1Name the behaviours a bank models in its banking book (prepayment, deposit rates, deposit volumes, drawdowns and outflows) and the data each one needs.
2Say which rules let a bank use its own behavioural assumptions and which prescribe them, for interest rate risk and for liquidity.
3Run a fitted deposit model through the Basel caps and explain why the caps, not the data, set the number.
4Describe how behavioural assumptions are governed: a register with an owner, a source, a version and a backtest.
Why it matters
Twenty years of RiskAtoms Bank plc's current-account balances say 97% of the £22bn is core, with a behavioural life of 30 years. The bank's books, set in L3.6.09: Non-maturity deposits, say 81% and an average maturity of 2.5 years. The Basel standard says no more than 90% and 5 years. Three numbers for one balance. This lesson sorts out which one runs which figure, and who has to sign for it.
Synthetic bank dataBooked figures standRiskAtoms Bank plc's banking book, to 31 December 2025
Synthetic: twenty years of month-end balances and client rates for the bank's four non-maturity deposit segments, generated by the course from fixed seeds against the real Bank Rate and built to end exactly on the booked balances and rates of M3.6: Interest rate risk in the banking book and ALM. A run on the bank's name in March and April 2023 is invented and placed in the real turmoil of that spring. No row is a real depositor, and unlike any real bank we know the true betas behind the data.
Synthetic: every two- and five-year fix the bank wrote from January 2012 to December 2025, 6,720 loans followed month by month (214,944 loan-months, 4,998 full prepayments), generated from a known prepayment model against the real quoted rates. The coupons of the fixes still on the book match the booked 4.6% and 3.8%, and the long-run prepayment rate matches the booked 8%. No row is a real borrower, and the true model is known, so the fit can be checked against it.
A bank's contractsContractual profile: the cash flows and repricing dates written in the contract, before any behaviour. say a current account can leave tomorrow and a five-year fixed mortgage stays five years. Neither is true on average. Customers hold options, and how they use them is a behavioural assumptionBehavioural assumption: how customers are expected to use the options in their products (to repay early, withdraw, draw down or stay), as opposed to what the contract says.. L3.6.02: Where interest rate risk in the banking book comes from named the option risk these create; L4.5.05: The IRRBB standard showed how the IRRBB standard treats them. This module builds the models behind them. Five behaviours matter most:
Prepayment: borrowers repay fixed-rate loans early, faster when rates fall.
Deposit rates: how much of a move in Bank Rate the bank passes to savers, and how fast (the beta).
Deposit volumes: how much of a balance stays (the core) and for how long.
Drawdowns: how much of a committed facility customers draw in a stress.
Outflows: how fast deposits leave when the bank's name is in doubt.
Each needs the bank's own history (balances, rates paid, prepayments and drawdowns, month by month) and the market it reacts to. The map below sets each of RiskAtoms Bank's books against the model this module builds for it.
RiskAtoms Bank plc at 31 December 2025: the books of Modules 3.5 and 3.6 and the behaviour each carries.
Book
Size
What customers do
Model in this module
Current accounts
£22bn
Rate pass-through; how long balances stay
Betas and floor; core share and life; replicating portfolio
Instant-access savings
£10bn
Rate pass-through; outflows when the bank lags
Betas; core; volume response to the rate gap; stressed run-off
Operational and call accounts (companies)
£8bn
Concentrated, uninsured, fast to leave
Betas; concentration; stressed run-off
Fixed-rate mortgages (2- and 5-year)
£24bn
Prepayment when refinancing pays; reversion
Hazard (S-curve) model; option value on rate paths
Committed facilities
£8bn undrawn
Drawn in a stress
Stressed drawdown from utilisation history
Liquid asset buffer
£8.5bn unencumbered
What it sells for, how fast
Market liquidity: exit cost, days to sell
The whole balance sheet
£67bn
NII, EVE, survival, transfer prices
Projections, cash flow stress, FTP, backtests and register
Two families of rules
For interest rate risk in the banking book (IRRBB), the bank measures with its own assumptions. The Basel standard (SRP31, Principle 5) asks that key behavioural and modelling assumptions be fully understood, conceptually sound and documented, and rigorously tested. It lists the products whose behaviour matters (SRP31.41): fixed-rate loans that can be prepaid, loan commitments, term deposits that can be redeemed early, and non-maturity deposits, for which banks “should document, monitor and regularly update key assumptions”. Assumptions should be consistent with historical experience (SRP31.42) and reviewed at least once a year (SRP31.47). Beside this, the standardised framework turns the same deposits into numbers with hard limits: banks split each category into coreCore deposits: the part of a non-maturity deposit balance that is stable and unlikely to reprice even when market rates move a lot. and non-core, non-core reprices overnight, and the core share and its average maturity are capped (BCBS 368, paragraphs 110 to 115 and Table 2). Prepayment there is the bank's base rate times a fixed multiplier, 0.8 when rates rise and 1.2 when they fall (paragraph 122, Table 3).
