Statistics and Computational Methods Seminar Series - Spring 2026

Speaker: Francesco Sanna Passino (Imperial College London)

Title: Estimation and community detection in network-informed time series models

Abstract:

Multivariate time series often exhibit dependence patterns that are naturally described by an underlying network. In many applications, however, the network is not observed and must instead be inferred from the time series themselves. This talk considers estimation and community detection in network-informed time series models, with a particular focus on models whose dependence structure is governed by a latent stochastic blockmodel. 
We first introduce Network Informed Restricted Vector Autoregression (NIRVAR), a model in which the autoregressive coefficient matrix is determined by latent network communities. The resulting structure provides a parsimonious approach to modelling high-dimensional time series while allowing the underlying communities to be estimated from the data. We then consider the theoretical problem of recovering these communities from the temporal dependence alone. In particular, we study spectral clustering based on the sample covariance matrix and establish conditions under which the latent communities can be exactly recovered, taking into account both the network size and the temporal dependence in the observations. 
The two perspectives are complementary: network structure provides a useful restriction for modelling and forecasting, while temporal dependence provides information with which that structure can itself be recovered. Together, they give a framework for understanding how latent network structure can be estimated and exploited in high-dimensional time series.

This work is based on the preprints https://arxiv.org/abs/2407.13314 and https://arxiv.org/abs/2608.02922, and it is in collaboration with Brendan Martin (UCLA & Imperial), Joshua Agterberg (Illinois), Mihai Cucuringu (UCLA & Oxford) and Alessandra Luati (Imperial & Bologna).

Link Teams: https://teams.microsoft.com/meet/311256699215585?p=KXJfFbqOMlenkyVmvx