06 · Operational risk

The model that measured everything except the future

Basel II's AMA let banks model operational risk with loss-distribution approaches, scenarios and correlation assumptions. The outcome was the credit-risk story in caricature: dispersion without credibility. Comparable banks held wildly different op-risk capital; the great post-crisis losses — LIBOR, FX-fixing, mis-selling, sanctions, US mortgage settlements, tens of billions each — arrived as litigation years after the conduct, in amounts no loss-distribution tail had contemplated. Modelling was measuring the past's noise while the actual risk was a legal process.

The SMA answer is deliberate crudeness: a single formula on income-statement scale (the BI), with coefficients rising in buckets. Basel's design lets a bank's own loss history modulate the result through the ILM; the EU set ILM = 1 — comparability over risk-sensitivity, the same trade as the output floor, made without apology. Loss history still matters twice: the ≥ €750m-BI bank must collect and disclose it, and Pillar 2 prices what Pillar 1 now refuses to model.

Marginal coefficients by BI bucket — crossing a threshold reprices only the increment, never the whole BI.

The lineage