Building An AI-Based Risk Computation Engine

Abstract

This paper discusses an approach for using AI-based systems (deep neural networks) in a very specific area of risk management, namely credit risk (counter-party risk), through efficient margin and limit management using an intelligent (non-formulaic) value at risk (VaR) computation engine.


This is opposed to traditional IT applications that employ Monte Carlo simulations with simplifying assumptions. The use of AI is primarily to uncover hidden patterns and signals in the daily price movements of the assets and/or portfolio of assets placed as collaterals to compute the VaR.  Opportunities for automating the processes associated with margin call management, asset liquidation activity (to limit risk), limit management, netting and collaterals management, and capital requirement computations have also been touched upon under the lens of an enterprise-scale AI.