Abstract

A latent Volterra contagion channel has no single observability level. This paper studies sequential information in the bivariate Gaussian Volterra setting and separates finite-resolution evidence, oracle short-horizon prediction, and prediction from observed Gaussian channels. The finite-resolution experiment recovers the path-detectability boundary $H_{XY}=H_Y+\frac{1}{4}$ as the rate at which likelihood evidence accumulates. The oracle prediction problem has a different boundary, $H_{XY}=H_Y$: rougher contagion gives a leading prediction gain, while smoother contagion is dynamically latent at leading order. After conditioning on the target asset's own observed past, the oracle gain can be screened by the observable filtration. Dynamic observability is therefore the residual predictive object left after pricing visibility, path-space detectability, and feasible prediction have been separated.

Citation

Vidal Llauradó, Joan. “Dynamic Observability of Latent Contagion: Sequential Revelation and Observable Screening.” 2026. doi:10.2139/ssrn.6539199