The internal model method and SFTs · RiskAtoms Academy

By the end you will be able to

1Compute IMM exposure from a simulation: effective EE, effective EPE, α and the stressed calibration.
2Compare IMM with SA-CCR on the same netting sets, and say why a bank might seek IMM approval.
3Compute E* for repos and reverse repos under the comprehensive approach, alone and netted under a master netting agreement.
4Apply the currency mismatch haircut and the Basel SFT haircut floors.

Why it matters

SA-CCR puts RiskAtoms Bank plc's derivatives exposure at £36.90m. The bank's own Level 3 simulation of the same trades, run through the internal model rules, gives £15.69m even on a stressed calibration: 43% of it. Meanwhile £3bn of reverse repos and £1.5bn of repo funding sit outside SA-CCR altogether. Module 4.1 counted the reverse repos and left the repo out. This lesson shows what a model would change, and what the repo adds.

Core explanation

Models for derivatives, haircuts for repos

Illustrative dataRiskAtoms Bank plc's capital book

The same made-up UK bank as Level 3: £67bn of assets and £4bn of equity at 31 December 2025, in £ millions. Module 4.1 adds its regulatory layer: own funds, deductions, risk-weighted assets by risk type, buffers, the leverage exposure and MREL, computed from the Level 3 credit, trading, derivatives and operational risk books. Every figure is illustrative, not any bank's data.

Pro forma. The book applies Basel 3.1, the 2017 finalisation of Basel III, to the 31 December 2025 balance sheet. On that date the bank still reported under the old UK CRR. The UK applies Basel 3.1 from 1 January 2027, and the internal model approach for market risk from 1 January 2028 (PRA PS1/26, paras 1.23 and 1.24). So the book shows the bank as it would look under the new rules.

The counterparty data. The four OTC derivative netting sets of the Level 3 derivatives book (M3.4: Counterparty credit risk): Halvergate Moor Manufacturing (HAL, a manufacturer), Corriemoor Bank (COR), Saltney Capital (SAL, a hedge fund) and Ebbsworth Pension Trustees (EBB), all invented. Module 4.1 gave them SA-CCR exposure of £36.9m and counterparty credit risk RWA of £31.2m. That figure is SA-CCR on these four netting sets only, at Basel's α of 1.4.

What it leaves out. The trades the bank clears through central counterparties (£0.42m of RWA, L4.4.04: Exposures to central counterparties) and the counterparty risk on its repo funding (£12.7m of RWA if each repo stood alone, nothing once netted with the reverse repos, L4.4.03: The internal model method and SFTs).

Illustrative. The internal model method needs the supervisor's approval (CRE53.1). RiskAtoms Bank plc has none, so its counterparty RWA are SA-CCR. Every IMM figure on this page shows what the method would give on the Level 3 exposure simulation, not what the bank reports.

The internal model method

L3.4.02: Exposure measures simulated each netting set's exposure: thousands of paths for rates, FX and the equity index, the trades revalued on each, the exposure averaged across paths. That average at each date is the expected exposure, EE, an expectation in the sense of M-C02: Expectation. The internal model method lets a bank use such a model for capital, with the supervisor's approval (CRE53.1), for derivatives, SFTs or both (CRE53.2). Without approval it uses SA-CCR for derivatives (CRE53.3).

The model output goes through three rules before it becomes an EAD.

Effective EE never falls (CRE53.12). Halvergate's EE drops from £3.19m to £1.94m at one year, when its FX forward matures; its effective EE stays at £3.19m. The rules assume short trades would be replaced.
Effective EPE averages effective EE over the first year (CRE53.13), and EAD = α × effective EPE with α = 1.4 (CRE53.11, 53.14). A bank may estimate its own α with approval, but not below 1.2 (CRE53.16): a floor in the sense of M-A11: Min, max, floors and caps.
The model must also be run on a stressed calibration, and the bank uses whichever gives the higher capital requirement for the portfolio as a whole, not counterparty by counterparty (CRE53.7). Margined sets must use an MPoR of at least ten business days, five for repo-style trades, twenty for large or illiquid sets (CRE53.24).

