Date & Time: June 7th, 14:00 - 15:00
Location: Seminar Room M 3170-73 at Dept. of Automatic Control, LTH 
Author: Oscar Gummesson Atroshi and Osman Sibai
Title: Deep hedging of CVA
Supervisor: Manu Upadhyaya
Co-supervisors:  Magnus Wiktorsson (Lund University),  Shengyao Zhu (Nordea Markets)
Examiner: Johan Eker

Abstract:
Large banks are exposed to several financial risks. Regulations have been introduced to manage these risks to avoid scenarios like the 2008 financial crisis. One such set of regulations is the Baselframework, developed by the Basel committee on Banking Supervision. A crucial part of this regulation is the credit valuation adjustment (CVA). To reduce the risks associated with CVA, banks need to hedge it. Hedging CVA poses significant challengesdue to its sensitivity to market conditions and the complexity of underlying assets, including nonlinear phenomena like cross-gamma and wrong-way risk, which significantly complicate the development of effective risk management and optimization approaches.This study uses an approach known as deep hedging that employs reinforcement learning to develop robust hedging strategies, circumventing the complexities traditionally encountered with analytic models. Experimental simulations compare traditional delta hedgingagainst our reinforcement learning-based strategy, revealing insights into their performance and actions. Two different market models are deployed and evaluated. The RL strategies show positive results, especially in the first smaller model, indicating thechallenges for successfully solving high-dimensional problems. This lays the groundwork for additional studies and demonstrates that reinforcement learning can support the execution of hedging strategies for CVA.