Abstract 
This work provides a novel framework for modeling time series displaying multiple seasonal patterns. The methodology presented builds on recent advancements in high-dimensional factor analysis, focusing on time series tensor factor models. The method is applied in a the domain of energy demand forecasting, considering hourly data of energy demand in the U.S. We show that the proposed method effectively captures the multi-seasonal patterns in the data, providing interpretable loading values in line with the expected characteristics of the underlying phenomena. Albeit the extraction of seasonal components is achievable through the simpler matrix factor models, we argue that a tensor factor model provide stronger asymptotic properties based on a thoughtful extension of the original dataset encompassing multiple electricity providers. Tensor factor models’ results are evaluated against classical vector factor models and functional time series methods, demonstrating the superior forecasting accuracy of the tensor approach. The analyses in this work provide a robust framework for future model extensions by effectively accounting for complex seasonal patterns. These models can be integrated into more complex empirical settings, allowing for the incorporation of additional variables to enhance accuracy in diverse forecasting scenarios.

Keynote speaker
Matteo Barigozzi is Full Professor of Econometrics and Political Economy at the Department of Economics of the Alma Mater Studiorum – Università di Bologna. Previously, he was Associate Professor in the Department of Statistics at the London School of Economics and Political Science. Matteo is a leading econometrician, with a particular interest in time series and in dynamic factor models. His scientific contributions are published in very prestigious journals such as the Journal of Econometrics, the Review of Economics and Statistics and the Journal of Business & Economic Statistics. He is also an affiliated member of CREST and ECARES.