Illustrative. £m. m3-4's 2,000-path simulation; stress: every factor move scaled by 1.5
Netting set	EEPE	IMM EAD	Stressed EEPE	Stressed EAD	SA-CCR EAD
Halvergate Moor (HAL)	2.146	3.004	3.258	4.561	6.934
Corriemoor Bank (COR)	1.377	1.928	2.059	2.882	5.689
Saltney Capital (SAL)	0.643	0.900	1.508	2.111	7.753
Ebbsworth Pension (EBB)	3.684	5.158	4.385	6.139	16.526
Total		10.989		15.693	36.902

The stressed calibration binds, so IMM exposure would be £15.69m illustrative, against £36.90m under SA-CCR: RWA of about £12.5m instead of £31.2m at the same risk weights. SA-CCR is built to be conservative. For Halvergate, α would have to be 3.23 for the model to match it. This is why big derivatives dealers seek approval, and why the output floor of L4.2.09: The output floor makes them compute SA-CCR anyway.

Repos and reverse repos

A securities financing transaction is a loan of cash against securities, or of securities against cash. SA-CCR does not cover them. A bank without IMM approval uses the comprehensive approach of L4.2.08: Credit risk mitigation: E* = max(0, E(1 + H_e) − C(1 − H_c − H_fx)), with ten-day haircuts scaled to the five-day holding period for repo-style trades (CRE22.57).

Reverse repos. The bank lends £3,000m of cash to banks against £3,060m of 1–3 year gilts. The 2% haircut is 1.41% at five days, so the gilts count as £3,016.7m, more than the cash: E* = 0. That is Module 4.1's figure.
Repo funding. In a repo the roles swap: the bank hands over £1,500m of gilts and receives £1,500m of cash. Its exposure is the gilts, and the haircut goes on them. Taking them as 5–10 year gilts (4%, so 2.83% at five days), E* = £42.43m and RWA £12.73m at the 30% bank weight. Module 4.1 left this out.
Netting. Under a master netting agreement that is enforceable on default, all repo-style trades with one counterparty form one netting set (CRE22.62), with the formula below (CRE22.65). The bank's repos are with the same four banks as its reverse repos, and the surplus gilts on each reverse repo cover the haircut on the repo. Netted, E* is zero for every counterparty.

Sort itIs it in scope of the Basel SFT haircut floors?

CRE56 sets minimum haircuts for some SFTs. Sort each trade.

A non-centrally cleared reverse repo: cash lent to a hedge fund against BBB corporate bonds
The same trade with Corriemoor Bank
Cash lent to a hedge fund against gilts
A repo with the Bank of England
A collateral upgrade: the bank lends gilts to a hedge fund against equities, and the fund may re-use the gilts

The formal version

IMM (CRE53.7, 53.11-53.16; the effective EE recursion is an image on bis.org, written here as in L3.4.02: Exposure measures):

\begin{aligned} \text{EffectiveEE}_{t_k} &= \max\big(\text{EffectiveEE}_{t_{k-1}},\ \text{EE}_{t_k}\big) \\ \text{EffectiveEPE} &= \sum_{k=1}^{\min(1\text{ year},\ \text{maturity})} \text{EffectiveEE}_{t_k} \times \Delta t_k \\ \text{EAD} &= \alpha \times \text{EffectiveEPE}^{\text{current or stressed}}, \qquad \alpha = 1.4,\ \alpha_{\text{own}} \ge 1.2 \end{aligned}

Current or stressed: one calibration for every netting set, whichever gives the higher capital requirement for the portfolio as a whole (CRE53.7).

SFTs under a master netting agreement (CRE22.65; the formula is an image on bis.org and was read as text in Article 220(3) of the Jersey FSC's CRM (CRR) Part, a Jersey adaptation of the UK CRR text, which writes ΣE − ΣC as the sum of net positions):

\begin{aligned} E^* &= \max\Big\{0,\ \sum_i E_i - \sum_j C_j + 0.4 \times \Big|\sum_s E_s H_s\Big| + 0.6 \times \frac{\sum_s E_s |H_s|}{\sqrt{N}} + \sum_{\text{fx}} |E_{\text{fx}}|\,H_{\text{fx}}\Big\} \\ H &= H_{10}\sqrt{T_M/10}, \qquad T_M = 5 \text{ days for repo-style trades} \end{aligned}

$E_i$ : cash and securities lent or posted; $C_j$ : cash and securities borrowed or held. $E_s$ : the net amount of each security issue, taken as positive; $H_s$ : its haircut, positive if the bank is a net lender of the issue and negative if a net borrower.
$N$ : the number of issues, leaving out any whose $E_s$ is under a tenth of the largest. $E_{\text{fx}}$ : the net position in each currency other than the settlement currency; $H_{\text{fx}}$ : 8% at ten days (CRE22.52).
CRE56 floors (non-centrally cleared, in scope): corporate debt 0.5% up to a year, 1.5% for 1–5 years, 3% for 5–10 and 4% beyond; securitised products 1% to 7%; main index equities 6%; other assets 10% (CRE56.6). For one trade lending cash, the haircut is H = C ÷ E − 1 (CRE56.9).

