FRTB in the UK: the standardised-versus-internal-models choice

In January 2026, the Prudential Regulation Authority (PRA) published Policy Statement PS1/26, settling the final shape of the UK’s Basel 3.1 market risk rules.¹ PS1/26 also started the clock on a capital strategy decision: rebuild internal models approval for 2028, or commit to the standardised route for good. The rules rest on the Fundamental Review of the Trading Book (FRTB), under construction since 2012.² FRTB replaces a regime the Basel Committee itself judged broken on two counts.² A value-at-risk (VaR) metric understated losses in the tail precisely when it mattered most, and a boundary between the trading book and the banking book invited regulatory arbitrage.

Why the pre-crisis regime failed

Introduced in 1996, the Market Risk Amendment to the Basel Accord let banks with supervisory approval use internal models calibrated to 99th-percentile VaR over a 10-day holding period.³

For capital purposes, VaR has a structural flaw: it marks the threshold of a loss distribution at a given confidence level, but says nothing about how large losses beyond that threshold could be. A 99th-percentile VaR of £100 million is equally consistent with a 99.1th-percentile loss of £105 million or one of £500 million; the capital requirement is identical in both cases. In the 2007-08 financial crisis, the losses that crystallised in trading books came precisely from that tail, and the approved models had never been calibrated to capture it.

The second failure was the trading-book boundary. Positions in the trading book were charged market risk capital, and positions in the banking book credit risk capital. The arbitrage was systematic: firms booked each instrument wherever the capital charge was lower, not where its economic intent belonged.

Patching the old regime had already been tried: Basel 2.5, in July 2009, added stressed VaR and the incremental risk charge to it in direct response to the crisis losses. The Basel Committee published the first final FRTB standard in January 2016, and that one rewrote the framework end to end.⁴ A January 2019 revision recalibrated the FRTB standard after a quantitative impact study exposed problems in the default risk charge and the sensitivities-based method’s correlations.⁵,

The internal models approach replaced VaR with expected shortfall (ES), which averages the losses across the entire tail beyond the confidence threshold. The trading-banking book boundary was redrawn around trading intent, with a presumptive list of what sits where, rather than around capital optimisation. A new standardised approach was introduced, designed to stand on its own as a credible measure of risk rather than as a rough approximation of the internal models number.

And a stress-scenario regime was added for risk factors with too little market data to model at all. Each of those four changes shapes the choice UK firms now face.

What are the three components of the FRTB standardised approach?

The UK calls the FRTB standardised approach the advanced standardised approach (ASA); a separate simplified standardised approach (SSA) remains available to smaller firms. The ASA is not a simplified fallback but a full sensitivity-based calculation, and it applies to every firm in the UK framework from 1 January 2027. The ASA also applies to every desk at every bank that uses internal models, because it sets the benchmark for the output floor, a minimum on total capital set as a share of the standardised calculation.¹

The standard puts it in one line at MAR20.4: the requirement "is the simple sum of three components: the capital requirement under the sensitivities-based method, the default risk capital (DRC) requirement and the residual risk add-on (RRAO)".⁵

• Sensitivities-based method (SBM): three sensitivity measures aggregated across all positions, delta (linear sensitivity to the underlying), vega (sensitivity to implied volatility), and curvature (the non-linear residual for instruments with optionality). Each is computed across the seven risk classes d457 defines, then risk-weighted and aggregated under supervisory correlations. Because those correlations shift under stress, the calculation runs for each of three correlation scenarios, and the charge is the largest of the three.⁵
• Default risk capital (DRC): captures jump-to-default risk on instruments subject to credit risk. It is capitalised through a discrete jump-to-default model rather than the continuous approach the banking book uses for the same names.⁵
• Residual risk add-on (RRAO): covers risks the SBM and DRC do not fully capture: instruments with exotic underlyings such as longevity, weather, or future realised volatility, and payoffs that cannot be written as a finite combination of vanilla options. The charge is the gross notional of those instruments multiplied by a supervisory risk weight. MAR23.8 sets it at "1.0%" for instruments with an exotic underlying and "0.1%" for those bearing other residual risks.⁵,

How does desk-level IMA approval work, and what is the PLA test?

The internal models approach (IMA) replaces the old bank-wide VaR model with a desk-level permission architecture. The core change is methodological. IMA capital is based on expected shortfall at the 97.5th percentile, one-tailed, rather than VaR at the 99th percentile. Because ES at 97.5% integrates the full distribution of losses beyond that point, it responds to the shape of the tail, not just where the tail begins. Liquidity horizons are applied at the risk-factor level, from 10 days for the most liquid factors to 120 days for the least, and the ES result scales the base 10-day figure accordingly.⁵

Two tests then run continuously. Profit-and-loss attribution is the one firms talk about, but backtesting sits alongside it: d457 requires that each desk "must satisfy backtesting requirements on an ongoing basis to be eligible to use the IMA", and treats the two assessments as complementary.⁵ Both run quarterly.

Approval itself is desk-level in a literal sense. A bank first satisfies the PRA that its firm-wide risk management meets the qualitative standards. It then nominates each desk it wants to run on the IMA, and the PRA approves or rejects each nomination on that desk’s own record. Desks that are never put forward, or that later fail the ongoing tests, use the standardised approach instead.⁵

The ongoing eligibility test is the profit-and-loss attribution (PLA) test, which each desk must pass quarterly to keep its IMA permission. It compares two P&L series:

1. Risk-theoretical P&L (RTPL): the P&L the desk’s risk model predicts.
2. Hypothetical P&L (HPL): the P&L implied by actual end-of-day market data.

