04 · A-IMA — expected shortfall models

Permission per desk, earned quarterly

The IMA's redesign is less about the risk measure than about accountability structure. ES at 97.5% calibrated to stress replaces VaR — pricing tail depth, scaled across liquidity horizons of 10–120 days so an exotic credit factor cannot pretend to index-future liquidity. But the deeper change: approval attaches to the desk, and stays only while the desk proves, every quarter, that the model's P&L is the desk's P&L (PLAT: Spearman correlation and KS-test between risk-theoretical and hypothetical P&L) and that exceptions stay countable (backtesting at 97.5% and 99%). Fail amber → capital surcharge; fail red → the desk falls back to A-SA. Risk factors themselves must earn modellability through real prices (RFET: 24 observations in 12 months, max gaps); the rest are charged as non-modellable via stress scenarios (SES) outside the diversification benefit. The 2008 problem — one bank-wide approval covering everything forever — is structurally gone.

• Sources: Art 325az–325bp; RTS on PLAT, RFET, NMRF scenarios, backtesting; the 3rd Delegated Act (Art 495v multiplier 2027–29). UK IMA from 2028 — cross-border desks live one year in two regimes.