One-Sided Dynamic Principal Components

Functions to compute the one-sided dynamic principal components ('odpc') introduced in Peña, Smucler and Yohai (2019) . 'odpc' is a novel dimension reduction technique for multivariate time series, that is useful for forecasting. These dynamic principal components are defined as the linear combinations of the present and past values of the series that minimize the reconstruction mean squared error.


odpc 2.0.0

  • The higher order ODPC are now defined as linear combinations of the observations (instead of linear combinations of the residuals of the previous fits). This way a smaller number of observations is lost when new components are added. The forecasting performance of the new definition is very similar to that of the old one.
  • Option to use gdpc as starting point for the iterations is no longer supported (for now).
  • Added functions to automate the choice of the tuning parameters for odpc: cv.odpc is based on minimizing the cross-validated forecasting error, crit.odpc is based on minimizing and information criterion.
  • Added several tests.

Reference manual

It appears you don't have a PDF plugin for this browser. You can click here to download the reference manual.


2.0.4 by Ezequiel Smucler, a year ago

Browse source code at

Authors: Daniel Peña <[email protected]> , Ezequiel Smucler <[email protected]> , Victor Yohai <[email protected]>

Documentation:   PDF Manual  

Task views: Time Series Analysis

GPL (>= 2) license

Imports methods, Rcpp, forecast, parallel, doParallel, foreach, MASS

Suggests testthat

Linking to Rcpp, RcppArmadillo

See at CRAN