EconoMonitor publishes a figure only after it has been rebuilt from the official file it comes from and tested against a second source or an accounting identity. This page explains how data moves from a publisher to a page, which checks run, and how revisions, withheld values, ranks and peer groups are handled.
Methodology by dataset
- US banks: FDIC Call Report data for the largest FDIC-insured banks, checked against FFIEC filings
- US macroeconomic series: Federal Reserve, Bureau of Labor Statistics, Treasury and New York Fed series
- Canada: Bank of Canada and Statistics Canada series, and OSFI capital data
- Australia: Reserve Bank of Australia, Australian Bureau of Statistics and APRA data
- United Kingdom: Bank of England and Office for National Statistics series
Definitions, units and formulas for every metric are also listed in one table in the Data dictionary.
How does data reach a page?
Every dataset passes through the same layers. The raw layer stores each file exactly as the publisher sent it, with the time it was retrieved. The clean layer converts values to standard units (full US dollars rather than thousands, ratios in percent as published), dates each value at the last day of its period, and applies one definition per metric. The entity layer gives every bank, country and indicator a stable EconoMonitor ID, linked to the publisher's own identifiers such as the FDIC certificate number and the RSSD ID. Derived metrics, such as the loan-to-deposit ratio, are calculated from the clean layer with the formulas shown in the data dictionary.
The website never queries the master database. When a dataset passes its checks, the pipeline exports static files and each page is built from them. Every page shows the period its data describes ("Data as of") and when its data last changed ("Last updated").
Which checks run before anything is published?
Check A compares the database with the downloaded source file. Every value is rebuilt from the raw file and must equal the stored value, the rows received must equal the rows stored, and there must be exactly one current value for each entity, metric and period.
Check B tests the data against something independent. For US banks, FDIC figures are compared with the Call Reports filed with the FFIEC, and the sum of all insured banks is reconciled with the FDIC's published national totals. For macroeconomic series, growth rates are recomputed from levels, monthly averages are recomputed from daily data, and totals are compared with the sum of their parts. Each dataset page below lists its checks.
A dataset is exported only when its latest review report says both checks passed. A check that compares nothing counts as not passed. Each data page shows its check results in the "Data checks for this page" panel, and the Status page shows them for every dataset.
What happens when a few values disagree?
When a check fails on a handful of individual values (at most 5 values in a check, and at most 0.5% of the values it compares), those values are marked provisional and are not published. Text, ranks and charts then use the latest verified period for that metric, and the page says so. The mark clears automatically when a later run matches. A larger failure, or one that affects a whole series, stops the dataset from being exported until it has been reviewed.
How are revisions handled?
Publishers revise past figures. When a value changes, the earlier value is kept with the date it was retrieved and the new value becomes current, so the database records what was reported and when. Values that a publisher withdraws keep their history but are no longer shown as current. Pages show the current values.
Mergers, failures and breaks in series
For US banks, every quarter in which a bank's total assets moved by more than 25% is checked against the FDIC's structure history for mergers, acquisitions and failed-bank purchases before the data is published. Bank pages list these events and warn that figures before and after an event are not like for like. Banks that merged, failed or closed after 2016 while among the 500 largest stay in the database with their full history.
How are ranks and peer groups built?
A rank is the position of a bank among the banks in our universe that report a value for that metric in that quarter. Pages state it as "rank among N banks reporting this metric". For most metrics 1 is the highest value; for noncurrent loans, net charge-offs and the efficiency ratio, where lower is safer, 1 is the lowest. Equal values share a rank. Unrealized gains and losses in dollars are not ranked, because they mostly track bank size. A rank is a position in a table, not a rating.
A bank's peer group is the 6 banks closest to it by total-assets rank in its latest verified quarter: 3 above and 3 below, shifted to one side at the top and bottom of the table.
Units, rounding and changes
Values are stored in base units and rounded only for display, using the number of decimals set for each metric. Changes in amounts are shown in percent; changes in ratios are shown in percentage points. A change is never calculated against a missing or withheld value.
Licences and attribution
The reuse terms of every source are checked before its data is used. Each page carries the attribution text its publisher asks for, and data from a source whose terms have not been confirmed for display is not shown. The Sources page lists every source, its licence and where its data may be used.
Corrections
If you think a figure is wrong, write to contact@economonitor.com with the page link and the value. Confirmed corrections are listed in the Changelog.