The Fundamental Review of the Trading Book (FRTB), drawn up by the Basel Committee on Banking Supervision, imposes a series of requirements for market risk capital. One of the main innovations defined by FRTB is the concept of Non-Modellable Risk Factors (NMRF), which require holding additional capital. Financial institutions seek to reduce this capital charge in order to increase their profitability.
This article presents a method to manage Non-Modellable Risk Factors using machine learning, and compares the results obtained with the standard market method.
The challenge of non-modellable risk factors
A risk factor becomes non-modellable when there is not enough observable market data to model it reliably. Under FRTB, these factors attract punitive capital add-ons. The core difficulty is therefore one of scarce data: how to reconstruct robust risk behaviour when observations are limited or illiquid.
A machine-learning approach
Machine learning offers a way to recreate reliable risk indicators despite data scarcity. The study explores how learned models can estimate the behaviour of non-modellable factors and benchmarks their accuracy against the traditional market-based approach, pointing to a path for reducing capital requirements without compromising risk coverage.