Two statistics then measure how closely the model tracks reality. The Spearman correlation asks whether the two series move together: when the model shows a gain, did the desk actually gain? The Kolmogorov-Smirnov (KS) distance asks whether the two series have the same distribution: are losses of the same size arriving at the same frequency? Both are computed over "the time series of the most recent 250 trading days", and MAR32.42 sets the zone boundaries: green needs the correlation metric above 0.80 and the KS metric below 0.09, while red follows if the correlation falls below 0.7 or the KS metric rises above 0.12.⁵,

The IMA also treats non-modellable risk factors (NMRFs) separately. These are factors that fail the risk factor eligibility test (RFET), which d457 defines by requiring "at least 24 real price observations per year" with no 90-day period in which fewer than four are identified.⁵ Under the old regime, thin-market factors were proxied or simply excluded. Under FRTB, each carries its own capital charge, calculated with a stressed version of expected shortfall (SES) and added on top of the ES charge for the factors the model can cover.⁵ The NMRF charge is a separate, mandatory capital layer for factors that cannot meet the data standard for full ES modelling, at least as the international standard stands today. For banks with material exposure to illiquid markets, that extra layer matters.

The UK timeline: 2027 for standards, 2028 for IMA

The UK’s Basel 3.1 implementation has a split effective date, confirmed in PS1/26, and our Basel 3.1 summary covers the wider package this sits inside.⁶ The advanced standardised approach, the trading-banking book boundary, and the output floor all take effect on 1 January 2027. The FRTB internal models approach is deferred to 1 January 2028.¹

During the interim year, from 1 January 2027 to 1 January 2028, firms keep their existing IMA permissions. Positions that fall out of scope under the new boundary rules, and positions the existing IMA framework has no methodology to capitalise, move to the ASA or the SSA. A firm may also move its whole trading book to the ASA rather than run a hybrid, and use the interim to build and test the new infrastructure. Our reading of the transition is that old permissions do not carry over: a firm that does not reapply and win approval under the new framework loses its IMA when the interim period ends.¹

The output floor requires firms, including those using internal models, to report risk-weighted assets of at least 72.5% of those the standardised approach would produce, once it is fully phased in.¹ The output floor is set on risk-weighted assets rather than on capital: at a Pillar 1 total capital requirement of 8%, that implies capital of about 5.8% of standardised-approach risk-weighted assets. The floor starts on 1 January 2027 at a lower level and steps up across a four-year transitional period. In effect that makes the ASA unavoidable for IMA firms, whatever their permissions: the floor is defined against the standardised number, so a firm cannot know where it sits without computing it.

How does the UK framework differ from the Basel FRTB standard?

Beyond the split timeline, PS1/26 makes several targeted adjustments to the international standard for UK conditions. None of these UK adjustments changes the choice between the advanced standardised approach and the internal models approach.

• Investment funds: a simpler treatment settles how funds are allocated and capitalised at the trading-book boundary. A fund belongs in the trading book where at least 90% of its net asset value is attributable to positions that would sit there.¹ The bank may treat those holdings as directly held where it can see inside at least 50% of the fund. Whatever it cannot see into carries a flat 70% risk weight.¹
• Residual risk: PS1/26 finalised an RRAO permissions regime letting a firm apply for an alternative calculation where the flat charge, 1.0% or 0.1% of gross notional, is disproportionate to the risk it actually runs, for instance on genuinely hedged positions.¹ It was consulted on in CP17/25.⁷
• Structural FX: a dedicated chapter covers the foreign exchange (FX) positions a bank holds not to trade but to keep its capital ratio steady when exchange rates move. With the PRA’s permission, those positions can be excluded from the market risk charge rather than capitalised as if they were trading exposures.¹

What should UK firms build before 2027, and should an IMA firm renew?

The 2027 effective date for the ASA and the boundary is just over six months away, and the practical priority is the same for every firm, whatever its IMA intentions: build the ASA. The ASA build is a step-change from today’s approach, which charges positions directly: delta, vega and curvature must now be computed as sensitivities to every applicable regulatory risk factor, then risk-weighted and aggregated within buckets and risk classes. Alongside it sit two judgements: where RRAO exposures fall in the current book, and, for smaller firms, whether the SSA simplification will do.

For firms with IMA permissions, the real decision comes in 2028: renewal is a choice, not a rollover. The question is whether the model still earns its keep. To win approval, a firm needs desk-level nominations, a year of backtesting and PLA results, an ES model with its NMRF capital layer in place, and quarterly PLA monitoring from then on.⁵ The reward for IMA approval is a capital number below the ASA’s, though the output floor caps how far below it can go. For banks with concentrated books and strong internal risk management, the trade can still pay. For banks whose IMA permission is a legacy of business they no longer trade, the ASA may give a more transparent and proportionate result.

For some firms the standardised approach will be a stepping-stone back to internal models; for others, a destination. Either way, the road starts in the same place: a 2027 deadline, an output floor that makes the ASA universal, and UK adjustments that tune the terms of the choice without changing it. For firms that go further, approval in 2028 and the PLA test thereafter will decide IMA eligibility one desk at a time.

Frequently asked questions