Sorry, you need to enable JavaScript to visit this website.
Skip directly to content

Join Dr. Jean-Marc Mercier to explore a range of algorithms for generation and prediction, underpinned by kernel techniques.

13:00 EDT / 18:00 BST / 19:00 CEST

Event Abstract:

We review some generative and predictive algorithms based on kernels (RKHS theory), and apply them to concrete Finance applications. We are using kernel methods because they are considered as very efficient and performing for small to medium dataset sizes, that is usually the case for finance.

For time series prediction, conditioned generative methods allowed us to extend and reinterpret most of the existing quantitative models, escaping the gaussian world. These extended models perfectly match historical observations and can be conditioned by other risk sources, such as liquidity or exogenous market information, for stress test or portfolio management purposes. An example coming from crypto trading is considered. Another interesting application of these algorithms comes from reverse stress tests, where one seeks to invert the PnL function. Such a function is usually not invertible, but the same optimal transport arguments used for generative methods allow us to invert it in a reasonable way. Numerical experiments show that this approach is quite accurate, and can safely extrapolate to rare events even from small datasets / computational power.

About the Speaker:

Dr. Jean-Marc Mercier is head of R&D at MPG-Partners, a consulting firm specialised in Risk Management. His main task is to develop an open-source kernel-based IA platform, named codpy, focusing on Finance applications for either prospective researches or clients projects. Jean-Marc is a former public researcher, PDE (Partial Differential Equation) specialised. He then turned to industry, and has several years of industry experience as quantitative engineer and business analyst. His research is involved in machine learning, artificial intelligence, computer sciences, with 30+ published papers. Jean-Marc holds a Ph-D in Applied Mathematics from Bordeaux University.