XVA calculation product of the year and Best use of machine learning/AI: IHS Markit

Stuart Nield, IHS Markit

To run a profitable derivatives business, banks need to understand all the costs involved in trading derivatives contracts. Recently, banks have been caught out by mispricing and writedowns of the trading book. To avoid this, banks use valuation adjustment (XVA) calculations to understand their true profit-and-loss and capital in derivatives trading. XVAs have grown in size and complexity over the past decade. The changing landscape, driven by new regulation, means traditional systems are becoming obsolete.

As well as quick calculations of XVAs and capital measures, IHS Markit’s XVA system allows banks to focus on more complex issues, such as the incremental effects of trading activity and market moves on the portfolio, and the portfolio’s sensitivities to market risk factors. These capabilities enable banks to better balance the increasingly complex regulatory requirements, while making informed decisions on the trading desk – pre-trade and in real time.

The computational expense of calculating XVAs and XVA sensitivities on traditional systems usually means they are computed just once a day. Ideally, banks would have the ability to rapidly compute them intraday to aid in real-time decision-making. IHS Markit’s system can calculate the standardised approach to credit valuation adjustment, for example, on a real-time basis, giving banks the ability to make trading decisions with an understanding of return on capital.

Neural networks and machine learning have revolutionised a range of computationally intensive modelling tasks. IHS Markit has applied these techniques to XVA calculation to address some of the toughest modelling and performance challenges in financial markets. IHS Markit’s system accelerates XVA computations and delivers XVA valuations and sensitivities to meet the challenges of modern financial risk pricing. The high-speed XVA pricer, based on neural net modelling and accelerated graphics processing unit computing, delivers accurate XVA valuations much faster than traditional methods. This allows for live calculation of XVAs for portfolios of swaps and cross-currency swaps.

IHS Markit has shown that applying deep neural networks (DNNs) to replicate instrument pricing functions offers high-performing alternatives to traditional pricing models. The trained DNNs return full ranges of sensitivities with negligible additional computation expense than that required to return the present-value. This enables direct computation of extremely data-intensive calculations, resulting in much-needed improvement in accuracy and speed for calculating measures such as margin valuation adjustment (MVA).
 

Judges said:

• “Strong enterprise risk solution.”
• “Pragmatic use of machine learning techniques to XVAs.”
• “Addresses the growing need to accelerate XVA computation.”
• “XVA calculation as a high-dimensional problem provides good application for DNN techniques. IHS Markit is one of the first to tackle the corresponding challenges.”
• “State-of-the-art product with a strongly quantitative approach.”
   

Stuart Nield, Global Head of Product, Financial Risk Analytics at IHS Markit, says:

“We are delighted to win these two awards, recognising our dedication to continually improving our XVA product. We are proud to have helped our clients navigate their challenges last year, such as the transition to risk-free rates, improved performance and the new regulatory focus on credit valuation adjustment in Asia. We believe our capabilities make us unique in the market. The best use of machine learning award is a testament to the excellent work of our research, financial engineering and software engineering teams to push the boundaries of what is possible in an XVA engine.” 

Sponsored content