deep-hedging

Implementation of the vanilla Deep Hedging engine

A curated list of resources dedicated to Deep Hedging

Open-source quantitative finance models in Python: option pricing and Greeks, volatility surfaces, deep hedging, portfolio construction and Monte Carlo. Every model runs on free data, with the full write-up on davidariasfinance.com

A package to learn optimal hedges by a deep feed forward neural network, to minimise the terminal error

Coding assignments of the "Machine Learning in Finance & Insurance" course at ETH Zürich (Fall 2024).

Deep Hedging neural network for dynamically pricing and managing derivatives risk under realistic market frictions

Code and reproducibility package for “What Does Deep Hedging Actually Learn? Delta Corrections, Regime Fragility, and Symbolic Distillation.”

Deep Hedging of a European call in a Heston market. TensorFlow implementation of Buehler et al.

Deep-learning option pricing and hedging: a neural surrogate for Asian options benchmarked against Monte Carlo, a rough Bergomi model for 0DTE, a CVaR deep hedging policy, live calibration, and an interactive dashboard.

Deep Hedging under market frictions.

Reproduction of J.P. Morgan's Deep Hedging papers (2019 pathwise, 2022 risk-averse actor-critic), checked number by number

LightGBM-based hedging strategy under Merton's jump diffusion with custom loss and delta approximation

Deep hedging with PPO: RL vs Black-Scholes delta hedging under transaction costs

Deep‑Hedging in PyTorch (MCPG): europäische & amerikanische Optionen mit RSQP‑Risiko, GJR‑GARCH‑Pfade, IV‑Features und Chebyshev‑Pricing inkl. Baselines.

Deep-hedging pipeline: trains a from-scratch PPO agent (JAX) to hedge options under discrete rebalancing + transaction costs, then compares it to Black-Scholes delta hedges across GBM/Heston regimes and two cost levels (5 bps vs 20 bps) on 3,000 paired paths. At low costs BS wins (as theory predicts). At higher costs the same PPO trades 30–45% less

Numpy-only replication of Deep Hedging (Buehler et al. 2019): a neural hedging policy trained via hand-written backprop-through-time to minimize entropic risk under transaction costs, benchmarked against Black-Scholes delta hedging.

Option-pricing and hedging models built from scratch and verified on real market data, with honest caveats. Black-Scholes to Heston to deep hedging.

Neural hedging policies (GRU, deep-set, no-trade-band) minimising CVaR, entropic and spectral risk under transaction costs, on fused CUDA kernels at billions of paths per second, with a deep BSDE pricer.

GPU-accelerated quantitative derivatives engine: FiLM-FNO rough volatility calibration (<1ms), multi-fidelity PDE ensemble autocall pricing (0.99 bps RMSE, SR 26-2 risk guardian), deep hedging, and Next.js/FastAPI streaming (1,126 tests).

Reinforcement Learning derivative hedging (Deep Hedging) vs Black-Scholes baseline.