R Time Series Issues I would like to know more details about R issues mentioned in Time Series Analysis and Its Applications: With R Examples. 
For e.g. the first problem still exists in R version 3.0.1
# generate an AR(1) with mean 50
set.seed(66)      # so you can reproduce these results
x = arima.sim(list(order=c(1,0,0), ar=.9), n=100) + 50   
mean(x)  
  [1] 50.60668   # the sample mean is close
arima(x, order = c(1, 0, 0))  
  Coefficients:
           ar1  intercept  <--  here's the problem
        0.8971    50.6304  <--  or here, one of these has to change
  s.e.  0.0409     0.8365

*direct copy from the webpage
Theoretical model of above simulation is:
$X_t = 5 + 0.9 X_{t-1} + Z_t$, where $\{ Z_t\} $ is the white noise
Is R following particular model convention or is R wrong?  If it is wrong, it is still in R core package and I wonder why it isn't changed.
Also, it would be very helpful if I get some hint on other five issues.
 A: In a case like this, first read the help page carefully. It might be that the implementation in R uses conventions and parametrizations that differ from what you expect. The arima help page clearly states in the Details section that the ARMA representation used applies to X - m if include.mean = TRUE, which is the default for ARMA models. The documentation of the include.mean argument says: 
Should the ARMA model include a mean/intercept term? 
Hence, from the coefficients output and reading the help page, the intercept parameter is m.
Surely, the casual user might expect that the intercept is something else, because he or she is using a different parametrization. This can be inconvenient, and if the conventions used by R differ from what most users expect, it can even be considered a design flaw. But it is not a bug! The implementation computes what is intended and documented. 
But why not just change the name of the parameter in the output to "mean"? Doing so might break other peoples code that rely on the parameter names, as well as break code internally in the stats package. Making a design change is really difficult, and even if many find it inconvenient that the parameter is called "intercept" and not "mean", you should not expect that this will ever be changed in R. 
And please don't put more faith into SAS than it deserves. 
