Description
Dynamic Principal Components for Periodically Correlated Functional Time Series.
Description
Method extends multivariate and functional dynamic principal components to periodically correlated multivariate time series. This package allows you to compute true dynamic principal components in the presence of periodicity. We follow implementation guidelines as described in Kidzinski, Kokoszka and Jouzdani (2017), in Principal component analysis of periodically correlated functional time series <arXiv:1612.00040>.
README.md
pcdpca
Implementation of "Dynamic principal components of periodically correlated functional time series".
Two examples in demo directory:
- pm10 data from Graz (comparizon with DFPCA paper)
- simplation with parametrized periodicity
Installation
library("devtools")
install_github("kidzik/pcdpca")
Running a demo
library("pcdpca")
demo("simulation")
demo("pcdpca.pm10")
Usage
Let X be a multivariate time series, a matrix with n observations and d covariates, periodic with period = 2. Then
FF = pcdpca(X, period=2) # finds the optimal filter
Yhat = pcdpca.scores(X, FF) # applies the filter
Yhat[,-1] = 0 # forces the use of only one component
Xhat = pcdpca.inverse(Yhat, FF) # deconvolution
cat(sum((X-Xhat)^2) / sum(X^2)) # variance explained