Initial full research published September 27, 2026. Historical events retain their dates; hypothetical examples and analytical recommendations are labeled.
The useful question is whose deposits reprice
Deposit beta measures how much a deposit rate changes relative to a reference rate. A 100-basis-point market-rate increase followed by a 40-basis-point deposit-rate increase produces a 40% beta. That arithmetic is simple; identifying the relevant balance, rate and period is the difficult part. A portfolio of operating accounts, online savings, promotional certificates and brokered funding does not have one stable behavioral coefficient.
The New York Fed's April 2023 analysis distinguishes deposit pricing behavior across banks and explains why funding needs and competitive conditions matter. It is historical research, not a September 2026 observation. Its companion analysis explains the economic value of a stable, relatively inexpensive deposit franchise while emphasizing depositor behavior. [1][2] This article uses those mechanisms to develop an analytical framework; it does not estimate any bank's current beta.
Separate three movements before interpreting the average
First, a bank can change the rate paid on an unchanged account. Second, customers can migrate from checking into higher-rate savings or certificates at the same bank. Third, customers can leave, forcing the bank to shrink assets or replace funding. All three affect economics, but only the first is pure repricing. An average rate can rise even when every posted product rate is unchanged because the mix has shifted.
The denominator also matters. Interest expense divided by interest-bearing deposits answers a different question from expense divided by all deposits. A bank losing noninterest-bearing balances can report a sharp increase in its total-deposit cost even if its interest-bearing product prices barely move. Quarterly expense divided by average balances, appropriately annualized, should not be compared directly with an end-of-quarter advertised savings yield.
A marginal beta describes a recent interval. A cumulative beta compares the present with the beginning of a rate cycle. Either can become unstable when the reference-rate change is small, crosses zero or reverses direction. A reported beta above 100% may reflect delayed repricing or mix change rather than an implausible contractual response. State the start date, end date, benchmark and balance definition beside every estimate.
Worked example: a lag can overwhelm an initial saving
Consider a hypothetical bank with $1 billion of interest-bearing deposits. Assume a 100-basis-point policy-rate increase, unchanged balances and a 30% immediate beta. The annualized incremental deposit expense is $1 billion × 1% × 30%, or $3 million. If pricing catches up to a 60% cumulative beta over the following year, the eventual annualized increase becomes $6 million. Neither number is necessarily the first year's actual expense: that depends on when each repricing occurs.
Now assume $200 million of noninterest-bearing deposits leave and are replaced at a hypothetical 4% annual rate. That replacement adds $8 million of annualized expense, before fees or liquidity costs. A model focused only on the original $1 billion would miss a larger exposure than the modeled repricing effect. These assumptions are illustrative and are not forecasts of current policy rates or any institution's balances.
On a rate decline, the response need not be symmetrical. Fixed-term certificates reprice at maturity, some savings rates face competitive resistance to cuts, and checking cannot fall much below zero. Meanwhile floating-rate loans may reset quickly. A bank can therefore experience falling asset yields before receiving the modeled funding benefit. Lag belongs in monthly cash flows, not in a verbal caveat appended to a static ratio.
| Illustrative funding change | Annualized incremental expense |
|---|---|
| $1bn; 100bp benchmark increase; 30% beta | $3m |
| Same balances; 60% cumulative beta | $6m |
| $200m zero-cost balances replaced at 4% | $8m, additional to repricing |
Earnings value and liquidity value require different tests
For net interest income, project monthly balances and rates by product, customer segment and maturity. For economic value, consider the duration and persistence of the deposit relationship. A low-cost deposit franchise can support considerable going-concern value, but that value depends on customers remaining and the bank continuing to operate. It is not a cash asset that can automatically be sold to meet tomorrow's withdrawals. [2]
An analytical stress should connect pricing and runoff. Holding deposit balances fixed while sharply underpaying competing alternatives can overstate earnings resilience. Conversely, assuming immediate full repricing and severe runoff simultaneously may double-count customer responses unless the scenario explains why both occur. At least one stress should include wholesale replacement costs, collateral requirements and borrowing capacity that remains usable after other obligations consume collateral.
Segmenting customers is useful without pretending that every account has a reliably estimated lifetime. Operational accounts with recurring payment activity, rate-shopping balances and concentrated uninsured relationships may behave differently. Those are hypotheses to test against observed flows, not labels that establish stability. A relationship manager's confidence should not substitute for reconciled account-level retention data.
Controls and the cost of better estimates
Recommended controls begin with a monthly bridge from ledger interest expense to modeled expense. Split changes into price, volume and mix. Retain actual opening and closing balances, promotional expirations and negotiated exceptions. Compare predicted migration with observed transfers between products. Backtesting should include rising and falling rate periods where available, and explicitly identify periods for which the bank lacks relevant experience.
More granular models require clean customer identifiers, product histories and consistent treatment of mergers and acquired accounts. Those investments have costs. For a small portfolio, a transparent segmented sensitivity may be more useful than a complex model fitted to a short, unusual cycle. Management still needs a range of plausible outcomes and an explanation of which assumptions drive it.
What would change the conclusion
Evidence of stable customer retention after sustained relative underpricing would support a lower behavioral beta for that segment. Faster-than-modeled migration, shorter certificate renewals, concentrated outflows or costly emergency replacement would weaken it. None can be inferred from a single quarter's net interest margin. The decision-relevant result is an earnings and liquidity range with dated assumptions, supported by observed behavior and reconciled funding costs.