Hidden Markov Linear Quantile Graphical Model

Martedì 28 maggio, alle ore 12.15, presso l'Aula Dono Giannessi del Dipartimento di Economia e Management, avrà luogo il seminario dal titolo "Hidden Markov Linear Quantile Graphical Model".
Relatrice sarà Beatrice Foroni (Università di Pisa).

L'evento fa parte dei Lunch Seminars to PhD students and early career researchers organizzati dal Dipartimento di Economia e Management.

Abstract

Graphical models are employed to describe the interrelationships among multiple variables. In diverse applications such as genome biology, finance, and environmental studies, data often exhibit temporal evolution influenced by hidden variables. Therefore, it is thus essential to model the evolution of interrelationships over time, and Hidden Markov Models (HMMs) are wellsuited to capturing the temporal dynamics of these variables. Estimation of graphical models in HMMs has been addressed by Stadler & Mukherjee (2013), using multivariate Gaussian emission distributions with sparse precision matrices,  which can be interpreted as state specific conditional independence graphs. However, the Gaussian assumption is often too strong to be met in actual applications. To investigate time-varying conditional dependence structures without assuming normally distributed data, we introduce a sparse hidden Markov linear quantile graphical model, showing that the conditional quantile carries information for inferring the conditional independence.