Skip to content

Fixed parameters in an AR (or arima) model

3 messages · Jenný Brynjarsdóttir, Brian Ripley, Henning Rust

Jenný Brynjarsdóttir
# 
Hello

I want to fit an AR model were two of the coefficients are fixed to zero
(the second and third ar-coefficients).

I used the "arima" function with the "fixed" argument but the ar3
coefficient is not set to zero:

==============================================
Call:
arima(x = Y, order = c(4, 0, 0), xreg = 1:23, fixed = c(NA, 0, 0, NA, NA,
NA))

Coefficients:
         ar1  ar2     ar3      ar4  intercept     1:23
       0.5370    0  0.4338  -0.8078     5.2991  -0.0421
s.e.  0.0735    0  0.0000   0.1096     0.1081   0.0079

sigma^2 estimated as 0.02665:  log likelihood = 6.77,  aic = -3.54
===============================================

Why?

Thanks,
Jenn? Brynjarsd?ttir
Brian Ripley
#
On Wed, 14 Jan 2004, Jenn? Brynjarsd?ttir wrote:

            

      
      
      
    

        
See the help page: you have forgotten about transform.pars. In the current
version of R (1.8.1) this works for me, with a warning, so I surmise your
version is not the current one (indeed looks like it is 1.7.0 or earlier).

            

      
      
      
    

        
Call:
arima(x = Y, order = c(4, 0, 0), xreg = 1:23, fixed = c(NA, 0, 0, NA, NA, 
NA))

Coefficients:
          ar1  ar2  ar3      ar4  intercept    1:23
      -0.4515    0    0  -0.0415    -0.3711  0.0236
s.e.   0.2414    0    0   0.2336     0.1991  0.0147

sigma^2 estimated as 0.4303:  log likelihood = -23.05,  aic = 56.11
Warning message:
some AR parameters were fixed: setting transform.pars = FALSE in: arima(Y, 
order = c(4, 0, 0), xreg = 1:23, fixed = c(NA, 0, 0,

If you really have only 23 observations you are expecting a lot here!

            

      
      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Henning Rust
    

    

      
      #
    

  


  

      
Hi,
arima expects an order 4 model by specifying order=c(4,0,0). It looks 
like you want to fit an order 6 model with parameters 2 and 3 fixed to 
0. Try order=c(6,0,0) and fixed=c(NA,0,0,NA,NA,NA,NA), the last NA is 
for the intercept.
Henning

        
Jenn? Brynjarsd?ttir wrote: