A deposit beta is the share of a change in market interest rates that a bank passes on to the rate it pays depositors. If the Federal Reserve raises its target by one percentage point and a bank raises its money market rate by 0.40 percentage points, that account’s beta is 0.40, or 40 percent. Betas are the most important assumption in most banks’ net interest income projections, because administered-rate deposits make up most of their funding and their pricing is a management decision rather than a contract. Deposit beta modeling is the work of estimating how betas will behave, by product and customer segment, over the speed and size of future rate moves, including lags, the difference between rising and falling rates, and the tendency of betas to climb as rates get higher. The 2022 to 2023 tightening showed how wrong the assumptions can be: deposit betas rose faster than in the previous cycle, customers moved money out of non-interest-bearing accounts, and banks that had modeled the deposits of the zero-rate years found their projections overstated. With the Federal Reserve raising its target range again on 16 September 2026, for the first time since 2023, up-cycle beta assumptions are no longer a scenario but a forecast.
This guide explains what a deposit beta is and the different ways it is measured, why betas matter to earnings, value and liquidity, what the last cycles showed, how betas are segmented and estimated, how lags, asymmetry and rate levels change them, how migration and decay fit with them, how the model should be governed, how to back-test it, and how internal audit tests it, with a worked example from a regional bank. The measures betas feed are compared in the site’s guide to EVE vs NII.
In this guide
- What a deposit beta is
- Why betas decide earnings, value and liquidity
- What the last cycles showed
- Segmentation: betas are not one number
- Estimation methods
- Lags, asymmetry and the level effect
- Setting assumptions for the cycle that began in September 2026
- Migration, decay and consistency with EVE
- Governance: the deposit model is a model
- Back-testing betas
- How internal audit tests deposit beta models
- Deposit study review checklist
- Worked example: Lakeshore Bancorp’s betas and the decay assumption
- Common mistakes
- Questions about deposit beta modeling
- Related guides
What a deposit beta is
The basic calculation is a ratio: the change in the deposit rate divided by the change in the market rate over the same period. The market rate is usually the federal funds rate or a short-term Treasury or SOFR rate. The simplicity hides several choices, and a beta quoted without them is ambiguous.
| Term | Meaning | Why it matters |
|---|---|---|
| Cumulative (cycle) beta | The change in the deposit rate from the start of a rate cycle to a point in it, divided by the change in the market rate over the same span | The measure most used to compare cycles and to set model assumptions |
| Period beta | The same ratio for a single quarter or month | Shows the lag: period betas are low early in a cycle and high late in it |
| Up-beta and down-beta | Betas measured separately for rising and falling rates | They differ, usually with down-betas lower and slower for most products |
| Interest-bearing beta | Measured on interest-bearing deposits only | The pricing decision itself |
| Total deposit beta | Measured on all deposits, including non-interest-bearing | Captures migration out of non-interest-bearing accounts, which raises the cost of funds even with no change in any posted rate |
| Product or segment beta | Measured for one product or customer segment | The level at which models should be built |
A bank can report a modest interest-bearing beta and a much larger increase in its total cost of deposits, because customers moved money from non-interest-bearing checking into money market accounts and time deposits. Both numbers are true, and a model that captures only the first will miss the second.
Why betas decide earnings, value and liquidity
Net interest income. In a rising-rate scenario, a bank’s asset yields rise as floating-rate loans reprice and new assets are booked at higher rates. Its funding costs rise with the betas. The difference between those two paths is most of the NII sensitivity result, so an understated beta flatters an asset-sensitive bank’s NII projection directly.
Economic value. Deposits that pay less than market rates, and stay, are valuable to a bank: the gap between the market rate and the deposit rate, earned year after year, is the deposit franchise. Academic research, notably a 2021 Journal of Finance paper by Itamar Drechsler, Alexi Savov and Philipp Schnabl, argues that this franchise hedges banks’ long fixed-rate assets, because deposit spreads widen as rates rise. The hedge only works if betas stay low and balances stay put, which is why the beta and decay assumptions together decide how much of the loss on long assets the deposits offset in an EVE calculation.
