Autocorrelation process

[master85]

Hello together,

does anybody know about the following problem within an ols-regression:

if I put ar-processes as independent variable (ar(1), ar(2)....), the number of included observations declines.
eg. before including ar-processes, I have 500,000 observations, after the introduction, I only have 200,000....

I have no idea why this happens. does anybody haven an advice?

Thank you for your help!!!

[EViews Gareth]

When you have an AR(1) you need data for not only the current observation, but also last period's observation. When you have an AR(2) you need data for the current observation, plus the previous two observations. This means for an observation to be included in a regression, not only does it require data for all the variables, it also requires data for the lagged values of those variables. Thus the number of observations that can be used drop significantly, especially if you have NAs in your data (because if you have a single NA, in a normal regression that would cause that single observation to be dropped, but in an AR(1), that single observation plus the next observation will be dropped. In an AR(2), that single observation plus the next two observations will be dropped).