Weighted Least Squares in practice - feasible GLS - part 2

[@vilandes]

Thanks Ben. It would be good to have an example with actual numbers.

[@semobrk700]

Thanks its the first time your video worked

[@m.hamzarahid2293]

Thank you for the great explanation. What exactly would you call a large sample? 100 data points? 1000 data points?

[@sajjadsrg4470]

Thanks dear Ben. I like to know, Can i use this method for panel data?

[@NnRNoAh]

Thanks a lot! I'm currently using non-linear sure, so I watched both your SUR estimation video and the one about FGLS - NLSUR uses FGNLS estimator. Although I was exposed to these tools in graduate courses I needed to refresh my mind about them. There is something that is not clear to me yet: You do not discuss the iterative process. I might be mistaken, but I remember that Feasible-GLS needs iterations when estimating the weighted matrix. In fact, it seems to me that STATA's NLSUR does iterate. Is the iteration process needed?

[@rebeccaramos7813]

Thanks Ben 🙏

[@mariosum1065]

Thank you very much for the video. Just to be sure: "v)", if you will, were to run OLS on the as-in-iv)-transformed model, or am I mistaken?

[@lastua8562]

what do you mean?

[@mariosum1065]

​@@lastua8562 I mean, that we take we ... obtain our estimates for the weights (iii), apply those weights to our dependent variable y and independent variables x (iv), and then, we take this transformed data and run just ordinary OLS [sic!] on it.

[@Elekko]

Hey, from what I understand so the regression of residuals log(u^2), we use the same observed data as for the ordinary regression right?

[@lastua8562]

yes, we use u^2 obtained from i.

[@i-fanlin568]

That is very helpful!!!!

[@keyulinghu4571]

thanks man! very helpful!

[@elizabethwomack6551]

Hello, I think the first equation needs a hat over the y?

[@lastua8562]

quite interestingly he does not do that often, seemingly consciously. You do get predicted values from the regression, so a y hat would be the "correct" version I think. So I suppose he does this to make clear that we use the actual data points in y before obtaining the predicted values. In fact, when using a statistical program, we use y (not hat) as the dependent variable).