Liquidity. Rate-sensitive balances leave when a bank does not match competitors’ rates. The same customers who produce high betas produce outflows when the bank holds rates down. Liquidity stress assumptions and beta assumptions describe the same depositors and should be consistent with each other, a point the site’s liquidity risk guide develops.
What the last cycles showed
| Period | Market rates | What happened to deposits |
|---|---|---|
| 2015 to 2019 | Federal funds target raised from near zero to 2.25 to 2.50 percent over three years | Cumulative interest-bearing deposit betas rose slowly, reaching about 40 percent by the end of the cycle, according to New York Fed research |
| March 2022 to July 2023 | Target raised from near zero to 5.25 to 5.50 percent, the fastest tightening in four decades | The New York Fed found the cumulative interest-bearing beta reached about 40 percent by the fourth quarter of 2022, a level the previous cycle took three years to reach; S&P Global Market Intelligence reported in October 2023 that the industry beta had passed both previous cycles; non-interest-bearing balances fell as customers moved money into interest-bearing accounts, time deposits and money market funds |
| March 2023 | Silicon Valley Bank, Signature Bank and later First Republic fail | Uninsured, concentrated deposits left in days; a later New York Fed retrospective found betas kept rising at a steady pace rather than jumping after the failures, and that super-regional banks paid up most to keep deposits |
| September 2024 to December 2025 | Target cut from 5.25 to 5.50 percent to 3.50 to 3.75 percent, 175 basis points in all | Deposit costs fell more slowly than they had risen for most products, as time deposits repriced only at maturity and posted rates on savings and checking had little room to fall |
| September 2026 | Target raised 25 basis points to 3.75 to 4.00 percent, the first increase since July 2023 | A new up-cycle begins from a rate level where depositors are already alert to what their money earns |
Two lessons came out of the 2022 cycle for modelers. Betas depend on how fast rates rise as well as how far: a quick cycle compresses the lag, so period betas late in the cycle are high. And betas rise with the level of rates. A 2023 Dallas Fed working paper on deposit convexity estimated that the aggregate beta rises from about a quarter near zero rates to around 60 percent when rates average close to 5 percent, which makes deposits behave like instruments with embedded options rather than fixed-proportion liabilities. A model calibrated on the zero-rate years understates both effects.
Segmentation: betas are not one number
A single bank-wide beta averages together depositors who behave completely differently. Models are built by segment, and the segmentation is one of the first things internal audit should test.
| Segmentation dimension | Typical distinction | Why it changes the beta |
|---|---|---|
| Product | Non-interest-bearing checking, interest checking, savings, money market, time deposits | Time deposits reprice at new-issue rates; checking rates barely move |
| Customer type | Retail, small business, commercial, public funds, brokered | Commercial treasurers and public entities manage cash actively; some public funds are contractually indexed to a market rate |
| Balance tier | Small balances versus large balances in the same product | Large balances are rate-sensitive and often above the insurance limit |
| Channel | Branch versus digital-only accounts | Digital accounts attract rate-shoppers and move fastest |
| Relationship depth | Operating accounts with lending or treasury services versus rate-only accounts | Operational balances are sticky; rate-only balances are not |
| Pricing structure | Posted rates versus exception pricing or indexed rates | Exception pricing is where betas rise first |
Segments should be few enough to estimate reliably and distinct enough that they behave differently. A segment that mixes indexed public funds with posted-rate commercial money market balances will produce a beta that fits neither.
