How the deposit betas are built

MDRM item mapping

Item codes were discovered from the files themselves (each schedule file's first row is MDRM codes, second row descriptions) and verified in both an old (2004Q1) and a recent (2023Q1) file. The time-deposit items changed families when the reporting threshold moved from $100,000 to $250,000: the last file carrying RIADA517/A518 is 2016Q4 and the first carrying RIADHK03/HK04 is 2017Q1, with no overlap; the build aborts if a file carries both or neither.

Deposit interest expense per bank-quarter = RIAD4508 + RIAD0093 + time-deposit items of the era + RIAD4172. Interest-bearing deposits = RCON6636 + RCFN6636. Cross-check at build: the component sum must not exceed RIAD4073 total interest expense (which also carries non-deposit interest) beyond a 0.5% share of bank-quarters, or the build aborts; the measured share is printed in the build log each run.

Year-to-date dequarterization

Schedule RI reports income YEAR-TO-DATE, so quarterly flows are differences within the bank-year: Q1 is the YTD value, and Qn is YTD_n minus YTD_n-1, taken only when the prior quarter's filing exists. Differencing produces a negative quarter when a bank amends a prior filing; those are floored to null, not zero, and counted: 1,054 of 665,597 quarterly expense observations (0.16%).

Named receipt, verified at build (the script aborts on failure): for JPMorgan Chase Bank NA (IDRSSD 852218), the four dequarterized 2023 quarters sum to $43,566,000 thousand, exactly the raw file's 2023Q4 year-to-date deposit interest expense. Its implied deposit rate moved from 0.05% in 2022Q1 to 2.62% in 2023Q3: beta 0.50.

Implied rate and drop rules

Implied deposit rate = 4 x quarterly deposit interest expense / average of the current and prior quarter-end interest-bearing deposit balances, in percent. Rates below 0% or above 25% are data errors (tiny denominators, mid-quarter charter events) and are dropped and counted: 32 of 646,678 computable bank-quarter rates (0.005%). The parquet keeps the unbounded raw rate beside the bounded one, so this drop rate is recomputable from the published data.

The product split

The same construction runs per deposit product: each Schedule RI expense item is paired with the Schedule RC-E balances of the accounts it is paid on, and the implied product rate is the annualized dequarterized expense over the average of the current and prior quarter's total balances for that product. The RC-E large-time items changed families at 2010Q1 (RCON2604 through 2009Q4, RCONJ473/J474 from 2010Q1, no overlap, verified in-file) while the RI expense items changed at 2017Q1, so between 2010 and 2016 the $100,000-and-over balance is reconstructed as RCONJ473 + RCONJ474. The small/large threshold therefore moves with the expense split in force: $100,000 through 2016Q4 and $250,000 from 2017Q1. The two regimes are not one continuous series, and the panel records which applies to every row.

The transaction denominator is total transaction accounts INCLUDING noninterest-bearing demand deposits, so the transaction rate is the blended cost of transaction funding, not the rate paid on interest-bearing checking. This is deliberate: since the 2011 repeal of Regulation Q's demand-deposit interest ban, banks may pay interest on demand deposits, that interest is booked in RIAD4508 (verified in the 2025Q1 file: 308 banks with demand books over $100 million and no NOW or ATS balances report material transaction interest), and no public call-report item isolates interest-bearing transaction balances. Dividing by transaction accounts net of demand would put interest-bearing-demand expense over a denominator that excludes the balances it is paid on, which distorts exactly the banks where the question matters.

Each row carries a coverage diagnostic: the four domestic product balances (with transaction taken net of demand) over RCON6636 interest-bearing domestic deposits. Near 1, the products tile the bank's interest-bearing book; below 1 the gap is mostly interest-bearing demand; above 1 it is noninterest-bearing nontransaction balances sitting in the savings and time denominators. Three checks run at build and abort on failure: MMDA plus other savings plus time must equal RCON2385 total nontransaction accounts (tolerance $2,000) in at least 99.5% of bank-quarters; the coverage median must sit within 1% of 1; and the per-product quarterly expenses must sum to the total deposit expense. The JPMorgan dequarterization receipt above is also enforced per product. Measured values are printed in the build log each run.

Cycle definitions

A cycle runs from the quarter containing the first hike to the quarter containing the last. The cumulative (end-to-end) beta per bank is the change in its implied deposit rate between those two quarters divided by the change in the quarterly average of FEDFUNDS, the standard summary in the literature. Banks must have a valid rate in both endpoint quarters.

Start and end rates are quarterly averages of the monthly FRED FEDFUNDS series; the move column is the difference of the two rounded rates shown.

External anchor

The New York Fed's Liberty Street Economics post "Deposit Betas: Up, Up, and Away?" (Kang-Landsberg, Luck and Plosser, April 11, 2023) reports an industry cumulative interest-bearing deposit beta of almost 0.4 by 2022Q4, on par with the 2015-19 cycle's peak, and a 2022Q4 industry interest-bearing deposit rate of 1.4 percent, from FR Y-9C holding-company filings. The same aggregate built from this page's bank-level panel (industry expense over industry average balances) gives a 2022Q4 rate of 1.40% and a cumulative beta of 0.35 from the 2022Q1 base, with the 2015-19 cycle's end-to-end aggregate at 0.37; by 2023Q3 the aggregate reaches 0.50. The deposit rate matches theirs at the reported precision; the beta sits at the low edge of "almost 0.4" because this page's base quarter (2022Q1, average FEDFUNDS 0.12%) and commercial-bank rather than holding-company filings differ from theirs. libertystreeteconomics.newyorkfed.org/2023/04/deposit-betas-up-up-and-away

Limitations

• The bulk files begin in 2001Q1, so the 1994 and 1999 hiking cycles are out of reach; three complete cycles is the panel’s maximum.
• End-to-end betas hide the path: a bank that repriced late looks like one that repriced early. The quarterly path chart covers this for 2022-23 only, and deposit rates kept rising after 2023Q3 while the fed funds rate plateaued, so end-of-cycle understates the eventual pass-through.
• Implied rates are average rates (expense over balances), not posted offer rates; there are no account-level rates in call reports. Balance averaging uses quarter-end stocks, not the intra-quarter averages holding companies report on the Y-9C.
• The denominator is interest-bearing deposits only. Betas measured on total deposits (including noninterest-bearing accounts) are mechanically lower; comparisons to published numbers must match the convention. The product panel differs again: its denominators are total product balances, and its transaction rate is a blended funding cost (see the product split section).
• Cycle betas exist only for banks filing in both endpoint quarters, so banks that failed or merged mid-cycle are absent from that cycle’s distribution.
• Foreign-office deposit expense (RIAD4172) and balances (RCFN6636) are included where reported, so large banks' betas reflect global deposit books, not only US retail pricing.