Tail Risk Diversification Using Deep Path Integrals Reinforcement Learning

Conditional Value at Risk (CVaR) is frequently employed as a tail risk measure since it is essential to effective tail risk management. While symmetric risk metrics consider both upside and downside risk, CVaR is an asymmetric risk measure that regulates and manages the downside risk of a portfolio. This study employs deep reinforcement learning under the framework of path integral formulation to propose a diversification model for financial derivatives. The developed model simultaneously estimates the option price using path integral based on the Black-Scholes model and also forecast the CVaR to develop an optimal hedging strategy using deep reinforcement learning. Using Monte Carlo simulation, this study finds that the proposed model outperforms conventional hedging strategy. Furthermore, to test the model with the real trading environment dataset, this study employs American options data and found similar performance