A method for high-dimensional probabilistic multivariate time series forecasting

Abstract

Probabilistic forecasting of high dimensional multivariate time series is a notoriously challenging task, both in terms of computational burden and distribution modeling. Most previous work either makes simplifying distribution assumptions or abandons modeling cross-series correlations. A promising line of work exploits scalable matrix factorization for latent-space forecasting, but is limited to linear relationships, is not probabilistic, and not trainable end-to-end. We introduce a novel probabilistic multivariate forecasting method addressing these shortcomings, and demonstrate improved performance on a variety of multivariate datasets.