Non-Modellable Risk Factors

FRTB, the post-crisis overhaul of bank trading-book capital rules, draws a hard line based on data availability rather than modeling sophistication: a risk factor only counts as "modellable" for internal-model capital purposes if the bank can point to at least 24 observable, real transaction prices in a rolling 12-month window, with no gap longer than a month between them. A risk factor can be economically well-understood and priced by a perfectly reasonable model, but if the desk can't produce that minimum trail of real trades, the regulation classifies it as a Non-Modellable Risk Factor (NMRF) anyway — the test is about evidence of trading, not model quality.

The consequence of landing in the NMRF bucket is a separate, generally much larger capital charge computed with a stressed scenario approach rather than the bank's normal internal value-at-risk model, and NMRF capital is calibrated to be punitive by design, to give banks a real incentive to source more market data rather than quietly running risk on thinly-traded factors under a favorable model. This hits exactly the risk factors you'd expect: exotic or long-dated derivatives, niche credit spreads, and any market segment where liquidity is naturally too low to generate two dozen observed trades a year.

For a desk, the practical implication is that pricing model quality and regulatory capital treatment are two separate questions — a well-calibrated model for an illiquid risk factor doesn't rescue it from the NMRF charge, and the only way out is to either source enough independent, verifiable transaction evidence or accept the higher capital cost as the price of trading that risk.