Try it

First the model. Pick a netting set and switch between the current and stressed calibrations: watch effective EE lift off EE when Halvergate's forward matures, and compare α × EEPE with SA-CCR.

Netting set

Calibration

Halvergate Moor (HAL). Illustrative: the bank has no IMM permission.

EEeffective EEshaded: the first year, which effective EPE averages (£m)

Effective EPE£2.146m

IMM EAD = 1.4 × EEPE£3.004m

SA-CCR EAD£6.934m

IMM ÷ SA-CCR43%

Stressed ÷ current EEPE1.52

Then the repos. Start on Corriemoor. Drop the gilts received to 100% of the cash and the netted E* turns positive; take the reverse repo to zero and only the repo's haircut is left.

Try itNet a counterparty's repos

Counterparty (resets to the book)

Corriemoor Bank AG, rated A, one master netting agreement.

Reverse repo: cash lent (£m)£1,000m

Gilts received, % of cash lent102.0%

Repo: gilts lent against cash (£m)£500m

Haircuts at 5 days: gilts received (1–3 years) 1.41%, gilts lent (5–10 years) 2.83%.

Each trade alone

£14.14m

Netted (CRE22.65)

£0.00m

Reverse repo alone, E*£0.00m

Repo alone, E*£14.14m

ΣE − ΣC−£20.00m

0.4 × net exposure£0.113m

0.6 × gross ÷ √N (N = 2)£12.120m

Netted E*£0.00m

RWA at 30%: alone · netted£4.24m · £0.00m

Netting removes it all: the surplus gilts on the reverse repo cover the haircut on the gilts lent.

Worked example

Corriemoor's repos, the FX haircut and the floor

£ millions. Haircuts at the five-day holding period. The FX and floor trades are illustrations, not in the book.

Three SFT cases

Step 1 of 4

1
Corriemoor, trade by trade
$\begin{aligned} \text{Reverse repo: } E^* &= \max\big(0,\ 1{,}000 - 1{,}020 \times (1 - 1.41\%)\big) = 0.00 \\ \text{Repo: } E^* &= \max\big(0,\ 500 \times (1 + 2.83\%) - 500\big) = 14.14 \end{aligned}$

Regulatory lens

Basel: CRE53 (IMM), CRE22 (the comprehensive approach for SFTs) and CRE56 (SFT haircut floors), all in the versions effective 1 January 2023. Jurisdictions as at 3 November 2026 as at 3 Nov 2026.

In the UK

Basel 3.1 applies from 1 January 2027 (PS1/26, para 1.23; L4.1.08: Basel 3.1 from 2027: UK, EU and US). The α of 1 for non-financial and pension counterparties of L4.4.01: SA-CCR: replacement cost and the multiplier sits in SA-CCR's exposure value article (Counterparty Credit Risk (CRR) Part, Art 274(2)); it does not change the α that multiplies effective EPE for a bank on the internal model method.

In the EU

CRR Article 284 sets IMM exposure at α × effective EPE, with α = 1.4, own estimates not below 1.2, and the higher of the current and the stressed calibration for all counterparties on the method (Art 284(3), (4) and (9)). The SFT netting formula is Article 220. Whether and when the EU and the UK apply the Basel SFT haircut floors of CRE56 is not covered here: this page shows them as Basel's rule.

Spark Atoms

Spark AtomThe model that needs a stress

A model fitted only to recent data shows low exposure after a calm spell, just when risk may be building. So the rules require a second run on three years of data that include a period of stress to the counterparties' credit spreads, and the higher of the two for the whole portfolio (CRE53.7, 53.51). On this book the stress raises IMM exposure from £10.99m to £15.69m, with a simple 1.5 times scaling of every move.

Spark AtomYour repo is someone else's reverse repo

For the bank's gilts to come back, the counterparty must return them. If it fails, the bank keeps the £1,500m of cash but must replace gilts that may have risen in price. That is why the haircut on a repo goes on the bank's own securities, and why handing over more gilts than the cash it receives would make the exposure larger still.

