Skip to content

Forcasting VAR/VEC

6 messages · Megh Dal, Pfaff, Bernhard Dr., Matthieu Stigler +1 more

Original
Megh Dal
# 
Hi dears, I am looking for some function to predict through horizon "h", say,
given the corefficients of a VAR/VEC model. I have moderately gone through
packages "urca" & "vars" but did not find any significant:
[1] "A"              "Acoef"          "arch"           "arch.test"      "B"             
"Bcoef"          "BQ"             "causality"     
 [9] "fanchart"       "fevd"           "irf"            "normality"     
"normality.test" "Phi"            "Psi"            "restrict"      
[17] "roots"          "serial"         "serial.test"    "stability"     
"SVAR"           "SVEC"           "VAR"            "VARselect"     
[25] "vec2var"
[1] "ablrtest"      "alphaols"      "alrtest"       "bh5lrtest"    
"bh6lrtest"     "blrtest"       "ca.jo"         "ca.po"        
 [9] "cajolst"       "cajools"       "cajorls"       "lttest"        "plot"         
"plotres"       "punitroot"     "qunitroot"    
[17] "summary"       "unitrootTable" "ur.df"         "ur.ers"       
"ur.kpss"       "ur.pp"         "ur.sp"         "ur.za"       

Can somebofy help in finding any relevant function for prediction?

Thanks,
Pfaff, Bernhard Dr.
# 
?vars::predict
4994.html
*****************************************************************
Confidentiality Note: The information contained in this ...{{dropped:10}}
1 day later
Megh Dal
# 
I tied to post it through Nabble, however it perhaps failed........ Here is my query. Your help will be highly appreciated.

I found a direct problem while implementing predict method. To use that, I need to give values for argument "object": 

object: An object of class ??varest??; generated by ?VAR()?, or an 
          object of class ??vec2var??; generated by ?vec2var()?. 

Here what I have I the estimated coefficients and last few values (required for prediction). Given those, how can I construct object of that class? 

For example, I have VAR[2] model with estimated coefficients: 

A1 <- t(matrix(c(-0.9444135,  0.5289205,  1.54338344, -1.2059849,  0.08658383, -0.7626375, 0.5384060,  1.34211012, -0.1540462,  0.07684465, 
-1.5000912,  0.5975748,  2.08629175, -1.3629879,  0.16405863,  0.5622497, -0.5749668, -0.01207330,  1.6547463, -0.37280903, -1.1190102, 
0.4834678,  1.04152221, -0.5084529,  0.91748918), 5)) 

A2 <- t(matrix(c(-0.5281084,  0.6750694, -0.00700632,  0.5744008, -0.540538174, -0.5115759,  0.6663074, -0.03690401,  0.1015040, 
-0.483984397, -0.6669159,  0.7810531, -0.08478713,  0.2837110, -0.634807434,  0.2474859, -0.2126529,  -0.04611398, -0.2851814, -0.003198122, 
0.1136257, -0.4070806, -0.15362813,  0.1043871, -0.570401589), 5)) 

## and deterministic terms with monthly seasonal dummy (11 dummy variables + constant): 

Mu <- t(matrix(c(0.01971314683, 7.392074e-05, 0.06745512042, 0.03066764063, -0.05716243051, -0.01666261294, -0.02648375478, -0.07739940822, 
-0.08537205872, -0.03454705454, 0.01026347102, 3.93408495893, 0.01124435027, 0.00826573416, 0.07232184812, 0.02643898799, -0.02290620244, 
0.00631679187, -0.02120713774, -0.06034222549, -0.06737971493, -0.02466344307, 0.01574220616, 3.67816873865, 0.01817359276, 0.00511780538, 
0.06717006338, 0.06206760303, -0.04353428003, -0.03033956305, -0.00535023233, -0.06992876937, -0.10020345332, -0.04323120694, 0.00399287988, 
3.01849213684,-0.00696500762, 0.00784220761, 0.00337317617, -0.00771413903, 0.03016416594, 0.02404299199, 0.00752992866, 0.01211247641, 
0.01442465011, 0.00162618787, 0.00646325736, 0.18377741575,0.06172376263, 0.03093119587, 0.00519357485, -0.00720735363, -0.02268228948, 
-0.03428031321, 0.01599783772, -0.01470977952, -0.00498472297, -0.03734476303, -0.03712904004, 2.95905626288), ncol=5)) 

## and last 4 values of TS 

YY <- t(matrix(c(-0.55367226, 3.276980, -0.57634682, 3.827854, 4.248918, -0.57806007, 3.254492, -0.61281320, 3.835869, 4.280090, -0.51697066, 
3.300236, -0.55756556, 3.815981, 4.274613, -0.47619582, 3.320164, -0.48689008, 3.793915, 4.216955), 5)) 

Given those information, how I can predict values for next 2 steps? 

Thanks,
--- On Mon, 12/13/10, Pfaff, Bernhard Dr. <Bernhard_Pfaff at fra.invesco.com> wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

            

      
      
      
    

  
  


  


      

    

    
1 day later

    

      
      

        


  

        

      Matthieu Stigler
    

    

      
      #
    

  


  

      
library(vars)
data(Canada)
modeleVAR<-VAR(Canada, p=2)

class(modeleVAR) # so you have your varest object!
predict(modeleVAR, n.ahead=3)

Hope this helps?

Matthieu

Le 14. 12. 10 18:38, Megh Dal a ?crit :

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    
1 day later

    

      
      

        


  

        

      Megh Dal
    

    

      
      #
    

  


  

      
Thanks Mat for your reply. However this is not what I wanted. I already have the coefficients estimated and using those estimated coefficients I want to forecast. Here your input is TS data, from that data you have estimated coefficients, which is of specific class. Therefore you could use directly predict() function.

As I have coefficients in Matrix class, I cant convert them to specific class for prediction. Is there any way to do that?

2ndly I wondering how without the initial values i.e. last few observations of TS, predict() function in your example could actually predict?

Thanks,

        
--- On Thu, 12/16/10, mat <matthieu.stigler at gmail.com> wrote:

        

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Brian G. Peterson
    

    

      
      #
    

  


  

      
On 12/17/2010 09:37 AM, Megh Dal wrote:

            

      
      
      
    

        
Not what you wanted is different from what you asked, which was how to 
predict a VAR/VEC model.  Bernhard and Mat have given you an answer 
which is not what you wanted to hear, which seems to be that someone 
else would magically have invented the same custom data format that you 
choose to use?

            

      
      
      
    

        
Then you will likely have to modify the predict() function to either 
take the coefficients directly, or you will have to take your custom 
non-package-compliant Matrix and format it into a vars object like that 
that predict() is .

see

?vars::VAR

for the format of the 'varest' list to give you input on what you will 
need to do to transform your data so that it will work with predict()

            

      
      
      
    

        
You need the data to make a prediction.  Period.

   - Brian