American-Option-LSM-Extension

Final exam for Dynamic programming at UCPH. Implementation and analysis of Longstaff–Schwartz Monte Carlo (LSM) for American option pricing including polynomial basis comparison, jump diffusion, quasi-Monte Carlo, and multi-asset extensions with neural network approximation.

📄 Final Report (PDF)
🎞️ Presentation Slides (PDF)

---

Repository structure

• notebooks/ – Single- and multi-asset LSM implementations
• reports/ – Final written report and presentation slides
• Multiasset_environment.yml – Environment file for the multi-asset notebook only

Reproducing results

1. Create environment (multi-asset only):

   conda env create -f Multiasset_environment.yml
   conda activate american-lsm

Post-submission update

Neural-network continuation value now works after applying Bengio’s practical training guidelines for deep networks (learning-rate schedule, initialization, normalization, early stopping, and random hyperparameter search).

Reference: Bengio, Y. (2012). Practical recommendations for gradient-based training of deep architectures. arXiv:1206.5533.