Better hedging of CVA with reinforcement learning
(2025) In Master's Thesis in Mathematical Sciences FMSM01 20251Mathematical Statistics
- Abstract
- This thesis investigates the application of reinforcement learning (RL) to the problem of hedging Credit Valuation Adjustment (CVA), a key component of counterparty credit risk in over-the-counter (OTC) derivatives. While prior studies have demonstrated the effectiveness of RL in hedging simple financial instruments, this work extends the literature by exploring whether RL, specifically the Proximal Policy Optimization (PPO) algorithm, can effectively
hedge CVA in the presence of realistic market conditions, including transaction costs and wrong-way risk. The study is conducted within a simulated environment that models interest rates using a one-factor Hull-White model and default intensity through a Jump-extended Cox–Ingersoll–Ross... (More) - This thesis investigates the application of reinforcement learning (RL) to the problem of hedging Credit Valuation Adjustment (CVA), a key component of counterparty credit risk in over-the-counter (OTC) derivatives. While prior studies have demonstrated the effectiveness of RL in hedging simple financial instruments, this work extends the literature by exploring whether RL, specifically the Proximal Policy Optimization (PPO) algorithm, can effectively
hedge CVA in the presence of realistic market conditions, including transaction costs and wrong-way risk. The study is conducted within a simulated environment that models interest rates using a one-factor Hull-White model and default intensity through a Jump-extended Cox–Ingersoll–Ross (JCIR) process. Multiple PPO variants are tested, evaluating the impact of reward functions, observation and action spaces, and hyperparameter settings. Results show that PPO is capable of learning competitive hedging strategies and can outperform traditional delta hedging under certain market
frictions. However, further research is required for the model to be applicable in the industry. Directions for future research include incorporating funding costs, liquidity constraints, and portfolio-level netting effects. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9197473
- author
- Byman, Isabelle LU and Karlsson, Oscar
- supervisor
- organization
- course
- FMSM01 20251
- year
- 2025
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- OTC, xVA, CVA, Counterparty Credit Risk, Interest Rate Swaps, Hull-White Model, Machine Learning, Reinforcement Learning
- publication/series
- Master's Thesis in Mathematical Sciences
- report number
- LUTFMS-3520-2025
- ISSN
- 1404-6342
- other publication id
- 2025:E45
- language
- English
- id
- 9197473
- date added to LUP
- 2025-06-12 09:52:08
- date last changed
- 2025-06-13 09:37:30
@misc{9197473, abstract = {{This thesis investigates the application of reinforcement learning (RL) to the problem of hedging Credit Valuation Adjustment (CVA), a key component of counterparty credit risk in over-the-counter (OTC) derivatives. While prior studies have demonstrated the effectiveness of RL in hedging simple financial instruments, this work extends the literature by exploring whether RL, specifically the Proximal Policy Optimization (PPO) algorithm, can effectively hedge CVA in the presence of realistic market conditions, including transaction costs and wrong-way risk. The study is conducted within a simulated environment that models interest rates using a one-factor Hull-White model and default intensity through a Jump-extended Cox–Ingersoll–Ross (JCIR) process. Multiple PPO variants are tested, evaluating the impact of reward functions, observation and action spaces, and hyperparameter settings. Results show that PPO is capable of learning competitive hedging strategies and can outperform traditional delta hedging under certain market frictions. However, further research is required for the model to be applicable in the industry. Directions for future research include incorporating funding costs, liquidity constraints, and portfolio-level netting effects.}}, author = {{Byman, Isabelle and Karlsson, Oscar}}, issn = {{1404-6342}}, language = {{eng}}, note = {{Student Paper}}, series = {{Master's Thesis in Mathematical Sciences}}, title = {{Better hedging of CVA with reinforcement learning}}, year = {{2025}}, }