Generative Models and Predictive Machines with Uncertainty Quantification for Financial Applications

We consider generative and predictive methods based on kernels (RKHS theory). This approach allowed us to consider various Finance applications, ranging from time series prediction with model-free models to intradays, real time pricing / hedging methods with predictive machines, or valuation algorithms of large portfolios depending on many risk sources for XVA or pricing purposes. The proposed approach is not only fast and efficient, but also reliable and explainable, because it is sheltered by solid error estimates, allowing us to fully understand the results. We illustrate our approach with several reproducible numerical examples, relying on an open source project. This led us to a modern and competitive approach of Portfolio Management, coupling machine learning and quantitative analysis, through an original approach mixing RKHS and optimal transport theory.

Download presentation