Generalized principal component analysis for moderately non-stationary vector time series

Research output: Contribution to journalArticle


This paper extends the principal component analysis (PCA) to moderately non-stationary vector time series. We propose a method that searches for a linear transformation of the original series such that the transformed series is segmented into uncorrelated subseries with lower dimensions. A columns' rearrangement method is proposed to regroup transformed series based on their relationships. We discuss the theoretical properties of the proposed method for fixed and large dimensional cases. Many simulation studies show our approach is suitable for moderately non-stationary data. Illustrations on real data are provided.
Original languageEnglish
Number of pages43
JournalJournal of Statistical Planning and Inference
Early online date12 Nov 2020
Publication statusE-pub ahead of print - 12 Nov 2020


  • dimension reduction
  • Eigenanalysis
  • moving cross-covariance
  • moving cross-correlation
  • multivariate time series
  • non-stationary data

Fingerprint Dive into the research topics of 'Generalized principal component analysis for moderately non-stationary vector time series'. Together they form a unique fingerprint.

Cite this