Estimation methods
| Method | How it works | Strength | Weakness | Audit question |
|---|---|---|---|---|
| Historical cycle betas | Cumulative betas measured over past cycles, by segment | Simple, transparent, anchored in the bank’s own data | Past cycles may not resemble the next one; small number of cycles | Which cycles were used, and why were others excluded? |
| Regression on rate changes | Deposit rate changes regressed on market rate changes, with lagged terms, by segment | Estimates both beta and lag from data | Sensitive to the period chosen; assumes a constant relationship | Is the relationship stable across periods, and how was the lag structure chosen? |
| Partial adjustment or error correction | Deposit rates adjust gradually toward a long-run spread to market rates | Captures catch-up after lags | More complex; harder to explain to ALCO | Are the long-run spreads plausible, and are they re-estimated? |
| Level-dependent betas | The beta rises as the rate level rises | Matches the 2022 experience and the research on deposit convexity | Needs a model of the level effect that is hard to estimate from one bank’s data | How does the model behave in a scenario starting from today’s rate level? |
| Market and competitor data | Peer rates and market surveys used to calibrate or check assumptions | Shows what competitors are paying, which drives outflows | Peer pricing reflects peer strategies | How close are the bank’s assumptions to observed market pricing? |
| Management pricing overlay | Management’s stated pricing strategy, applied to model outputs | Reflects plans the data cannot see | Can be optimistic, and is where bias enters | Is the overlay documented, approved and back-tested? |
Most banks combine two or three: regression or cycle betas by segment for the core estimate, a level effect or scenario-specific betas for large moves, and a documented overlay for strategy. The combination should be written down in a deposit study that ALCO approves and model risk management validates.
Lags, asymmetry and the level effect
Three dynamics distinguish a realistic beta model from a static one.
Lags. Deposit rates follow market rates with a delay, and the delay differs by product. Time deposits follow as they mature and are replaced; money market rates follow within weeks when competition is intense; savings and checking rates may not move for quarters. A model that applies a full-cycle beta instantly overstates early funding costs and understates later ones, which is why the NII result in months 13 to 24 often differs sharply from months 1 to 12.
Asymmetry. Banks raise deposit rates reluctantly and cut them quickly where competition allows, but for many products the cut is limited: a savings account paying 0.20 percent cannot fall by 1.75 percentage points. Time deposits keep paying their contractual rate until maturity. The result is that realized down-betas are usually lower and slower than up-betas for most products. A model that assumes symmetric betas overstates the funding-cost relief in falling-rate scenarios and so understates the NII risk of rate cuts.
Level effect. Near zero, depositors have little to gain from moving money and banks have little reason to pay up. As rates rise, both change, and the beta climbs. A new up-cycle that begins at 3.75 percent does not start where the 2022 cycle started at zero, and betas estimated on the early months of 2022 are a poor guide to the early months of a cycle beginning from today’s level.
Setting assumptions for the cycle that began in September 2026
The Federal Reserve’s increase on 16 September 2026 turned up-rate scenarios from a hypothetical into a forecast, and it arrived at a point that no bank’s history describes well. The last up-cycle started at zero, with deposits swollen by pandemic-era inflows and banks holding more liquidity than they needed. This one starts at 3.50 to 3.75 percent, after a falling-rate period in which many deposit rates came down less than market rates, so the spread between what depositors earn and what they could earn elsewhere is narrower than it was in early 2022. Money market funds and high-yield digital accounts are established alternatives, and depositors learned in 2022 and 2023 how easily money can move.
None of that says what the new cycle’s betas will be, and a model should not pretend to know. It does suggest five disciplines for the assumptions:
- Start from where the last cycle ended, not where it began. Early-cycle betas estimated on 2022 describe depositors who were earning nothing and expected to keep earning nothing.
- Show a faster pass-through scenario. Alongside the base assumption, run NII with betas at the level reached late in the last cycle from the first quarter, and show the difference to ALCO.
- Watch exception pricing. The volume and rates of exception-priced accounts are the earliest sign that posted-rate betas are about to move.
- Back-test monthly in the first quarters. A fast cycle can make a quarterly back-test too slow to matter.
- Revisit migration. Non-interest-bearing balances that stayed through the falling-rate period are the ones most likely to move if the gap to market rates widens again.
Migration, decay and consistency with EVE
Betas describe pricing; migration and decay describe balances. Both matter, and they have to be consistent.
- Migration is the movement of money between products as rates change: from non-interest-bearing checking into interest-bearing accounts, and from savings and money market accounts into time deposits. It raises the bank’s cost of deposits even if no posted rate changes, so NII models need an explicit migration assumption, not only product betas.
- Decay is the rate at which balances leave the bank, usually expressed as an annual runoff rate or an average life. It drives the value of non-maturity deposits in an EVE calculation.
