FINCAD was featured in a recent Risk.net article, Structured Products Desks Join the AAD revolution. The article discusses how derivatives dealers are now electing to apply adjoint algorithmic differentiation (AAD) to complex pricing and analytics problems. FINCAD has been a pioneer in this area, having incorporated its own implementation of AAD into its F3 Platform, which it launched in 2010. Firms that utilize AAD typically experience a dramatic uptick in computational speeds when compared to traditional risk methods such as “bumping.”
One reason dealers have been drawn to the AAD model is that they are facing increasingly complex risk management challenges. They must find ways of addressing these challenges both quickly and accurately. But the need to calculate risk sensitivities, or Greeks, across a vast, and many times, diverse book of products, is placing major processing strain on issuers’ pricing and risk models.
As a potential solution, many dealers are turning to the mathematical technique, AAD, which is offering them a more efficient way of calculating Greeks. In the article, Adil Regahai, Head Quantitative Analyst at the French investment banking company, NATIXIS, shares his success with the model. He describes that prior to using AAD, a calculation would take 24 hours and involve 10,000 machines. But using AAD, the same calculation takes just two hours and uses 2,500 machines.
The article goes on to discuss how some firms are successfully utilizing FINCAD’s implementation of AAD, known as Universal Algorithmic Differentiation (UAD), for structured products. Clients highlighted include a major Taiwanese bank using UAD for hedging target redemption forwards, an Austrian bank using the model for managing equity basket exposures and a large European bank using it for CVA calculations on a portfolio covering 15 currencies.
As with any new approach, financial institutions may experience minor challenges when adopting AAD. For instance, implementing AAD typically requires firms to make significant code changes to their existing technology. When modifications are performed in-house, this undertaking can represent a major time investment for IT personnel who are generally already stretch thin. For this reason, many firms are turning to established providers such as FINCAD that have already done the coding work and can help them get AAD up and running quickly— without significant interruption to business as usual processes.
Another challenge is that while AAD is efficient at generating first-order Greeks (e.g. delta, vega and theta), it is less performant when computing second-order Greeks (e.g. gamma, vanna and charm). Russell Goyder, Director of Quantitative Research and Development at FINCAD, explains that this dilemma can be easily overcome. He advises firms to use the traditional bump method on subsets of the Greeks generated by their AAD.
The article explains that part of the increased interest in AAD may be fueled by firms’ anticipation of regulations soon to be put into place by the International Swaps and Derivatives Association (ISDA). Starting in September 2016, major derivatives firms will be called on to have a transparent and standardized risk-based margining approach. Implementing AAD may be a good way for firms to meet this mandate.
To view the original article as it appeared on Risk.net, click here.
For more information on FINCAD’s Universal Algorithmic Differentiation (UAD), watch our webinar, Improve Trading Performance with Algorithmic Differentiation.