[R-SIG-Finance] version 1.4.7 of vars: HC estimation

[Matthieu Stigler]

Hi

As also announced on r-pkgs, the package vars has been updated to 
version 1.4.7

The main change concerns the compatibility with the sandwich/lmtest 
package, thanks to the nice help of Achim Zeileis and Bernhard Pfaff.

One can hence now use heteroskedasticity-consistent covariance matrix 
for inference in VAR models. This is available through function 
coeftest() from lmtest package:

library(vars)
data(Canada)
va<-VAR(Canada, p=2)
coeftest(va, vcov.=vcovHC)


The second change is an implementation of this procedure for the granger 
causality() test, leading to a heteroskedasticity-robust version of the 
test:

causality(va, vcov.=vcovHC)

As the distribution of the test is different whether there is really 
heteroskedasticity or not, a wild bootstrap procedure has been implemented:

causality(va, vcov.=vcovHC, boot=TRUE)


Those procedures above are especially interesting when one is modelling 
a VAR and finds  heteroskedasicity but is reluctant to model a 
multivariate GARCH if the focus is only on the conditional mean 
parameters, as advocated by Hafner and Herwartz (2009).

Matthieu Stigler

Hafner, C. M. and Herwartz, H. (2009) Testing for linear vector
     autoregressive  dynamics under multivariate generalized
     autoregressive heteroskedasticity,  _Statistica Neerlandica_,
     *63*: 294-323