Aplikace strojového učení pro výpočet CVA (Credit Valuation Adjustment)

This thesis investigates the application of neural networks, specifically Adam and AdamW networks, to accelerate the Credit Valuation Adjustment (CVA) calculation process, traditionally performed using Monte Carlo methods. The goal is to significantly reduce computational time while maintaining accurate results. By leveraging the predictive capabilities of neural networks, this research aims to optimize financial computations in CVA assessments. A dataset of synthetic Interest Rate Swaps (IRS) was generated, varying in selected parameters. The neural networks were trained to predict the Expected Positive Exposure (EPE). Results show that neural networks can achieve comparable accuracy to Monte Carlo methods, with significantly faster computation times. However, the process of generating artificial datasets and training the neural network is time-consuming. The findings contribute to the ongoing efforts in financial risk management using advanced machine learning techniques.