m*********g 发帖数: 6 | 1 I am doing this break point tests with inflation data.
I need to estimate the multiple unknown break point in a MZ regression, with
serial correlated errors.
I use the single point test statistics in Bai and Perron (1998) , also
Andrews' (1998) econometrica papers.
But it seems quite strange, that the size of those test statistics under the
null( no break point) is quite large. I did some simulation, at 5% critical
value, the rejection rate from simulation could be as high as 30% in small
sample | z****j 发帖数: 131 | 2 Not surprising to me.
test statistics is big means that the null hypothsis is false!
here it means "no break points" hypothesis is false,
which means "there are break points".
I am a biostatistician and read a "two phase linear regression model"
paper in stat journal recent days. hehe.
with
the
critical
small
help
【在 m*********g 的大作中提到】![](/moin_static193/solenoid/img/up.png) : I am doing this break point tests with inflation data. : I need to estimate the multiple unknown break point in a MZ regression, with : serial correlated errors. : I use the single point test statistics in Bai and Perron (1998) , also : Andrews' (1998) econometrica papers. : But it seems quite strange, that the size of those test statistics under the : null( no break point) is quite large. I did some simulation, at 5% critical : value, the rejection rate from simulation could be as high as 30% in small : sample
|
|