Key takeaways

IMM: EAD = α × effective EPE, with effective EE non-decreasing and averaged over the first year; α = 1.4, own estimate floored at 1.2; the higher of current and stressed for the whole portfolio (CRE53).
Illustrative: on the bank's four netting sets IMM gives £15.69m (stressed binds) against SA-CCR's £36.90m. The bank has no approval, so it reports SA-CCR.
SFTs use the comprehensive approach: haircuts scaled to five days for repos. Reverse repos: E* = 0. The £1,500m repo funding alone: E* £42.43m, RWA £12.73m, which Module 4.1 leaves out.
Under master netting agreements, CRE22.65 nets SFTs by counterparty with a 0.4 systematic and 0.6 / √N idiosyncratic haircut term. Netted with the reverse repos, the repo adds nothing.
A currency mismatch adds 8% (5.66% at five days). Basel's SFT haircut floors turn an in-scope trade below the floor into an unsecured loan: £50m to Saltney at a 1% haircut counts in full.

Reaction

Check what stuck

Question 1 · objective 1

A netting set has effective EPE of £5m on the current calibration and £4m on the stressed one, and the bank's portfolio as a whole is also higher on current data. What is its IMM EAD at the standard α, in £m?

£m

Question 2 · objective 1

Halvergate's EE drops at one year, when its FX forward matures. What happens to effective EE?

AIt stays at its earlier peakBIt drops with EECIt jumps to the SA-CCR exposureDIt is reset to zero

Question 3 · objective 3

The ten-day haircut on 5–10 year gilts is 4%. What is it at the five-day holding period for repos, in %? (Two decimals.)

%

Question 4 · objective 3

The bank repos £200m of those gilts for £200m of cash, alone with no netting. What is E*, in £m?

£m

Question 5 · objective 4

Which of these SFTs is in scope of the Basel SFT haircut floors (CRE56)?

AA non-centrally cleared reverse repo lending cash to a hedge fund against corporate bondsBA reverse repo lending cash to a bank against corporate bondsCA reverse repo lending cash to a hedge fund against giltsDA repo with the Bank of England

Glossary

Internal model method (IMM): Internal model method (IMM): a bank's own simulation of future exposure, used for counterparty credit risk capital with the supervisor's approval, for OTC derivatives, SFTs or both (CRE53.1-53.2).
Effective EE: Effective EE: expected exposure made non-decreasing over time: at each date, the larger of that date's EE and the previous effective EE (CRE53.12).
Effective EPE: Effective EPE: the average of effective EE over the first year, or until all trades mature if sooner (CRE53.13).
Stressed calibration: Stressed calibration: model parameters estimated on three years of data including a period of stress to the counterparties' credit spreads; the bank holds the higher of the current and stressed capital requirement for its whole portfolio (CRE53.7, 53.51).
Securities financing transaction (SFT): Securities financing transaction (SFT): a repo, reverse repo, securities lending or borrowing, or margin loan, where the value of the trade depends on market prices and it is often margined.
Repo: Repo: the bank sells securities and agrees to buy them back, in effect borrowing cash against them. Its exposure is the haircut on the securities it handed over.
E*: E*: the exposure after collateral under the comprehensive approach: exposure plus its haircut, less collateral after its haircuts (CRE22.46).
Master netting agreement: Master netting agreement: a legally enforceable agreement that lets the bank net all its repo-style trades with one counterparty on default (CRE22.62).
SFT haircut floor: SFT haircut floor: the minimum haircut on non-centrally cleared SFTs that lend cash against non-government collateral to unregulated counterparties; below it the SFT counts as unsecured (CRE56.6-56.7).

Deep Dives & Origins

Deep Dives

M6.5: Counterparty risk and XVA: exposure simulation in depth, from the paths to effective EPE and XVA. Advanced.

Origins

Basel Committee on Banking Supervision, CRE53: Internal models method for counterparty credit risk, paras 53.1-53.16, 53.24 and 53.51, version effective 1 January 2023 (Swiss SIF copy).
Basel Committee on Banking Supervision, CRE22: Standardised approach: credit risk mitigation, paras 22.46-22.65, bis.org.
Basel Committee on Banking Supervision, CRE56: Minimum haircut floors for securities financing transactions, paras 56.1-56.9, version effective 1 January 2023 (Swiss SIF copy).
Regulation (EU) No 575/2013 (CRR), Articles 220 and 284, as amended by Regulation (EU) 2024/1623.
Jersey Financial Services Commission, Credit Risk Mitigation (CRR) Part, May 2026, Articles 220, 223 and 224 (a Jersey adaptation of the UK CRR text).
Prudential Regulation Authority, PS1/26, 20 January 2026, para 1.23; PRA Rulebook: Counterparty Credit Risk (CRR) Part, Article 274, version from 1 January 2027.

Watch this lesson

Video version

Prefer to watch? The same lesson as a 12-minute video will sit here, made from this page's script.