For decades, the Black-Scholes framework has served as the bedrock of option pricing, yet real-world markets are rarely so forgiving. Trading frictions, liquidity limits, and stochastic jumps routinely break traditional replication. Designed specifically for quantitative analysts and machine learning practitioners, this book introduces a paradigm shift in derivative risk management. By replacing rigid analytical Greeks with dynamic neural network policies, you will learn to optimally hedge complex portfolios in incomplete markets where classical theories inevitably fall short.
You will master the end-to-end pipeline of designing and training deep hedging algorithms. The text guides you through constructing differentiable market simulators, engineering scale-invariant features, and tailoring loss functions to institutional risk measures like Expected Shortfall. Moving beyond theoretical models, you will train robust policy networks that directly internalize market frictions. By explicitly accounting for transaction costs and position limits, these models learn to balance risk reduction with execution intensity, uncovering optimal, cost-aware trading strategies.
Bridging the gap between academic research and trading desk reality, the material focuses heavily on the engineering required for robust production deployment. You will explore synthetic data generation, domain randomization to combat model misspecification, and rigorous walk-forward backtesting protocols. Complete with safety guardrails, drift detection, and explainability frameworks, this guide equips you with the
