selects the sample, the crisis chronology used, and every limitation.

How the FinObservatory Composite bank-health score is built: formula, component weights, peer groups, the honest 2023-failures backtest, and the mandatory not-CAMELS caveats.

How FinObservatory's Bangladesh banking module is built: Bangladesh Bank source lineage, the aggregates-only design decision, definitions and coverage denominators, the IMF FSI perimeter comparison, and staleness.

How the FinObservatory credit-union layer is built: the NCUA 5300 Call Report, the whole-dollar units, the December-cycle convention for annual series and the partial-filer caveat behind it, net-worth availability from 2000Q4, and the validation anchors.

How the FinObservatory deposits layer is built: the FDIC Summary of Deposits survey and its June 30 convention, the DEPDOM reconciliation anchor, the institution-share HHI construction with the DOJ/FTC concentration thresholds, and every stated caveat.

How FinObservatory builds county banking markets from the FDIC Summary of Deposits: the branch-count HHI and the deposit HHI, the verification that the county rollup reproduces the branch file exactly, the central-booking distortion, the null-HHI case, the retired-FIPS register, and every limitation.

How the FinObservatory FR Y-9C layer is built: the form, the pre-2006 lower-tier double count, the noncontrolling-interest identity break, the Tier 1 nulls, the reporting-threshold steps, the source cutoff, and what the panel cannot compute.

How the FinObservatory lending layer is built: the HMDA screen, the composition standardization, the string sentinels that are not zeros, the schema drift between the 2018 and 2019 files, the county FIPS validation, and what the public register cannot tell you.

How the FinObservatory mortgage layer is built: the HMDA public Loan/Application Register, the privacy modifications on the published files, the two schema eras and their exact action-taken and loan-purpose code mappings, the denial-rate definition, the unit conversions, and the snapshot-revision caveat.

How the FinObservatory conditions layer is built: the 8-component PCA FCI, the Basel III one-sided-HP credit-to-GDP gap, the credit impulse, the curated yield-curve and stress panel, the SLOOS credit cycle and Z.1 sectoral debt, NY Fed money-market rates (SOFR/EFFR and the FOMC target range, with the pre-2000 FRED DFF splice), the household balance sheet, and the discount-window lender-of-last-resort disclosures, with their windows, citations, the NFCI validation, and every stated limitation.

How FinObservatory computes the Fama-French factor returns: percent units, geometric compounding, missing-value sentinels, the two conflicting Mkt-RF series, ragged industry coverage, and what the data cannot support.

How the FinObservatory rates layer is built: Treasury's own description of the daily par yield curve (monotone convex method, bid-side quotations), the tenor-set evolution by era with the locate-by-name rule, the FiscalData auction fields, the announcement-vs-result distinction, and the units of every figure.

How FinObservatory computes real returns on equities, housing, bonds and bills from the Jorda-Schularick-Taylor Macrohistory Database: deflation, geometric averaging, the war years, and what the housing series cannot tell you.

How the FinObservatory systemic-risk series is built: LRMES, MES, Delta-CoVaR and the absorption ratio, their windows and citations, the universe, the engine bug worked around, and why the dollar SRISK level is an open gap.

How the FinObservatory valuation layer is built: the two source files and their vintages, the CAPE definition and its reproduction check, the overlapping-window problem and how the non-overlapping sample is drawn, the Damodaran identities that are asserted at build time, and every trap found in the data.

Where the China lending figures come from, how they are constructed, what they exclude, and why they are not comparable with borrower-reported external debt.

The exact BIS selection behind every figure on /claims: one measure, one basis, aggregates excluded, and the traps in the locational banking statistics that this layer is built to avoid.

How the FinObservatory cross-border layer is built: the BIS locational (residence) and consolidated (nationality) banking statistics, the exact dimension cut, the claims-versus-liabilities distinction, units, and the reporting-population and confidentiality caveats.

How FinObservatory builds the external balance-sheet pages from the External Wealth of Nations database: the equity-versus-debt split, why crisis episodes are deduplicated, and what the data cannot say.

How the FinObservatory external-debt layer is built: the World Bank Debtor Reporting System, the 24 IDS indicators and their exact definitions, the derivation of the bilateral residual, the cut between actual years and the contracted debt-service schedule, the balanced panel, and the coverage caveats.

How the FinObservatory remittances layer is built: the WDI balance-of-payments series, the GDP denominator, the balanced panel, the page rule, and why the bilateral corridor matrix is a model output rather than a measurement.

How the FinObservatory reserves layer is built: the IMF COFER dataset and dimensions, the World-aggregate allocated-reserve shares, the five-currency subset limitation, the recent-quarter revision caveat, and the exact queries behind every figure.

How the FinObservatory sovereign-debt layer is built: the IMF debt series and their fallback chain, the debt-at-crisis construction, the era distributions, the current-elevation benchmark, and every stated caveat.

How the FinObservatory analyst works: the never-generate-numbers rule, the StatGPT architecture, the fixed tool set, the citation format, what it will not do, and the model and privacy disclosures.

How the FinObservatory country pages are composed: the union of layers that defines which countries get a page, the exact GFDD and BIS series shown, each source's coverage and staleness, and the license the composition inherits.

How the /filings module is constructed from the SEC Financial Statement Data Sets: the universe, the curated tag lists, the five conditions that make a fact safe to sum, and what the module refuses to claim.

How FinObservatory's financial-crime reference layer is built: the two FATF lists quoted verbatim, why the R.1-R.40 / IO.1-IO.11 ratings are not reproduced, the OFAC aggregates-only decision, vintages, and license notes.

How the FinObservatory inclusion layer is built: the Findex survey rounds and how 2022 folds into the 2021 round, the indicators published, the coverage breaks, why GFDD is not mixed with Findex, and every stated caveat.

How the FinObservatory research library was acquired and what it may show: per-corpus acquisition detail, the recorded license language for BIS, Federal Reserve, IMF, and SEC material, the catalog schema, the federalreserve.gov URL mapping, and the honest-absence rule.