Tools for modeling sparse high-dimensional multivariate time series
For a demonstration of the package's capabilities, see the recently updated BigVAR Tutorial, the Shiny App, or the slightly out of date user guide available on Arxiv.
Note: This package utilizes C++11, so it requires a compiler with C++11 support (which should include most modern compilers) and a later version of R (version 3.1 is the oldest that I can confirm works).
To install the development version of BigVAR, after installing the devtools package, run the following commands
library(devtools)
install_github("wbnicholson/BigVAR/BigVAR")The stable version is available on cran.
If you experience any bugs or have feature requests, contact me at wbn8@cornell.edu.
A minimalist Python implementation (partially inspired by this abandoned effort) has been released. Currently, it only has the capability to fit Basic or Elastic Net penalty structures. Feel free to suggest other functionality or submit pull requests.
In order to install the Python implementation, clone the repository, navigate to the python directory and run
pip install -e .An example script is below
importnumpyasnpfromBigVAR.BigVARSupportFunctionsimportMultVARSim, CreateCoefMatfromBigVAR.BigVARClassimportBigVAR,rolling_validate# example coefficient matrixk=3;p=4B1=np.array([[.4,-.02,.01],[-.02,.3,.02],[.01,.04,0.3]])
B2=np.array([[.2,0,0],[0,.3,0],[0,0,0.13]])
B=np.concatenate((B1,B2),axis=1)
B=np.concatenate((B,np.zeros((k,2*k))),axis=1)
A=CreateCoefMat(B,p,k)
Y=MultVARSim(A,p,k,0.01*np.identity(3),500)
VARX={}
# construct BigVAR object:# Arguments:# Y T x k multivariate time series# p: lag order# penalty structure (only Basic and BasicEN supported)# granularity (depth of grid and number of gridpoints)# T1: Start of rolling validation# T2: End of rolling validation# alpha: elastic net alpha candidate# VARX: VARX specifications as dict with keys k (number of endogenous series), s (lag order of exogenous series)mod=BigVAR(Y,p,"Basic",[50,10],50,80,alpha=0.4,VARX=VARX)
res=rolling_validate(mod)
# coefficient matrixres.B# out of sample MSFEres.oos_msfe#optimal lambdares.opt_lambda