The Basel caps on core non-maturity deposits (BCBS 368, April 2016, Table 2; the Basel Framework's SRP31 keeps them).
Category
Cap on core share
Cap on average maturity of the core
Retail transactional
90%
5.0 years
Retail non-transactional
70%
4.5 years
Wholesale
50%
4.0 years
For liquidity, the roles are reversed. The LCR's run-off rates are prescribedPrescribed rate: a run-off or drawdown rate set by the rule itself, the same for every bank.: 5% for stable retail deposits, 10% for less stable, 25% for operational deposits and 40% for other non-financial corporate deposits over 30 days (Basel LCR40). No behavioural model changes the ratio. The bank's own view goes into its ILAAPILAAP (internal liquidity adequacy assessment process): the bank's own assessment of its liquidity and funding risks, with its own stress assumptions, reviewed by the supervisor.: in the UK, the PRA expects a firm to make “appropriate assumptions, both quantitative and qualitative” about its key risk drivers, starting with the likely run-off of each part of its retail book (SS24/15, paragraph 2.18A), and may set Pillar 2 liquidity guidance beyond the LCR buffer for risks the LCR does not capture.
Sort itPrescribed by the rule, or the bank's own?
For each figure, decide whether the rule sets it for every bank or the bank estimates it, within limits.
The 10% 30-day run-off on less stable retail deposits in the LCR
The savings run-off in the bank's own 90-day name-specific stress
The core share of current accounts in the bank's internal IRRBB measure
The 1.2 multiplier on the base prepayment rate when rates fall, in the standardised framework
The pass-through beta on instant-access savings
The 15% of Tier 1 threshold for the EVE outlier test
What the caps do to a fitted model
The engine's deposit model, fitted to each segment's twenty years of balances, finds cores far above the booked ones. Then the caps cut every one of them, on the core share and on the maturity:
The engine's volume model at 99% over one year on synthetic balances, January 2006 to December 2025; booked = Module 3.6's stable share × (1 − β) and half its slotting period.
Segment
Category
Fitted core
Fitted life
Cap
Core used
Maturity used
Booked
Personal current accounts
Retail transactional
97.2%
30.0 years
90%, 5.0y
90%
5.0 years
81%, 2.5y
Instant-access savings
Retail non-transactional
92.4%
30.0 years
70%, 4.5y
70%
4.5 years
30%, 1.5y
Operational corporate deposits
Wholesale
91.2%
15.3 years
50%, 4.0y
50%
4.0 years
50%, 1.5y
Corporate call accounts
Wholesale
78.2%
8.5 years
50%, 4.0y
50%
4.0 years
10%, 1.0y
In IRRBB, after the caps, the statistics hardly matter. That is what the caps are for: a twenty-year history with steady growth would let a bank call almost every deposit core and hedge it like a thirty-year bond, yet deposits that sat still for years can leave in days, as 2023 showed (L6.6.07: Stressed outflow and drawdown models). The caps put a ceiling on how much interest rate risk a bank can hide in a model. Across the four segments the average repricing date of the £40bn is 17.3 years on the raw fit, 2.57 years after the caps and 1.33 years as booked. The booked figures stand; the gap between them is a sensitivity, which later lessons of this module carry into NII, EVE and transfer prices.
Governance: the register
Behavioural models are models. Each assumption goes on an assumption registerAssumption register: the list of every behavioural assumption a bank's figures use, with its value, source, owner, approval and the result of its latest backtest. with its value, the model and data behind it, an owner, a version and a backtest. RiskAtoms Bank's register has 14 rows for this module: a beta, a core and a run-off rate for each deposit segment, and the fitted and booked prepayment rates. Three of them:
RiskAtoms Bank plc's assumption register, three of its rows (synthetic data). Their backtests and traffic lights are in the backtesting lesson of this module.
Assumption
Value
Model
Owner
Version
Instant-access savings: deposit beta (up / down)
0.68 / 0.76
Asymmetric lagged betas (engine)
ALM (Treasury)
2025-12 v1
Instant-access savings: core share
Model 93.0% (99%, one year; data to Feb 2023); in the book after caps 70%, 4.5 years
Volume model, 99% one-year (engine)
ALM (Treasury)
2023-02 v1
Fixed-rate mortgages: booked constant CPR
8% (× 0.8 / 1.2 in shocks)
Constant (m3-6)
ALM (Treasury)
Level 3
L6.6.11: Backtesting behavioural assumptions backtests every row. In the UK the PRA's model risk principles (SS1/23) apply to banks with internal model approval for capital, and ask for models to be identified and tiered, governed, developed with tests, independently validated, and covered by mitigants such as overlays where they are uncertain. A behavioural model that drives IRRBB, the ILAAP and transfer prices is a model in that sense.
