Fusion Confess

Inspiration

As the risk manager or the risk operation officer I need to reconcile and validate Residual Risk Add-on (RRAO) charges imposed by Fundament Review of the Trading Book (FRTB) regulation. The charges are computed on trades with non-linear risk exposure in cross-asset portfolio. The trades that are subject to RRAO are manually identified and marked by the traders. I need to identify the exceptions and classify them to avoid both over and under charge.

What it does

To comply with minimum capital risk requirements, the traders must identify the trades with non-linear risk component to be submitted to RRAO charges. The current way to select the instruments is subjective to trader understanding of regulations Risk Manager faced with the challenge to validate RRAO charges as the only data available is market risk sensitivity and p&l vectors uploaded from Front Office. As sensitivity and P&L data are highly correlated, the combination of un-supervised novelty detection and supervised classification ML algorithms allow identify the subset of instruments for review and classified them. ML algorithms are leveraging the data already computed in FO for end of the day process and uploaded to Enterprise Risk Management System. The results of ML processing is published to UXP dashboards to Risk Officer to review The feather analysis of the deal exception is done by query back to Front Office core system to retrieve additional details about the trade After completing the review Risk Officer accepts or corrects the RRAO charges