README.md

In markov10000/factorcpt: Simultaneous Change-Point and Factor Analysis

factorcpt

implements a two-stage methodology for consistent multiple change-point detection under factor modelling. It performs multiple change-point analysis on the common and idiosyncratic components separately, and thus automatically identifies their origins. The package also provides options to implement Binary Segmentation, Wild Binary Segmentation, Sparsified Binary Segmentation, Wild Sparsified Binary Segmentation, or Double CUSUM Binary Segmentation algorithms, which are proposed for multiple change-point detection in high-dimensional panel data with breaks.

See

M. Barigozzi, H. Cho and P. Fryzlewicz (2018) Simultaneous multiple change-point and factor analysis for high-dimensional time series.

Y.-N. Li, D. Li and P. Fryzlewicz (2022) Detection of multiple structural breaks in large covariance matrices.

Installation

The developmental version can be installed from within R using the devtools-package:

library(devtools) install_github("markov10000/factorcpt")

Usage

We can generate an example dataset

set.seed(1)
x1 <- matrix(rnorm(40*50),nrow=40,ncol=50)
x2 <- matrix(rnorm(40*50),nrow=40,ncol=50)*1.3
x <- cbind(x1,x2)

Fit

model1.res <- factor.seg.alg(x, r=1, do.parallel=0, idio.diag=F)

Or use get.args() to get arguments

args2 <- get.args(x,ref='baseline') #ref='baseline', 'BCF18', or 'LLF22'
model2.res <- factor.seg.alg(x, args=args2)
#or
#model2.res <- factor.seg.alg(x, r=1, args=args2)

Print results of model1 and model2

model1.res$common.est.cps
model2.res$common.est.cps

model1.res$idio.est.cps
model2.res$idio.est.cps

markov10000/factorcpt documentation built on Feb. 4, 2022, 8:55 a.m.