[問題] OLS 與 GLS

看板Statistics作者 (飛~)時間15年前 (2009/05/14 12:15), 編輯推噓1(1推 0噓 8→)
留言9則, 1人參與, 最新討論串1/1
我有個問題一直困擾很久，就是OLS跟GLS的差別在哪裡呢？我是知道當用OLS估計的結果 如果不具效率的話要改用GLS啦，但除此之外，有沒有分線性模型只能用OLS估計，非線 性模型要用GLS這樣的分別呢？ -- ※ 發信站: 批踢踢實業坊(ptt.cc) ◆ From: 118.169.230.211
推
07/07 13:55, , 1F
simply put, one could apply GLS while the problem of
07/07 13:55, 1F
→
07/07 13:55, , 2F
heteoskedastic errors occur, so that we want to find
07/07 13:55, 2F
→
07/07 13:56, , 3F
a nonsingular and squared matrix to transform the data
07/07 13:56, 3F
→
07/07 13:56, , 4F
However, GLS is often unable to applied because we do
07/07 13:56, 4F
→
07/07 13:57, , 5F
not know what Var(y) is (so does the transformed M).
07/07 13:57, 5F
→
07/07 13:57, , 6F
Hence, we use FGLS instead. But another problem is
07/07 13:57, 6F
→
07/07 13:58, , 7F
that FGLS often doesn't have the nice properties in
07/07 13:58, 7F
→
07/07 13:58, , 8F
finite sample and could be a less efficient estimator
07/07 13:58, 8F
→
07/07 13:58, , 9F
than OLS.
07/07 13:58, 9F
‣ 更多 candypinky 的文章
文章代碼(AID): #1A2vhD5s (Statistics)