- Consistency. A segment assumed to be highly rate-sensitive for NII, with a high beta, cannot sensibly be assumed to be very long-lived and stable for EVE. Customers who demand market rates are the customers who leave when they do not get them. The two assumptions should come from the same deposit study, estimated on the same data and reviewed together.
The consistency test is simple and often failed, because betas and decay rates are frequently estimated at different times, by different people, for different reports. The worked example below shows what that looks like. The effect of decay assumptions on EVE is covered in the site’s guide to EVE vs NII.
Governance: the deposit model is a model
The deposit beta and decay models are models in the sense of the banking agencies’ model risk guidance, and among the most consequential ones a bank runs, since they feed NII projections, EVE, liquidity stress testing and often funds transfer pricing.
The Federal Reserve, OCC and FDIC replaced the 2011 model risk guidance with revised guidance on 17 April 2026 (SR 26-2 and OCC Bulletin 2026-13). The revision keeps the three elements of validation, conceptual soundness, ongoing monitoring and outcomes analysis, and the principle of effective challenge; it narrows the definition of a model to complex quantitative methods, and says the guidance is most relevant to banking organizations with more than 30 billion dollars of assets. A deposit model built on regressions and scenario logic remains squarely a model. For a bank below 30 billion dollars, the criteria are the bank’s own model risk policy, as the site’s guide to auditing model risk management explains.
The 2010 interagency advisory on interest rate risk management adds expectations specific to these assumptions: document and support them, explain changes between periods and their effect, run sensitivity analysis to show which assumptions matter most, and give rate-sensitive, higher-cost deposits such as brokered and internet deposits higher decay rates than core relationships. Together, the two documents describe what a deposit study should contain and what ALCO should see.
Back-testing betas
Back-testing compares what the model assumed with what happened. For betas, it is straightforward and underused.
| Segment | Assumed beta for the period | Realized beta | Difference | Within tolerance? | Explanation and action |
|---|---|---|---|---|---|
| One row per segment | From the model version used for the period | Deposit rate change divided by market rate change, measured the same way as the assumption | Realized minus assumed | Against a tolerance set in advance, for example 0.10 | Why it differed; whether the assumption changes |
Four disciplines make back-testing useful. Measure realized betas exactly the way the assumption is defined, cumulative or period, interest-bearing or total. Back-test every quarter during a rate cycle, not once a year. Back-test migration and balances as well as rates, since a bank can price exactly as modeled and still lose balances it assumed would stay. And report the results to ALCO with the explanation, not just the numbers, so that a persistent miss leads to a model change rather than a footnote.
How internal audit tests deposit beta models
| Area | Test | Evidence | Red flags |
|---|---|---|---|
| Data | Deposit rates and balances by segment reconcile to the core system and general ledger | Data extracts, reconciliations | Rates taken from posted-rate sheets rather than rates actually paid |
| Segmentation | Segments are distinct in behavior and cover all balances | Segment definitions; balance coverage | Indexed or public funds balances mixed into posted-rate segments |
| Calibration period | The estimation window includes the most recent full rate cycle | Deposit study | Studies built on 2015 to 2021 only |
| Method | The method suits the segment and is documented | Deposit study; model documentation | One bank-wide beta; unexplained lag choices |
| Asymmetry and level effects | Down-betas and up-betas are estimated separately; the model responds to the starting rate level | Model documentation; scenario outputs | Symmetric betas; identical betas whatever the starting rate |
| Overlays | Management overlays are documented, approved and back-tested | ALCO minutes; overlay log | Overlays that always reduce projected funding costs |
| Consistency | Betas, migration and decay assumptions come from the same study and tell the same story | Deposit study; EVE and liquidity assumptions | High betas paired with long average lives for the same segment |
| Back-testing | Quarterly comparison of assumed and realized betas and balances | Back-testing reports | Annual only; misses reported without action |
| Sensitivity | NII and EVE results shown under higher and lower beta assumptions | ALCO packs | Single-point results |
| Change control | Every assumption change approved, with its effect on results and limits | Change log; before-and-after results | Changes that coincide with limit pressure |
| Validation | Independent validation of the deposit models, including the study behind them | Validation reports; issue log | Validation scope that excluded the deposit study |
| Use | The same assumptions are used for NII, EVE, liquidity stress testing and funds transfer pricing, or differences are explained | Assumption inventory | Different betas in different models for the same deposits |
The consistency and use tests are where internal audit adds most, because they cut across teams that each own one model. The site’s guide to vendor model validation covers the case where the deposit model sits inside a purchased asset-liability management engine.
