1# saudada
I don't know whether I get your point?
It seems like cluster data, 100 companies as 100 cluster.
for data in cluster: following methods could be used
1.gee: [xtgee ycont x1 x2 x3 x4, id(id) robust] for example, ps. but it not support stepwise
2.robust standard error: [sw, pr( ): regress y x1 x2 x3 x4, vce(cluster id)] for example
3.randome effects model