Finance-Insurance
Time: 11.00 am
Date:05th of March 2026
Room 3001

Dario PALUMBO (Université de Venise) “Multivariate Score-Driven Models for Strictly Positive Variables”

Abstract :The paper presents a novel approach for the joint modelling of strictly positive time series. A new score-driven specification based on the multivariate GB2 (MGB2) distribution is introduced. The structure of the MGB2 implies that the joint moments depend on the marginal shape parameters, so that allowing these parameters to evolve over time provides a flexible mechanism to capture time variation not only in scale but also in cross-sectional dependence. In this framework, dynamics in a limited set of shape parameters can induce coherent movements in the implied correlation matrix, offering a parsimonious alternative to fully parameterised correlation models. The paper also introduces a multivariate model for the logarithms of strictly positive variables, based on the multivariate exponential GB2 (MEGB2) distribution. Estimation is equivalent to that of the MGB2 model, but the log formulation facilitates comparison with multivariate models defined on the real line, such as the Gaussian and Student’s t. The empirical performance of both the MGB2 and MEGB2 specifications is assessed using a dataset of realised volatilities. The results indicate that the implied dependence structure is broadly consistent with that obtained from multivariate Gaussian and t score-driven models, while relying on a comparatively parsimonious dynamic specification as the cross-sectional dimension increases.

Organizers:  Jean-Michel ZAKOIAN & Christian FRANCQ