Outliers in High-Dimensional Dynamic Factor Models

• Grant number: 24/20677-0
• Support Opportunities: Scholarships abroad - Research Internship - Scientific Initiation
• Start date: January 01, 2025
• End date: March 01, 2025
• Field of knowledge: Physical Sciences and Mathematics - Probability and Statistics - Applied Probability and Statistics
• Principal Investigator: Luiz Koodi Hotta
• Grantee: Cauã Pereira Masseu
• Supervisor: Pedro Galeano
• Host Institution: Instituto de Matemática, Estatística e Computação Científica (IMECC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil
• Institution abroad: Universidad Carlos Iii De Madrid, Campus De Getafe, Spain
• Associated to the scholarship: 24/08171-3 - Study of Robustness of Some Methods Applied to Financial Time Series, BP.IC

Abstract

High-dimensional time series have become increasingly common across various fields. For instance, they are prevalent in climatology, where measurements are taken at numerous locations, as well as in economic panel data and finance. A popular approach to modeling these types of series, in order to overcome the so-called "curse of dimensionality," is to use factor models. Meanwhile, a common challenge in time series analysis is the presence of outliers. To address this issue, the most commonly used approaches are either robust methods or models that explicitly account for outliers, with the latter requiring the initial detection of these anomalies.Galeano and Peña (2024) [Detecting Outliers in High-Dimensional Time Series by Dynamic Factor Models. In Barigozzi, M., Hörmann, S., and Paindaveine, D., editors, Recent Advances in Econometrics and Statistics: Festschrift in Honour of Marc Hallin] adopt the former approach. The objective of this visit is to discuss this paper with the first author, to examine its robustness against other types of outliers beyond those discussed in their study, and to explore two extensions: (1) developing methods that are robust to a broader range of outlier types, including time-varying outliers, and (2) modeling conditional heteroscedasticity and outliers within Dynamic Factor Models (DFMs).