s*********e
发帖数: 1051 | |
A*******s
发帖数: 3942 | 2 赞。
Insurance company比较喜欢搞这类的double GLM。
很少见到银行的应用
也许dispersion可以用来搞stress test
【在 s*********e 的大作中提到】
: Jorgensen's dispersion models (1987)
: http://statcompute.wordpress.com/2013/07/14/dispersion-models/
|
s*********e
发帖数: 1051 | 3 it has nothing to do with double GLM or not.
from theoretical standpoint, all exponential error GLM is just a special
case of dispersion GLM and can be represented in a general formula form as
shown in Song's paper.
in other words, there are some models fallen outside of classic GLM based on
exponential errors, such as simplex model (which is a very good candidate
models for unity outcomes).
【在 A*******s 的大作中提到】
: 赞。
: Insurance company比较喜欢搞这类的double GLM。
: 很少见到银行的应用
: 也许dispersion可以用来搞stress test
|
A*******s
发帖数: 3942 | 4 i think it is just about the naming.
Double GLM is nothing but modeling the mean and dispersion simultaneously
via traditional GLM framework and this terminology is widely accepted in
actuarial models nowadays.
you may refer to an much ealier paper of Smyth 1988 GLM with varying
dispersion.
on
【在 s*********e 的大作中提到】
: it has nothing to do with double GLM or not.
: from theoretical standpoint, all exponential error GLM is just a special
: case of dispersion GLM and can be represented in a general formula form as
: shown in Song's paper.
: in other words, there are some models fallen outside of classic GLM based on
: exponential errors, such as simplex model (which is a very good candidate
: models for unity outcomes).
|
c****t
发帖数: 19049 | |
