What makes a risk factor non-modellable under FRTB, and how are NMRFs capitalized separately?

Under FRTB's Internal Models Approach, risk factors that fail the modellability assessment must be capitalized separately through a stress scenario capital add-on (SES) rather than being included in the expected shortfall model. This prevents banks from relying on sparse or stale data for their internal models.\n\nModellability Criteria:\n\nA risk factor is deemed modellable only if it has:\n1. At least 24 real price observations per year, AND\n2. No gap exceeding 1 month between consecutive observations, AND\n3. The observations must represent actual transactions or committed quotes (not indicative prices)\n\nIf any criterion fails, the risk factor is classified as non-modellable (NMRF).\n\nCommon NMRFs:\n- Bespoke correlation parameters for structured products\n- Long-dated implied volatility surfaces for illiquid underlyings\n- Basis spreads for off-the-run bonds\n- Credit spreads for unrated or infrequently traded issuers\n- Dividend forecasts beyond 2-3 years\n\nCapital Treatment:\n\nEach NMRF receives a stressed scenario capital charge (SES_i) calculated as:\n\nSES_i = max(Loss from stress scenario at 97.5% confidence, 0)\n\nThe stress scenario should reflect a period of significant financial stress relevant to that specific risk factor. The bank must:\n- Identify a plausible extreme movement for the risk factor\n- Apply that shock to the portfolio\n- Calculate the resulting P&L impact\n\nAggregation of NMRFs:\n\nIndividual NMRF charges are aggregated assuming zero correlation between them (conservative) or partial correlation if the bank can demonstrate dependence:\n\nSES_total = sqrt(sum of (SES_i)^2) (zero-correlation case)\n\nOr with limited diversification:\n\nSES_total = rho x sum(SES_i) + (1-rho) x sqrt(sum(SES_i^2))\n\nwhere rho is a regulatory correlation parameter (currently proposed at 0.5-0.6).\n\nWorked Example:\nLongshore Securities has three NMRFs in their trading book:\n\n| NMRF | Description | SES_i |\n|---|---|---|\n| 1 | 15-year dividend yield on Cerulean Tech | $2.8M |\n| 2 | CDS spread for Blackmoor Industries (unrated) | $1.5M |\n| 3 | Basis correlation for bespoke CDO tranche | $4.1M |\n\nWith rho = 0.5:\n\nSES_total = 0.5 x ($2.8M + $1.5M + $4.1M) + 0.5 x sqrt($2.8M^2 + $1.5M^2 + $4.1M^2)\n= 0.5 x $8.4M + 0.5 x sqrt($7.84 + $2.25 + $16.81)M\n= $4.2M + 0.5 x sqrt($26.90M^2)\n= $4.2M + 0.5 x $5.19M\n= $4.2M + $2.59M = $6.79M\n\nThis NMRF add-on is layered on top of the IMES capital charge, creating a strong incentive for banks to source real price data and move risk factors into the modellable category.\n\nStudy FRTB capital frameworks in our FRM Part II materials.