FRTB Non-Modellable Risk Factors: AI-Supported NMRF Identification and Basel III Capital Calculation Optimization

A non-modellable risk factor (NMRF) is a trading book risk factor that fails the Real Price Observation test (RPO test). Specifically, this means fewer than 24 real price observations exist within a twelve-month period, or more than one month elapses between two consecutive observations. Risk factors that fail this test are classified as non-modellable and require a separate capital charge via the Stressed Expected Shortfall (SES). Typical NMRF include illiquid credit spreads, exotic volatilities, or correlations for which insufficient market data is available.

The RPO test checks whether sufficient real price observations exist for each risk factor. The requirement is: at least 24 real prices within the past twelve months, with no gap between two consecutive observations exceeding one month. Real prices include actual transactions, firm quotes, and committed quotes. Indicative prices or model valuations do not count. The bank must document the price observations and be able to demonstrate them to the supervisor. If a risk factor passes the RPO test, it is deemed modellable and enters the regular Expected Shortfall calculation.

Stressed Expected Shortfall (SES) is the capital metric for non-modellable risk factors. A separate SES is calculated for each NMRF, capturing the loss potential under stress conditions. The bank must define an appropriate stress scenario for each risk factor, based on historical extreme events or regulatory prescribed scenarios. The individual SES values are then aggregated – typically assuming limited diversification, which leads to significantly higher capital requirements than the regular ES calculation. The SES capital add-on is added to total IMA capital.

The 24 observations threshold is the critical cutoff for risk factor modellability. Missing it means the bank must treat the risk factor as NMRF and compute a separate SES capital charge. In practice, the NMRF capital add-on often accounts for 30–60% of total IMA capital requirements. Every risk factor just below the threshold is therefore a direct lever for capital optimization. Banks actively invest in additional data sources and vendor partnerships to push risk factors above the 24 observations threshold and reduce the capital add-on.

Banks use several levers to lower the NMRF capital charge. First, data enrichment through additional market data sources, vendor connections, or consortium data platforms to obtain more real price observations for critical risk factors. Second, proxy mapping, where a risk factor is assigned to a sufficiently similar modellable factor. Third, risk factor aggregation, combining granular factors into broader, better-observable categories. Fourth, SES calibration optimization, where stress scenarios are calibrated using data to avoid unnecessarily conservative assumptions. ADVISORI supports prioritization of these measures based on capital impact.

The Standardised Approach (SA) has no explicit NMRF treatment – capital requirements are calculated via sensitivity-based risk weights, regardless of data availability. Under the Internal Models Approach (IMA), banks must test each risk factor individually for modellability. Non-modellable factors receive the SES capital add-on, while modellable ones enter the regular ES calculation. This makes IMA more data-hungry, but with good data quality it offers lower capital requirements than SA. The choice between IMA and SA depends significantly on the NMRF proportion in the trading portfolio.

Supervisory authorities expect comprehensive documentation of NMRF treatment as part of IMA approval. This includes: an up-to-date risk factor inventory with modellability classification, the methodology and results of the RPO test for each risk factor, calibration of stress scenarios for NMRF including the historical data and assumptions used, the aggregation methodology for SES values, and processes for ongoing monitoring and reassessment of modellability. Authorities also require evidence of data quality and the provenance of price observations used.