s*******e
发帖数: 226 | 1 panel data of firms over years.
How to use STATA and SAS to program models with errors of AR(1) and
clustering by firms? |
l***a
发帖数: 12410 | 2 to mix multiple firms into a single ar(1) model? is that a good idea?
【在 s*******e 的大作中提到】
: panel data of firms over years.
: How to use STATA and SAS to program models with errors of AR(1) and
: clustering by firms?
|
s*r
发帖数: 2757 | 3 proc mixed can estimate different covariance matrices by group |
s*******e
发帖数: 226 | 4 My current program is
proc mixed method=reml empirical noclprint data=beta;
class ID time;
model Y = ID x1 x2 x3/s;
repeated/type=ar(1) subject=ID;
but it does not account for clustering by firms.
Any suggestions?
【在 s*r 的大作中提到】
: proc mixed can estimate different covariance matrices by group
|
s*r
发帖数: 2757 | 5 100 wb?
【在 s*******e 的大作中提到】
: My current program is
: proc mixed method=reml empirical noclprint data=beta;
: class ID time;
: model Y = ID x1 x2 x3/s;
: repeated/type=ar(1) subject=ID;
: but it does not account for clustering by firms.
: Any suggestions?
|
o****o
发帖数: 8077 | 6 you give me some wb...
【在 s*r 的大作中提到】
: 100 wb?
|
c**d
发帖数: 104 | 7
【在 s*******e 的大作中提到】
: panel data of firms over years.
: How to use STATA and SAS to program models with errors of AR(1) and
: clustering by firms?
|
c**d
发帖数: 104 | 8 我想楼主第一要把自己想model的问题想清楚。
（1） 比如你这个panel是什么？应该是firms吧。
（2）比如你想compare mean response over time by firms是你关心的问题
（3)你的ID 应该是单个的subject and clustered by firms
proc mixed
class id time firms;
model y = time firms time*firms x1 x2 x3/s;
repeated /type = ar(1) subject = id(firms);
random id(firms); /* random intercept model */
run;
proc mixed method=reml empirical noclprint data=beta; |
s*******e
发帖数: 226 | 9 my panel is of firms over time.
ID is ID of each firm.
【在 c**d 的大作中提到】
: 我想楼主第一要把自己想model的问题想清楚。
: （1） 比如你这个panel是什么？应该是firms吧。
: （2）比如你想compare mean response over time by firms是你关心的问题
: （3)你的ID 应该是单个的subject and clustered by firms
: proc mixed
: class id time firms;
: model y = time firms time*firms x1 x2 x3/s;
: repeated /type = ar(1) subject = id(firms);
: random id(firms); /* random intercept model */
: run;
|
c******a
发帖数: 725 | 10 You can remove (part of) the first order serial correlation by first differencing data.
Then you apply the fixed effect model to the differenced data using
clustering errors.
In order to fully remove the serial correlation, you may apply Prais-Winsten estimation (using quasi-differenced data). See Wooldridge (2006), chapter 12 for details.
【在 s*******e 的大作中提到】
: panel data of firms over years.
: How to use STATA and SAS to program models with errors of AR(1) and
: clustering by firms?
|
s*******e
发帖数: 226 | 11 It makes sense. Will try. Thank you very much.
Thanks everyone up there for the input. Love this board so much.
differencing data.
Winsten estimation (using quasi-differenced data). See Wooldridge (2006),
chapter 12 for details.
【在 c******a 的大作中提到】
: You can remove (part of) the first order serial correlation by first differencing data.
: Then you apply the fixed effect model to the differenced data using
: clustering errors.
: In order to fully remove the serial correlation, you may apply Prais-Winsten estimation (using quasi-differenced data). See Wooldridge (2006), chapter 12 for details.
|
