This project implements and compares two approaches to calibrating the Heston stochastic volatility model to real SPX options market data: classical least-squares optimization and a neural network surrogate model.
The Heston model (1993) prices European options by modeling volatility as a mean-reverting stochastic process rather than a constant. Calibrating the model to observed market prices requires solving an optimization problem that is slow under classical methods. This project trains a neural network to approximate the Heston pricing function and uses it as a fast oracle for calibration.
Week 1: Black-Scholes implementation from scratch (baseline model).
notebooks/ - Jupyter notebooks documenting each stage of the project src/ - Production Python modules data/ - SPX options data collected via yfinance paper/ - Final write-up
See requirements.txt
Black, F. and Scholes, M. (1973). The Pricing of Options and Corporate Liabilities. Journal of Political Economy, 81(3), 637-654.
Heston, S. L. (1993). A Closed-Form Solution for Options with Stochastic Volatility with Applications to Bond and Currency Options. Review of Financial Studies, 6(2), 327-343.
Horvath, B., Muguruza, A., and Tomas, M. (2021). Deep Learning Volatility. Quantitative Finance, 21(1), 11-27.