The method will save significant computation time, energy and costs. Saving computation costs will also mean that companies can run more calculations to calculate the risks of their portfolios more precisely. They will be able to include more complex - but more precise - models into their libraries, such as rough volatility models. “The work also enables members of the public – with a little knowledge of Python, but without access to significant computing power – to download the publicly available code and double check option prices for themselves on their own PCs under a number of different models in real time.
04 February 2020
Lecturer in Financial Mathematics Dr Blanka Horvath wins Risk.net Rising Star Award
Dr Blanka Horvath, Lecturer in Financial Mathematics in the Department of Mathematics, has won the Risk.net Rising star in quantitative finance award with the paper ‘Deep Learning Volatility: A deep neural network perspective on pricing and calibration in (rough) volatility models’.
The Risk.net Rising star in quant finance award recognises new talent in quantitative finance. The award winning paper was co-authored with PhD students Aitor Muguruza and Mehdi Tomas and was also shortlisted for Research Paper of the Year in the RiskMinds Awards.
Dr Horvath’s research explores the properties of stochastic processes - mathematical objects usually defined as families of random variables – used to model the dynamics of financial markets.
Her recent paper using deep neural networks is the first to enable the pricing of rough volatility models– in real time with a precision and a computation that is standard in the industry for other model families. Rough volatility models have entered the landscape of mathematical finance recently in a series of pioneering articles (2007, 2011, 2014 and 2015) exploring their advantageous properties.
These models refine classical stochastic volatility models by allowing a variability in the regularity of the volatility process. More precisely, while classical volatility models have a Hölder regularity of H=0.5, these models allow for any regularity parameter H (0, 0.5]. This added flexibility quants to model market dynamics much more accurately than in classical models. This young movement has grown into an active field of research in academia and with this work we hope that it finds its way to industry applications as well.
The recent results and the calibration speedups effectively have the consequence that rough volatility models can now enter standard pricing libraries. Not only does this method speed up pricing and calibration option prices under rough volatility models, but under all previously used standard models too, to a factor of 30,000. Companies using this method could therefore gain a competitive advantage and reduce their risk exposures.
Speaking about the potential real-world impact of this breakthrough, Dr Horvath said:
The next phase of Dr Horvath’s research will be to research the properties of more data-driven models and their interplay with standard models such as rough volatility models, and the possibilities and challenges that arise from a close, data-driven modelling of market dynamics.