The formal version
From a balance history to the repricing profile
For a segment with month-end balances Bt, monthly log changes rt = ln Bt − ln Bt−1, horizon h (years), confidence q and category c, the engine's volume model (fitted in L6.6.05: Deposit volumes and replicating portfolios) and deposit book give:
μ=12rˉ,σ=12sd(r),z=Φ−1(q)core=min(1,emin(μ,0)h−zσh,t,k≤12hminBtBt+k),λ=max(zσ−μ,301),L=min(30,1/λ)coreused=min(core,cˉc),Mused=min(L,Mˉc)overnight=B(1−coreused)+Bcoreusedβ,ladder=Bcoreused(1−β) spread evenly over [0,2Mused]average repricing date=coreused(1−β)Mused
c̄c and M̄c are the caps of category c (BCBS 368 Table 2); β is the pass-through beta (L6.6.04: Deposit rates and betas), which reprices overnight; L is the behavioural life. The booked rule of M3.6: Interest rate risk in the banking book and ALM is simpler: core = stable share × (1 − β), capped, spread evenly over its slot. The EU and, from 2027, the UK add a cap on the average repricing date of the whole balance: five years per currency (EBA/GL/2022/14 paragraph 111; PRA SS31/15 paragraph 2.9B(vi)). Sources: BCBS (2016); the Risk Engine's deposit calibration and deposit book.
Try it
Pick a segment and move the confidence and horizon of the volume model. With the caps applied, the core used does not move, because every fitted core sits above its cap; switch them off to see what the data alone would claim. Watch the average repricing date: it is what an EVE shock acts on.
Try itFit a deposit segment, then apply the Basel caps
Segment
Basel caps
Fitted core share92.4%
Behavioural life (fitted)30.0 years
Basel cap (retail non transactional)70%, 4.5 years
Core share used70.0% (capped)
Average maturity of the core used4.5 years (capped)
Booked core and average maturity30%, 1.5 years
Average repricing date of the balance, model0.96 years
The same, booked slotting0.45 years
Monthly drift and volatility of the balance (a year)5.0%, 3.4%
Where the £10bn reprices
Overnight: volatile part 30%Overnight: core × β (β 0.70) 49%Core ladder to 9.0 years 21%
Booked (Module 3.6): core slotted evenly over 3 years
Overnight 70%Core ladder to 3 years 30%
The engine's volume model: core = the smaller of exp(min(μ, 0)h − zσ√h) and the worst fall of the balance over any window of h, on twenty years of synthetic month-end balances; life = 1 / max(zσ − μ, 1/30), at most 30 years. The booked core is the stable share × (1 − β), already net of the pass-through, so the booked bar has no “core × β” part. Average repricing date = the core ladder's share × its average maturity.
In the Risk Engine
Analysis → ALM, the Deposits loader and Deposit behaviour calibration
Load a deposit book and calculate: the Deposits table lists each segment with its category, balance, “core share used” and “core maturity used”, and marks a value cut to its Basel cap as “(capped)”. Open Deposit behaviour calibration, load a monthly history (balance, client rate, market rate by segment), choose the confidence, horizon, lags and backtest months, and Calibrate: the table gives each segment's beta, floor, core share and life. “Write to book” puts them into the deposit book, where the caps apply. The Loans loader shows each loan segment's CPR, and the Assumption register lists every behavioural assumption with its source object, owner, status and backtest.
Worked example
Instant-access savings through the caps
From twenty years of balances to the repricing profile
Step 1 of 4
1
The fit
The £10bn of instant-access savings grew by 5.0% a year on average with a volatility of 3.4%. At 99% over one year the model core is 92.5%; the worst one-year fall in the history leaves 92.4%. The smaller, 92.4%, is the fitted core. Growth offsets most of zσ, so the run-off rate zσ − μ falls below 1/30 and floors there: a life of 30 years.
Regulatory lens
Basel: SRP31 (Principle 5) requires behavioural assumptions to be understood, conceptually sound, documented and tested; SRP31.41 lists the behavioural products and SRP31.47 asks for at least annual review. The standardised framework of BCBS 368 (April 2016; SRP31.94 to SRP31.129) caps core deposits (Table 2) and scales prepayment by 0.8 or 1.2 (Table 3); supervisors may mandate it, and should require a bank whose internal system is deficient to improve it and/or use the standardised framework (paragraphs 92 and 99). In the EU, the EBA's guidelines on IRRBB and CSRBB (EBA/GL/2022/14, 20 October 2022, applying from 30 June 2023) cap the behavioural repricing date of retail and non-financial wholesale deposits at a weighted average of five years per currency (paragraph 111) and keep deposits from financial customers out of behavioural modelling, except operational deposits (paragraph 110). In the UK, SS31/15 (January 2026 version, from 1 January 2027) sets the same five-year average (paragraph 2.9B(vi)) and asks for review at least annually and when markets change significantly (2.9L). Liquidity: the LCR prescribes its rates (Basel LCR40; the PRA Rulebook's Liquidity Coverage Ratio (CRR) Part); own assumptions go in the ILAAP (SS24/15, January 2026). Model governance: SS1/23's five principles, first effective 17 May 2024.
