Generative Models
Generative Market Paths
Generative models for synthetic asset-price paths, judged on market stylised facts and on recovering known stochastic-volatility dynamics, with a built-in benchmark from simulators whose moments are known.
2026In progress
Trains score-based diffusion on returns and a signature-based generator, with GBM and Heston samplers as ground-truth references.
Judges output against the stylised facts of real markets, heavy tails, volatility clustering, the leverage effect, and aggregational Gaussianity.
Measures distributional distance to reference returns and tests whether the model recovers Heston and rough-Bergomi moments when trained on their paths, before a final real-data test.