Deposit study review checklist
Scope. Which deposits are covered, by segment, with balances reconciled to the ledger.
Data. The period used, why it was chosen, and whether it includes the most recent full rising and falling cycles and any stress period.
Betas. Up-betas and down-betas by segment, cumulative and period, with lags; the method used and its fit.
Level effect. How betas change with the starting rate level, and what the study assumes for a cycle beginning from current rates.
Migration. Assumed movement between products, including out of non-interest-bearing accounts, with support.
Decay and average life. Runoff rates by segment, with support, and consistency with the betas for the same segment.
Overlays. Management adjustments, their rationale and their approval.
Back-testing. Assumed versus realized betas, migration and balances for the last cycle, with explanations.
Sensitivity. NII and EVE under higher and lower assumptions for the most influential segments.
Approval and validation. ALCO approval, model risk review, and open findings.
Worked example: Lakeshore Bancorp’s betas and the decay assumption
Lakeshore Bancorp, the nine-billion-dollar regional bank with 60 branches used across this site, funds itself with about 7.5 billion dollars of deposits. Its 2026 audit of interest rate risk management, whose EVE finding is described in the site’s guide to EVE vs NII, also tested the deposit beta model.
| Segment | Balance, June 2026 | Model beta before 2024 update | Realized beta, 2022 to 2024 up-cycle | Model beta after update | Realized down-beta, September 2024 to June 2026 |
|---|---|---|---|---|---|
| Non-interest-bearing | 1.9 billion | Not applicable | Not applicable | Not applicable | Not applicable |
| Interest checking | 1.3 billion | 0.10 | 0.14 | 0.15 | 0.05 |
| Savings | 0.9 billion | 0.15 | 0.21 | 0.20 | 0.08 |
| Money market | 1.6 billion | 0.45 | 0.61 | 0.60 | 0.41 |
| Time deposits, new issue | 1.35 billion | 0.80 | 0.83 | 0.85 | 0.64 |
| Public funds, within commercial money market | 0.45 billion | 0.45 | 0.92 | 0.60 | 0.88 |
What the bank had done well. After its 2023 NII projections overstated actual NII for three consecutive quarters, the ALM team re-estimated betas in 2024 from the bank’s own 2022 to 2024 data, by segment, using a regression with lags and a check against cycle betas. The updated betas were close to what had happened in the up-cycle, within the bank’s back-testing tolerance of 0.10 for every segment.
What the audit found. Three findings came from the beta model.
- Beta and decay assumptions contradicted each other. The betas were estimated on the 2022 to 2024 cycle, in which interest checking and savings balances proved rate-sensitive and interest checking balances fell by 9 percent. The average-life assumption for the same balances, extended to 6.8 years in late 2025, was estimated on 2015 to 2021 data, in which they looked almost immobile. The same depositors were modeled as rate-aware for NII and as immovable for EVE. This was part of the High finding reported in the EVE audit.
- Down-betas were assumed symmetric, and back-testing had not caught it. The model applied the up-cycle betas to falling-rate scenarios. Back-testing was annual, and the 2025 back-test compared the updated betas with the up-cycle they had been estimated on, so they matched by construction; no back-test had been run against the falling rates since September 2024. The realized down-betas were far lower for every product. Re-run with the realized down-betas, the minus 200 basis point NII result in the June 2026 ALCO package moved from minus 5.8 percent to minus 8.4 percent, still inside the minus 10 percent limit but much closer to it. Rated Medium.
- Public funds were mixed into commercial money market. About 450 million dollars of public funds deposits, many indexed by agreement to a market rate, sat inside the commercial money market segment. Their realized beta was 0.92. Blended with the posted-rate balances, the segment’s realized beta was 0.66, within tolerance of the 0.60 assumption, which is why back-testing had not flagged it. Separating them raised the projected cost of funds in up-rate scenarios. Rated Medium.
