[R-SIG-Finance] How estimate VAR(p)-model robustly?

Brian G. Peterson brian at braverock.com
Mon Apr 20 19:21:38 CEST 2009

Irene Schreiber wrote:
> Hello,
> Does anyone know about robust estimation of vector autoregressive models
> (VAR(p)) in R? Or in Matlab?
> Currently I am using the function ar().
> The problem is, that the variances of my data change a lot with time, and we
> also have some outliers in the data. That is why, I presume, that we would
> get quite different results when estimating robustly.
> I would be very grateful if someone could help!
> Thanks a lot!
> Irene.
I'll try to remember to respond in greater detail after 
http://www.RinFinance.com/ this Friday/Saturday, but I'll suggest two 
avenues now.

A Bayesian smoothing method should improve your forecasts.  There are 
several Bayesian time series implementations in R.

Also, you may want to take a look at our (Boudt,Peterson,Croux) Journal 
of Risk paper from last year and the Return.clean method implementation 
in PerformanceAnalytics, which implements a robust filtering of time 
series outliers aimed squarely at making better risk predictions out of 


  - Brian

Brian G. Peterson
Ph: 773-459-4973
IM: bgpbraverock

More information about the R-SIG-Finance mailing list