Seminar by Christa Cuchiero (University of Vienna)

Event Date: 

Monday, April 27, 2020 - 11:00am to 12:00pm

Seminar by Christa Cuchiero (Univeristy of Vienna). Please note that it will be via Zoom and seminar time is 11:00 am - noon in PST. 

Title: Neural S(P)DEs for infinite dimensional calibration and prediction problems

Abstract: We consider calibration of local stochastic volatility models and yield curve prediction and view these problems from an optimal control perspective. The underlying model class are neural S(P)DEs, i.e. S(P)DEs  with neural networks as coefficients where the parameters of the networks correspond to the controls. In the spirit of generative modeling the optimization problem consists in transporting the measure induced by the neural S(P)DE to a target measure given by the market. The choice of the metric depends on the problem, e.g. in the case of local stochastic volatility models we can choose a usual calibration functional. We can therefore learn the leverage function from the implied volatility surface without any kind of interpolation. In the context of yield curve modeling we learn the volatility function in a Heath-Jarrow-Morton framework from time series data to create arbitrage-free yield curve scenarios. The talk is based on joint work with Wahid Khosrawi and Josef Teichmann as well as Claudio Fontana and Alessandro Gnoatto.