ccrostirolla / dynfactor Goto Github PK

This is a public repository for dynfactoR, a package for R which facilitates estimation of dynamic factor models.

Current implementation of main dfm function supports vector auto-regressive type dynamics for factors, missing observations and some statistical identification restrictions. A dynamic factor model estimation will typically return 3 estimates, namely principal component estimator, a two-step estimator as well as quasi-maximum likelihood (QML) estimator. The two latter estimators are based on Kalman filtering and QML estimator is a particular case of EM-algorithm.

dynfactoR is easy to install with the help of devtools:

devtools::install_github("rbagd/dynfactoR")

dynfactoR also provides a dataset from National Bank of Belgium which contains monthly Belgian business and consumer survey data over a sufficiently long period. You can call it with:

data("NBBsurvey")

If you are not familiar with related literature, it could be useful to read a short introduction to dynamic factor models which is packaged as a vignette. Some academic references are also made available.

vignette("dynamic-factors")

In the future, the package will be extended to support more general dynamics as well as Markov-switching factor loading or transition matrices. Currently, these features are early experimental.