The goal of this project was to turn a full GJR-GARCH volatility model into a compact neural surrogate that reproduces its return distributions with high-level accuracy. This required more than training a network. We first… How We Built the GJR-GARCH Pretrained Model