[R] Bug in arima?

[Prof Brian Ripley]

Here's a reproducible example (R 1.7.0)

set.seed(1)
x <- rnorm(100)
arima(x, order=c(7, 1, 0), fixed = c(NA, NA, NA, 0, 0, 0, NA),
      transform.pars = FALSE)

Coefficients:
          ar1      ar2      ar3  ar4  ar5  ar6     ar7
      -0.6838  -0.4126  -0.2236    0    0    0  0.0454
s.e.   0.0986   0.1115   0.0988    0    0    0  0.0843

and you did not set transform.pars.

Looks like the comment

   fixed: optional numeric vector of the same length as the total
          number of parameters.  If supplied, only non-`NA' entries in
          `fixed' will be varied.  `transform.pars = TRUE' will be
          overridden if any AR parameters are fixed.

has been copied from help(arima0) and is unimplemented in arima.


On Wed, 30 Apr 2003, Richard A. Bilonick wrote:

> I'm using the fixed argument in arima. Shouldn't ar4, ar5, and ar6 
> display as zero in the output?
> 
> Call:
> arima(x = window(log(hhprice), start = c(1990, 1), end = c(2003, 3)), 
> order = c(7,
>     1, 0), xreg = window(ts.union(exa1 = lag(exa, -1), exa12 = lag(exa,
>     -12), exb1 = lag(exb, -1), exc1 = lag(exc, -1), exc12 = lag(exc,
>     -12)), start = c(1990, 1), end = c(2003, 3)), include.mean = FALSE, 
> fixed = c(NA,
>     NA, NA, 0, 0, 0, NA, NA, NA, NA, NA, NA))
> 
> Coefficients:
>          ar1      ar2      ar3      ar4      ar5   ar6      ar7      
> exa1      exa12      exb1      exc1      exc12
>       0.0922  -0.1279  -0.2661  -0.0577  -0.0277  0.02  -0.2167   
> -0.3015     0.3424    0.0281    0.0519     0.1715
> s.e.  0.0789   0.0801   0.0742   0.0000   0.0000  0.00   0.0853    
> 0.0503     0.0515    0.0295    0.0257     0.0329
> 
> Also, is the documentation wrong?
> 
>  From ?arima:
> 
> fixed: optional numeric vector of the same length as the total
>           number of parameters.  If supplied, only non-`NA' entries in
>           `fixed' will be varied.  `transform.pars = TRUE' will be
>           overridden if any AR parameters are fixed.
> 
> The non-NA entries in my fixed argument are zeroes. Aren't these "fixed" 
> to zero so they don't vary when a call is made to optim? I thought that 
> was the purpose of the argument. I only wan ar1, ar2, ar3, and ar7 in 
> the model so I'm setting ar4, ar5, and ar6 to zero.
> 
> My main concern is that the predict.Arima function works correctly when 
> using the fixed argument. I'm assuming, output display notwithstanding, 
> that ar4-ar6 are actually fixed to zero when using fixed. When I try to 
> manually make the forecast, the result is slightly different than what 
> predict.Arima reports. I'm wondering if that is due to these 
> coefficients not being set to zero?
> 
> Rick B.
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-help
> 

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595