The response, and the timing. Management agreed to rebuild the deposit study so that betas, migration and decay are estimated together on data including the full 2022 to 2026 period; to estimate down-betas separately; to split public funds into their own segment with contractual indexation modeled directly; and to add a level-dependent overlay for up-rate scenarios. By the time the report reached the audit committee, the Federal Reserve had raised its target range for the first time since 2023. The committee’s question was the right one: what would NII look like if the new cycle’s betas started where the last cycle’s ended rather than where it began? The ALM team’s answer, presented at the next ALCO, showed the plus 200 basis point NII benefit falling from 3.1 percent to 1.4 percent, which changed the conversation about whether the bank needed the swap program it had been discussing.
Common mistakes
| Mistake | Consequence | Better practice |
|---|---|---|
| One beta for all deposits | Rate-sensitive balances are averaged with sticky ones, and the model fits nothing | Segment by product, customer type, tier and pricing structure |
| Calibrating on the zero-rate years | Betas and outflows are understated for any cycle starting above zero | Include the 2022 to 2026 period |
| Symmetric up and down betas | Falling-rate NII risk is understated | Estimate down-betas separately |
| Ignoring migration | Total deposit costs rise faster than product betas imply | Model movement out of non-interest-bearing accounts explicitly |
| Betas and decay from different studies | The same depositors are rate-sensitive in one model and permanent in another | One deposit study, one data set, reviewed together |
| Annual back-testing | Misses during a fast cycle go unnoticed for quarters | Back-test quarterly during rate cycles |
| Undocumented pricing overlays | Management optimism enters the model unseen | Document, approve and back-test every overlay |
| Different betas in different models | NII, liquidity and transfer pricing disagree about the same deposits | Maintain one assumption inventory |
Questions about deposit beta modeling
What is a deposit beta?
The share of a change in market interest rates that a bank passes on to the rate it pays on deposits. A beta of 0.40 means the deposit rate moves 40 basis points for every 100 basis point move in the market rate.
How do you calculate a cumulative deposit beta?
Divide the change in the deposit rate from the start of a rate cycle to the point of measurement by the change in the market rate, usually the federal funds rate, over the same span. Measure by product or segment, and state whether the result covers interest-bearing deposits only or all deposits.
Why are deposit betas higher when rates are higher?
Because depositors have more to gain from moving money and competitors have more reason to pay up. Research from the Dallas Fed found aggregate betas rising from about a quarter near zero rates to around 60 percent at rates near 5 percent.
Are deposit betas the same when rates fall?
Usually not. For most products, banks cut deposit rates more slowly or by less than they raised them, time deposits keep their rates until maturity, and low posted rates have little room to fall, so realized down-betas are generally lower than up-betas.
What is the difference between deposit beta and deposit decay?
Beta describes how the rate paid on deposits responds to market rates. Decay describes how quickly the balances themselves leave. Betas drive NII projections; decay drives the value of deposits in EVE. Both should come from the same analysis of the same depositors.
How often should deposit beta assumptions be updated?
Back-tested every quarter during a rate cycle, and re-estimated at least annually or whenever back-testing shows a persistent miss, with every change approved by ALCO and its effect on NII, EVE and limits reported.
Related guides
- EVE vs NII — the two measures deposit assumptions feed.
- Interest rate risk: an overview — the concepts behind the models.
- IRR vs IRRBB vs structural interest rate risk — the vocabulary.
- The complete guide to interest rate risk — tools and internal audit perspectives.
- Liquidity risk — the outflow side of the same depositors.
- Auditing model risk management — the deposit model inside the inventory, under the April 2026 guidance.
- Vendor model validation — deposit models inside purchased ALM engines.
- Why internal audit watches Fed rate decisions — the policy moves that drive betas.
- Bank runs in history — what happens when deposits decay faster than modeled.
- Risk appetite statements — where interest rate risk limits belong.
- The model risk library — every guide on model risk.
- How to Audit Model Risk Management: End-to-End Program — a work program for MRM under SR 26-2, from inventory to validation to vendor models
- Liquidity Stress Testing: Scenarios, Horizons, Death Spirals — scenario design, survival horizons and the death spiral loop.
Leave a Reply