11. since covariance is not close to zero as t changes to infinite, it is not weakly dependent. so the autocorrelation issue is at stake

durbin -watson test is centered around 2, by which you can judge whether it is or not

12, heteroskedascity issue. when it is unknown, robust way to do.

feasible GLS is to know what variance looks like and then transform the original equation into newly variance-constant one, so that better efficiency.

read the book from woodridge, it is pretty easy to do this. more advanced topic is not touched here.