Deep Reinforcement Learning for Delta Hedging

This repository implements and compares different approaches to delta hedging, from classical Black-Scholes-Merton (BSM) to modern deep reinforcement learning methods.

Project Overview

The project is structured in three main phases:

1. Data Generation and Simulation Framework
2. Classical Delta Hedging Implementation
3. Deep Reinforcement Learning Implementation (Coming Soon)

Current Features

Synthetic Data Generation

• Geometric Brownian Motion (GBM) for multiple asset paths
• Planned: Heston, SABR, Merton Jump, and Variance Gamma models
• Customizable parameters for volatility, drift, and time horizon

Classical Delta Hedging

• BSM option pricing and Greeks calculation
• Dynamic delta hedging implementation
• P&L calculation and performance metrics
• Visualization tools for hedge performance

Coming Soon: Deep Hedging

The ultimate goal is to implement deep reinforcement learning for hedging options. This approach offers several advantages over classical methods:

1. Model-Free Hedging

   • No assumptions about underlying price dynamics
   • Learns directly from market data
   • Can adapt to changing market conditions

2. Transaction Costs

   • Naturally incorporates transaction costs in the hedging strategy
   • Balances hedging error vs trading costs

3. Market Frictions

   • Handles discrete-time hedging
   • Accounts for bid-ask spreads
   • Manages inventory constraints