Anchor-Conditioned Probabilistic World Models for Multi-Asset Realized-Volatility Paths
Work in progress, 2026
This project formulates financial volatility forecasting as a passive probabilistic world-model problem. A causal HAR-IV anchor models the predictable level, while a recurrent state-space model learns market-shared and asset-specific innovation dynamics. The project evaluates forecast accuracy, calibration, and joint path realism separately.
Status: Work in progress. Public PDF to be added after the manuscript is finalized.
Recommended citation: Tongren Xiao. "Anchor-Conditioned Probabilistic World Models for Multi-Asset Realized-Volatility Paths." Work in progress, 2026.