Hey,
I have a few doubts with regard to the usage of the auto.arima function from
the forecast package in R.
*Background:*
I have a set of about 50 time-series for which I would like to estimate the
best autroregressive model. (I want to estimate the coefficients and order
of p). Each of the series is non-stationary and are also have a non-normal
distribution. The data is non-seasonal.
My objective is to group these 50 odd time-series into 6-7 groups and apply
the same auto-regressive model.(Essentially want a best fit auto-regressive
model for each of the groups).
For a single time-series if I apply:
fit<-auto.arima(<series1>,d=NA,D=0,max.p=6,max.q=0,max.order=6,stationary=F,ic=c("aic"),trace=T,allowdrift=F)
will the differencing be done internally and the final coefficients for the
AR parameters be outputted by the coef(fit) function? Or do I have to make
the series stationary before I apply the auto.arima function?
I.e if finally my result is something like
Coefficients:
ar1 ar2 intercept
0.1561 -0.4495 635.1266
s.e. 0.2076 0.1967 22.0342
for the above command.
will my model be yt=0.1561*yt-1 -.4495*yt-2 + 635.1266 only?
Kindly shed some light on the above issues.
Also if I want my final model to be a composite one involving expoential
smoothing and autro-regressive terms what is the best mode of action?
Thanks and regards,
Kishan
--
A. Phani Kishan
3rd Year B.Tech
Dept. of Computer Science & Engineering
IIT MADRAS
Ph: +919962363545
[[alternative HTML version deleted]]
