Seminário de Pesquisa 03/12
Deixe um comentário / Economia Aplicada, Seminários de Pesquisa / Por Helena Camillo
Ocorrerá no dia 03/12/2025, quarta-feira, às 14h na sala 402 – FCE, o Seminário de Pesquisa “Clustering of High-Dimensional Functional Time Series”, na área de Economia Aplicada, ministrado pelo Prof. Klaus Boesch (UFRGS / IFSul – Charqueadas).
Resumo:
Analysing large, high-dimensional time series such as global yield curves poses substantial methodological and computational challenges. We propose a robust two-stage clustering framework tailored for high-dimensional functional time series (FTS), addressing an important gap between the clustering literature and functional time series research. Our approach combines a generalised principal components analysis (GPCA) of lag-covariance operators for dimension reduction with subsequent clustering of the estimated factor loadings, using both standard and robust clustering algorithms. This strategy exploits the latent factor structure to capture temporal dependence and to obtain a compact representation of the multivariate functional structure. A Monte Carlo study spanning a range of signal-to-noise ratios and contamination levels compares the proposed estimators with traditional benchmarks. It shows that GPCA-based methods achieve accurate subspace recovery and superior clustering performance, particularly in robust clustering implementations under challenging scenarios with outliers. Applied to daily government bond yield curves from 31 countries (2013–2022), the framework yields a stable partition into six economically interpretable clusters, separating advanced, resource-dependent, and several groups of emerging and structurally distinct economies.