Internal Models vs the Standardised Approach
Banks can calculate regulatory capital using their own approved risk models or a fixed regulatory formula, and the choice changes both the capital charge and the incentives it creates.
Prerequisites: Risk-Weighted Assets
Two banks hold identical trading books. One calculates its capital requirement using a regulator-approved in-house risk model, tailored to its own portfolio and correlations. The other uses a fixed formula that regulators apply to every bank, regardless of the specifics of its book. Both numbers are meant to measure "how much loss could this book plausibly produce," but they can come out very differently — and which method a bank is allowed to use is itself a regulatory decision.
The standardised approach is a common formula every bank can use; the internal models approach lets a bank use its own risk model instead, but only after regulators approve it and keep checking it works.
Two philosophies
The standardised approach assigns fixed risk weights and formulas to categories of exposure — this type of bond gets this weight, this kind of option gets that add-on — using rules set centrally by regulators (as under Basel's FRTB or the counterparty-credit-risk standardised approach). It is simple, comparable across banks, and doesn't require the regulator to trust any individual bank's modeling.
The internal models approach (IMA) lets a bank calculate its own capital requirement using an in-house model — often a form of value-at-risk or expected shortfall built on the bank's actual historical data and actual portfolio correlations. It can produce a lower capital charge for a well-diversified, well-hedged book, because it captures diversification benefits the standardised formula ignores. But it requires the bank to earn and keep regulatory approval, backed by ongoing model validation and backtesting.
Worked example
A bank's trading desk holds offsetting positions in correlated interest-rate products across several currencies. The standardised approach, which weights each currency bucket largely on its own, produces a capital charge of $800 million. The bank's internal VaR-based model, which captures the real hedging benefit across currencies, produces a charge of $550 million for the same book. The bank has a strong incentive to seek IMA approval — but it must first demonstrate the model passes rigorous backtesting, or regulators will require it to fall back to the standardised number.
What this means in practice
Regulators haven't let internal models run unchecked: after the 2008 crisis and repeated evidence that bank-reported risk-weighted assets varied wildly for similar portfolios, frameworks like FRTB tightened IMA approval standards and, for some risk categories, mandated the standardised approach as a floor beneath whatever the internal model produces. A bank losing IMA approval for a desk — because backtesting exceptions piled up — can see that desk's capital charge jump overnight back to the standardised number.
A lower capital number under an internal model isn't automatically a "truer" one. It reflects the model's assumptions about correlation and tail behavior, which can be wrong in ways that only show up in a crisis — exactly when the extra capital would have mattered most.
Further reading
- Basel Committee, 'Minimum Capital Requirements for Market Risk' (FRTB)