What is the acceptable value range for p,d and q in ARIMA?
Question
Why we can’t have large value for d?
in progress
3
Machine Learning
4 years
1 Answer
1200 views
Grand Master 0
Answer ( 1 )
The answer is subjective and there is no ideal answer.
It depends on the frequency of your data as well as the no of observations you have.
If you have quarterly data, may be up to 3 lags, if monthly data then up-to 6 lags,
if the data is at very granular level like no of hours,minutes etc. then the no of lags would increase.
If you do not have large no of observations, then it is better to keep lesser no of lags as
AIC takes into consideration no of lags while imposing penalties and we need AIC to be as low as possible.
when it comes to value of d, generally if you require your time series to difference more than 3 times,
then generally the series you are trying to model is very complex and will not give accurate predictions even
on modeling.