Skip to content

help on vector auto-regressive model

7 messages · Luna, Megh Dal, Spencer Graves +2 more

Megh Dal
# 
I guess same question was asked in R-forum and already answered. Before
throwing same question in different forum you better follow it up properly.
Luna Moon wrote:

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Spencer Graves
    

    

      
      #
    

  


  

      
Hi, Megh: 


      What do you mean by "R-forum", and how can I search it? 


      I just searched the "Nabble R Forum" for "goodness-of-fit in a VAR 
model", and didn't see anything that looks like this.  I did see copies 
of this question asked on R-help and Rmetrics.  However, the "Nabble R 
Forum" is not a mailing list by itself on which people ask questions, as 
your comment seems to imply, but a search facility like "RSiteSearch" 
for the archives from "R help", "Rmetrics", "R devel", and "Rcom-l".  I 
also tried RSiteSearch, with the same negative results.


       What is the question and answer to which you referred? 


      Thanks,
      Spencer Graves

        
megh wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Megh Dal
    

    

      
      #
    

  


  

      
this question was asked (and answered) on http://www.nabble.com/help-on-vector-auto-regressive-model-td25099737.html

If anyone post the same question in multiple forums then it would be always courtesy to mention that like, same question was there in so-so forums and no satisfactory answers were provided. It would also be good to mention why it is not satisfactory. When an expert answer an query he spends (I would say invest) lot of times on that and therefore after spending so much time on that, suddenly he might discover that was already answered. Therefore if it is mentioned that was already posted in some other places then he might look into that and perhaps can provide a better solution.

I do agree that ""Nabble R  Forum" is not a mailing list by itself" however I did not mean that even, rather meant to say, it was already answered in r-help (address given above).

Thanks

        
--- On Mon, 8/24/09, spencerg <spencer.graves at prodsyse.com> wrote:

        

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

            

      
      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Matthieu Stigler
    

    

      
      #
    

  


  

      
Right it's good to mention if the question was asked on antoher list,
to avoid cross-posting.

I'm not convinced by the answer there, neither by the question itself:
I'm indeed not sure goodness of fit is a relevant notion for time
series data.

To the answer: ACF/PACF and normality tests just tell you if you the
assumption needed for inference seem to hold or not. In analogy to the
usual linear regression case, this is not the point whe nassesing
goodness of fit I believe.

To the question: for time series model where you add lags, indicators
such R^2 go to 1 due to the presence of the lagged variables, so are
not indicative... You may still look at individual R^2 and make
comparisons which variable is well explained in the model and which is
not (for that rather use FEVD maybe)... But overall R^2 does not exist
to my knowledge (nor makes sense in my opinion).

Well this is rather a personal view...

Hope this helps

Mat

2009/8/24 Megh Dal <megh700004 at yahoo.com>:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Christofer Bogaso
    

    

      
      #
    

  


  

      
By R-square, if you meant to say, how much variance is explained by the
explanatory variables then, I would like to suggest to see Forecast variance
decomposition.

        
matifou wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Spencer Graves
    

    

      
      #
    

  


  

      
Hi, Megh: 


      Thanks.  I found that post but failed to see how to view the 
reply.  With your confirmation that a reply was there, I was now able to 
find it -- and learned something about how to use Nabble. 


      If that doesn't answer Luna Moon's question, it needs to be 
rewritten as you say, acknowledging the reply from the earlier respondent. 


      Best Wishes,
      Spencer

        
Megh Dal wrote: