How Banksparency computes its figures

Written and computed byAviel Fahl

One person builds the collection pipeline, writes the code that derives each figure, and writes the text around it. That is why this page states the method in full and why the correction log at the foot is open to the reader.

Rates come from the banks themselves. Banksparency records rates as observed on each bank's website; a change may have occurred before the date it was observed. Every rate on this site carries the date it was observed.

Banksparency applies identical methodology to affiliate and non-affiliate banks. No bank can buy a rate, a rank or a date. Sponsored placements are labeled and sit outside every ranking; the advertising disclosure sets out the rule.

The savings, money market and CD indices are built from the same observations, under their own inclusion rules. The index methodology is published separately.

Products sit in one of two tiers. The daily tier is re-observed at 10:00 UTC from Monday to Saturday. The rest of the products are re-observed once a week, at 02:00 UTC on Wednesday.

Sunday is left out on evidence, not habit. Over a measured window of about three months, Sunday recorded zero rate changes in 1,191 observations, while Saturday repriced at the weekday rate. A run on Sunday would read the same rates back.

A re-observation does not mean a rate moved. It means the page was read again on that date, which is the date printed beside the rate.

They come from the quarterly call reports that every FDIC-insured bank files, read through the FDIC BankFind Suite financials API. The figures are as reported by the bank, and the FDIC may amend them later.

The national average savings and money market rates shown beside a bank's own rate come from the FDIC monthly national rate series, and each one carries its month.

Source: FDIC BankFind Suite. Banksparency is not connected with the FDIC, uses no FDIC logo, and claims no endorsement by it.

It is the quarter's savings interest expense, multiplied by four to annualize it, over the average savings and money market balance in the quarter.

The numerator is the call report field ESAVDPQ, which is the FDIC's own per-quarter figure, so no year-to-date differencing is applied to it. The denominator is the mean of the current and the prior quarter-end balance of NTRSMMDA plus NTRSOTH, or the current balance alone when the prior quarter is missing. The denominator actually used is stored with the figure.

The FDIC line for savings includes money market deposit accounts and business savings deposits and does not separate consumer accounts. That is why the published label names both: "Rate paid on savings and money market deposits".

No bank files a rate. It files interest expense and balances, and the rate follows from them. That is also why a quarter can be withheld as not computable.

The same shape, over the whole deposit book: the quarter's deposit interest expense, multiplied by four, over the average interest-bearing deposit balance.

Here the numerator needs differencing. The call report field EDEP is year-to-date, so the quarterly amount is the difference against the prior quarter of the same calendar year, and the first quarter stands alone. The denominator is the mean of the current and the previous quarter's DEPI, and that previous quarter crosses the year boundary: the first quarter uses the prior fourth quarter.

The two rates answer different questions, so they sit side by side and are not combined.

It is count-weighted and strictly below. The percentile counts the banks that paid strictly less on savings and money market deposits, one vote per bank, among the banks that reported a figure for the quarter. One bank is one vote, whatever its size, so a large bank does not move the rank more than a small one.

The percentile compares all FDIC filers, count-weighted. A missing quarter is suppressed, never interpolated.

Each bank page prints the number of banks the percentile counted and the quarter it counted them in, read from the record stored with the figure. A record that does not carry the counts gets the sentence without them, never an invented number.

When the inputs cannot support a reliable rate. The rate is then stored as null with a reason code, and the page prints the reason instead of a number. A suppressed figure is never interpolated, never carried forward from an earlier quarter, and never shown as zero: a break in a chart line is a suppressed quarter.

Because the FDIC API returns 0 for many fields that a bank did not report, a rule that keys on null alone catches none of it. A zero at a bank that holds no deposits is usually correct. The same zero at a bank that holds billions in deposits is usually "not reported".

So each call report field is swept before it is published: how often it is absent, how often it reads 0, how often it reads 0 while the bank reports deposits, and whether a reported 0 is true for that field or a stand-in for nothing. A field with no such record does not reach a page.

On the published pages the rule runs the other way as well. A missing figure is a suppressed figure, never a zero, and no tile prints 0.00% in place of an absent rate.

Because a call report covers a quarter that has already ended. The newest quarter on a bank page is the last one the FDIC published, and it is printed with its quarter-end date.

The ingest runs monthly, at 06:00 UTC on the 15th, and reads the last six quarters by default, so an amendment inside that window is picked up on a later run. The percentile and atlas pass runs monthly as well, at 06:00 UTC on the 20th.

This site prints the date of the data it holds. It does not print a date for data it does not hold yet, because the FDIC, not Banksparency, decides when a quarter is published.

Write to info@banksparency.com. Please include three things: the address of the page, the figure you are reading, and the source you compared it against.

Those three make a report checkable in one pass. A figure that is wrong is corrected at the source and recorded in the log below.

No corrections have been recorded since this log opened on 2026-09-14.