# Using ARMA when data is missing

I am using ARMA over a dataset with missing samples. How do I treat them? Would you suggest to make linear/nonlinear interpolation or just keep them out and consider two samples with missing data in between as consecutive samples?

-

There is no need to do anything. An ARMA model can easily be estimated with missing values within the time series. You need to use the state space representation of an ARMA model to compute the likelihood. If you use R, this is already handled automatically via the arima() function.

-
thanks for your comment. What do you mean exactly by computing the likelihood? Are you saying that I should do some kind of regression for the missing samples? –  Bob Jun 25 '12 at 3:36
The best approach to estimating an ARMA model is maximum likelihood estimation. To do that, you need to calculate the likelihood of the data given the model and parameters. Any decent ARMA software will do that, but not all of them do it using a state space representation which is needed in order to handle missing values. –  Rob Hyndman Jun 25 '12 at 18:48
could you, please, provide a reference? I need to implement it myself :) –  Bob Jun 29 '12 at 6:17