Observations and AR model

[lentjeuhz]

Hi,

I have a sample of firm data for which I want to run multiple regressions. I use the same independent variables in every regression, but different dependent variables. Only complete observations (for which all variables are available) are included in the analysis. When estimating I include an AR(1) to correct for autocorrelation.

My problem is that when using different dependent variables I end up with varying number of observations, something that does not happen when I exclude the AR(1) from my model.

How can I set up my AR(1) model using exactly the same observations for all regressions?

Thank you for your help.

Lennart

[startz]

AR(1) needs one lagged observation. You're probably missing that for some variables. Set the smpl to reflect complete observations for all dependent and independent including accounting for the lag. As a guess, just exclude the first observation in your current smpl.

[lentjeuhz]

Hi startz,

Thanks for your help.

That's what I thought as well, therefore I added the following selection criteria to my sample criteria when estimating to ensure that only complete observations are used:

sample: 1998 - 2009 if var1 <> na and var2 <> na ... etc

But it seems that this method doesn't include these selection criteria for the lagged variables needed for the AR(1) process.

Any ideas?

Thanks

[lentjeuhz]

I have solved the problem by generating new series for my dependent variables, setting them equal to NA if one of the other dependent variables is NA as well.