[R] Different results with arima in R 2.12.2 and R 2.11.1

[Uwe Ligges]

On 10.09.2011 01:54, Stephen Sefick wrote:
> Those are not "very" different.  I would suspect they are within
> computer rounding error.  Only numbers of log2 can be represented
> exactly, could that be it?  Did you compiled the two versions
> differently, or could something have changed between 2.11.1 to 2.12.2 -
> both of which are old versions of R?  If you provide a reproducible
> example I will gladly run it to compare your results with 2.13.1.  Maybe
> somebody else will have a more helpful response.


No, you are so right: We do not no OS, architecture, 32 vs 64-bit, BLAS 
in use, compiler differences etc. on that machine, and minor numerical 
differences can be caused by anything.

Uwe Ligges

> FWIW,
>
> Stephen Sefick
>
> On 09/09/2011 04:43 PM, Luis Felipe Parra wrote:
>> Hello , I have estimated the following model, a sarima:
>>
>> p=9
>> d=1
>> q=2
>> P=0
>> D=1
>> Q=1
>> S=12
>>
>>
>> In R 2.12.2
>> Call:
>> arima(x = xdata, order = c(p, d, q), seasonal = list(order = c(P, D, Q),
>> period = S),
>>       optim.control = list(reltol = tol))
>>
>> Coefficients:
>>            ar1     ar2      ar3     ar4     ar5     ar6      ar7      ar8
>> ar9
>>         0.3152  0.8762  -0.4413  0.0152  0.1500  0.0001  -0.0413  -0.1811
>>    0.0646
>> s.e.  0.0865  0.0885   0.1141  0.1181  0.1196  0.1220   0.1120   0.0908
>>    0.0865
>>             ma1      ma2     sma1
>>         -0.0221  -0.9779  -0.7635
>> s.e.   0.0539   0.0534   0.0834
>>
>> sigma^2 estimated as 1.965e+17:  log likelihood = -3316.07,  aic = 6658.13
>>
>>
>> and in In R 2.11.1
>> Call:
>> arima(x = xdata, order = c(p, d, q), seasonal = list(order = c(P, D, Q),
>> period = S),
>>       optim.control = list(reltol = tol))
>>
>> Coefficients:
>>            ar1     ar2      ar3     ar4     ar5    ar6      ar7      ar8
>> ar9
>>         0.3152  0.8761  -0.4413  0.0153  0.1500  0.000  -0.0413  -0.1810
>>    0.0646
>> s.e.  0.0865  0.0885   0.1141  0.1181  0.1196  0.122   0.1120   0.0908
>>    0.0865
>>             ma1      ma2     sma1
>>         -0.0221  -0.9779  -0.7635
>> s.e.   0.0539   0.0534   0.0834
>>
>> sigma^2 estimated as 1.965e+17:  log likelihood = -3316.07,  aic = 6658.13
>>
>> and as you can see in the results some coefficients (for example ar2 and
>> ar8) are different in the different R versions. does anybody know what might
>> be going on. Was there any change in the arima function between the two
>> versions?
>>
>> Thank you
>>
>> Felipe Parra
>>
>>
>>
>> ______________________________________________
>> R-help at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-help
>> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
>> and provide commented, minimal, self-contained, reproducible code.
>
>
> 	[[alternative HTML version deleted]]
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.