Spark Atoms
✦Spark AtomThe cap the EBA promised to watch+
When it published its IRRBB guidelines on 20 October 2022, the EBA said it would closely scrutinise one part of them in particular: the five-year cap on the repricing maturity of non-maturity deposits, to see whether it had unintended effects.
✦Spark AtomTen years of volumes+
The 2016 Basel standard asks banks to split stable from non-stable deposits using observed volume changes over the past ten years. RiskAtoms Bank's synthetic history runs twenty, and its fitted cores still sit above every cap.
Key takeaways
Contracts say when cash flows are due; behaviour says when they really come. Banks model prepayment, deposit rates and volumes, drawdowns and outflows.
In IRRBB the bank measures with its own documented, tested assumptions; the standardised framework caps core deposits and fixes the prepayment multipliers.
The caps decide: fitted cores of 78% to 97% are cut to 50% to 90%, and every segment hits its cap on core and maturity.
After the caps the average repricing date of the £40bn of deposits is 2.57 years, against 1.33 years booked and 17.3 years on the raw fit; the booked figures stand.
In liquidity the LCR prescribes its rates; the bank's own run-off and drawdown assumptions belong in the ILAAP.
Every assumption goes on a register with an owner, a source, a version and a backtest.
Reaction
Check what stuck
Question 1 · objective 3
A bank's model says 96% of its retail current accounts are core, with a behavioural life of 12 years. In the standardised IRRBB framework, what core share may it use? (Per cent.)
%
Question 2 · objective 3
£10bn of savings has a core share of 70% after the caps and a pass-through beta of 0.6. In the engine's split, how much reprices overnight? (£m.)
£m
Question 3 · objective 2
RiskAtoms Bank's own data says its savings would lose at most 4.8% in a month. Can it use that figure in place of the LCR's 10% run-off for less stable retail deposits?
Question 4 · objective 1
Which of these is a behavioural assumption rather than a contractual one?
Question 5 · objective 4
A new model raises the current accounts' fitted core from 93% to 97%, and the standardised IRRBB figures do not move. What is the most likely reason?
Glossary
Behavioural assumption
Behavioural assumption: how customers are expected to use the options in their products (to repay early, withdraw, draw down or stay), as opposed to what the contract says.
Contractual profile
Contractual profile: the cash flows and repricing dates written in the contract, before any behaviour.
Core deposits
Core deposits: the part of a non-maturity deposit balance that is stable and unlikely to reprice even when market rates move a lot.
Basel caps
Basel caps: the largest core share and average maturity of core deposits the standardised IRRBB framework allows, by category of deposit.
Behavioural life
Behavioural life: how long the core balance is expected to stay, from the rate at which it would run off in a stress.
Prescribed rate
Prescribed rate: a run-off or drawdown rate set by the rule itself, the same for every bank.
ILAAP
ILAAP (internal liquidity adequacy assessment process): the bank's own assessment of its liquidity and funding risks, with its own stress assumptions, reviewed by the supervisor.
Assumption register
Assumption register: the list of every behavioural assumption a bank's figures use, with its value, source, owner, approval and the result of its latest backtest.
Basel Committee on Banking Supervision, Interest rate risk in the banking book, BCBS 368, April 2016: Principle 5 and the standardised framework, paragraphs 99 to 132. Intermediate.
Prudential Regulation Authority, SS24/15 The PRA's approach to supervising liquidity and funding risks, January 2026: what the PRA expects of an ILAAP's assumptions. Advanced.
Origins
Basel Committee on Banking Supervision, Interest rate risk in the banking book, BCBS 368, April 2016.
Basel Committee on Banking Supervision, Basel Framework, SRP31 (version effective 1 January 2026).
European Banking Authority, Guidelines on IRRBB and CSRBB, EBA/GL/2022/14, 20 October 2022; and press release of 20 October 2022.
Prudential Regulation Authority, SS31/15 The ICAAP and the SREP, January 2026 (effective 1 January 2027).
Prudential Regulation Authority, SS24/15 The PRA's approach to supervising liquidity and funding risks, January 2026.
Prudential Regulation Authority, SS1/23 Model risk management principles for banks, May 2023.
RiskAtoms Risk Engine, ALM and liquidity features (deposit books, calibration, register), November 2026.
Watch this lesson
Video version
Prefer to watch? The same lesson as a 14-minute video will sit here, made from this page's script.