Multiple Change Point Detection in Structural VAR Models
block fused sparse group lasso step (first step).
block fused lasso step (first step for BSS).
local screening step (second step).
Shrinkage function for low-rank soft-thresholding
Shrinkage function for sparse soft-thresholding
Function to deploy simulation with LSTSP algorithm
Simulation function for TBSS algorithm
Backward selection function for the second screening step
BIC and HBIC function
cluster the points by neighborhood size a_n
Compute local loss function.
Auxiliary function to calculate loss at the estimated change points
Main loss function for quardratic loss
Prediction function (single observation)
Function to print the change points estimated by VARDetect
Proximal function for nuclear norm penalty
Proximal function with nuclear norm penalty updating
Proximal function with l1-norm penalty updating
An auxiliary function in FISTA algorithm
helper function for detection check
Backward elimination algorithm for screening in the second step
Single change point detection for low-rank plus sparse model with cros...
cross-validation index function, separate train and test sets
a function to apply cross-validation to select tuning parameter by min...
Single change point detection for low-rank plus sparse model structure
Function for detection performance check
Evaluation function, return the performance of simulation results
First step rolling window function
A function to solve low rank plus sparse model estimation using FISTA ...
A helper function for implementing FISTA algorithm to estimate low-ran...
Gradient function of quardratic loss
Function for Hausdorff distance computation
Select the lag of the VAR model using total BIC method
Main function for the low rank plus sparse structure VAR model
Nuclear norm penalty for low-rank component
Objective function
Plotting the output from VARDetect.result class
Function to plot the sparsity levels for estimated model parameters
Function to plot Granger causality networks
Plot the AR coefficient matrix
Prediction function (block)
Generate VAR(p) model data with break points
L1-norm penalty for sparse component
Function to summarize the change points estimated by VARDetect
A function to summarize the results for simulation
Block segmentation scheme (BSS).
Exhaustive search step (third step).
Implementations of Thresholded Block Segmentation Scheme (TBSS) and Low-rank plus Sparse Two Step Procedure (LSTSP) algorithms for detecting multiple changes in structural VAR models. The package aims to address the problem of change point detection in piece-wise stationary VAR models, under different settings regarding the structure of their transition matrices (autoregressive dynamics); specifically, the following cases are included: (i) (weakly) sparse, (ii) structured sparse, and (iii) low rank plus sparse. It includes multiple algorithms and related extensions from Safikhani and Shojaie (2020) <doi:10.1080/01621459.2020.1770097> and Bai, Safikhani and Michailidis (2020) <doi:10.1109/TSP.2020.2993145>.