Rough Volatility
MNO outperforms Neural SDE and Neural CDE on rough volatility at all tested Hurst exponents, with the largest margin at H=0.1, the most non-Markovian regime. Targeting the terminal marginal directly without modeling the full non-Markovian…
1 sources - 3 claims
MNO outperforms Neural SDE and Neural CDE on rough volatility at all tested Hurst exponents, with the largest margin at H=0.1, the most non-Markovian regime. Targeting the terminal marginal directly without modeling the full non-Markovian path history can outperform sequential baselines optimized for pathwise fidelity on rough volatility tasks. Autoregressive reuse of MNO whitens temporal roughness toward Brownian scaling, marking the boundary of the model's valid scope.