A1252
Title: Fitting latent variable models to multivariate time series
Authors: Michael Eichler - Maastricht University (Netherlands) [presenting]
Abstract: In systems that are affected by latent variables conditional independences are often insufficient for inference about the structure of the underlying system. One common example is a system in which four observed variables $X_1$, $X_2$, $X_3$, and $X_4$ are conditionally independent given a fifth unobserved variable $Y$. While there are no conditional independences among the observed variables, they must satisfy the so-called tetrad constraints. In the time series case, these can be expressed in terms of the spectral matrix of the observed variables. We discuss how these constraints can be used to fit time series models that involve latent variables. In particular, we consider so-called latent variable models in which the observed variables are independent given the latent variables while the latent variables are interconnected.
