0

From my model selection based on information criteria such as SBIC returns me a ARMA(0,0) model. Should I choose it or reject it? and Why?

Need some guidance on this.

lakshmen
  • 1,133
  • 2
    This is impossible to answer as is. Please provide a lot more context. What are your variables? What models were tested with SBIC? What did autocorrelation and partial autocorrelation look like? How much serial correlation is there? A plot of the variable would also help. – Peter Flom Dec 15 '13 at 15:09
  • (1) Why did you include an ARMA(0,0) model in the set of candidate models in the first place? It hasn't got less plausible through having the lowest information criterion value of the lot. (2) Is there any reason to worry that none of the candidate models is much good? - are there signs of assumption violations or more complex structure than allowed for in any? – Scortchi - Reinstate Monica Dec 16 '13 at 12:56

1 Answers1

1

It seems that your data is "white noise", so the information criterion suggests you an ARMA(0,0) model.

For example in R:

library(forecast)
set.seed(123)
d<-rnorm(1000)
auto.arima(d)

gives you an ARMA(0,0) model. So there seems to be no structure in your data which can be exploited in the context of an ARMA model.

DatamineR
  • 1,627
  • 4
  • 19
  • 26
  • 2
    It is true that white noise will yield a ARMA(0,0) model, but it is not necessarily true that all ARMA(0,0) models are white noise. There could be something else going on, not captured by the model. Or N could be low. Or there could be a lot of noise on top of something actual. – Peter Flom Dec 15 '13 at 15:30