quicker regression using loop or something?

[Xiyuan]

we have 2 datasets: dataset 1: for each id1-id2 pair, inter_id have so many values dataset 2: for each id2-inter_id pair, y and x can have so many values final we want b for each id1-id2 pair: e.g. b_AM is calculated by regressing y on x using (K1+K2+K3) observations

[Xiyuan]

It is best not to merge like proc SQL Create because there are too many observations for datasets 1 and 2，like dataset1 can be as large as 1 million

[Xiyuan]

by the way i have a macro when regressing, specifically, when we determine the whole data used for regression for ID1= A and ID2 =M, we can simply use %fn (variable1=Y, variable2=X) and output a dataset with coefficients of X, standard deviation, etc.
What we need to do further is to label this dataset as AM, and assign a new tag such as BM and AN when new dataset is output later. Finally, we merge to get the final result.

[PeterClemmensen]

So what does b_AM equal in the first obs?

[Xiyuan]

sorry i don't give the complete K1+K2+K3 data, so there is no answer, but just regress y on x can got it, in other words, i don't really care beta, i just want to use all the y and x values which can be used for id1=A and id2=M, because i have a macro to help me regressing

[Xiyuan]

well i got it
i find my computer allows this large dataset after merging 1 and 2,
i just need to solve how to execute the macro grouped by id1-id2

[PaigeMiller]

I'm not following the need for a macro. If you do the merge, then you can do all the regressions with a BY statement.