m********5
发帖数: 619 | 1 First stage
x1=z1 z2 (z1 z2 are instrumental variables)
Second stage
y=x1 x2
both estimated using OLS
第一个问题，我用SAS proc model也好，Stata ivreg也好,都会把x2也算成
explanatory variables for first stage。如果我直接用predicted value of x1来
run second stage regression，就要自己调Murphy and Topel correction. 有没有有
经验的啊
第二个问题是如果first stage estimated using OLS and second stage estimated
using probit
同样的问题，stata, ivprobit一样会把x2算成first stage里面。
第三个问题，如果distribution本身是censored, 比如volatility是>=0，这算
censored
data，需要用tobit regression |
t****g
发帖数: 715 | 2 First stage
x1=z1 z2 (z1 z2 are instrumental variables)
Second stage
y=x1 x2
both estimated using OLS
第一个问题，我用SAS proc model也好，Stata ivreg也好,都会把x2也算成
explanatory variables for first stage。如果我直接用predicted value of x1来
run second stage regression，就要自己调Murphy and Topel correction. 有没有有
经验的啊
ivreg y x2 (x1=z1,z2), here you go. No need to manually fix anything.
第二个问题是如果first stage estimated using OLS and second stage estimated
using probit
同样的问题，stata, ivprobit一样会把x2算成first stage里面。
Can not understan |
m********5
发帖数: 619 | 3 如果你用ivreg y x2 (x1=z1, z2), first
然后出来得会是
first stage
x1=z1 z2 x2
second stage
y=x1 x2
你可以随便找个dataset看看是不是这个情况
【在 t****g 的大作中提到】
: First stage
: x1=z1 z2 (z1 z2 are instrumental variables)
: Second stage
: y=x1 x2
: both estimated using OLS
: 第一个问题，我用SAS proc model也好，Stata ivreg也好,都会把x2也算成
: explanatory variables for first stage。如果我直接用predicted value of x1来
: run second stage regression，就要自己调Murphy and Topel correction. 有没有有
: 经验的啊
: ivreg y x2 (x1=z1,z2), here you go. No need to manually fix anything.
|
x********4
发帖数: 405 | 4 it is ok... in the first stage, x2 is the instrument for itself... it does
not bise your coefficient of x1 in the second stage... actually, you should
include all exogenous variables x2, x3.... x119 in your 1st stage. exclusion
restriction only asks z1 and z2 not be in the second stage...
【在 m********5 的大作中提到】
: 如果你用ivreg y x2 (x1=z1, z2), first
: 然后出来得会是
: first stage
: x1=z1 z2 x2
: second stage
: y=x1 x2
: 你可以随便找个dataset看看是不是这个情况
|
m********5
发帖数: 619 | 5 那做table的时候是报还是不报coefficients on x2 in the first stage
报的话，没有理论支持
不报的话，我觉得好像在cheating....
should
exclusion
【在 x********4 的大作中提到】
: it is ok... in the first stage, x2 is the instrument for itself... it does
: not bise your coefficient of x1 in the second stage... actually, you should
: include all exogenous variables x2, x3.... x119 in your 1st stage. exclusion
: restriction only asks z1 and z2 not be in the second stage...
|
t****g
发帖数: 715 | 6 Yes, you are right. But the way I said above is the correct way to do this.
If you do not see why, open your 1st year metrics notes.
【在 m********5 的大作中提到】
: 如果你用ivreg y x2 (x1=z1, z2), first
: 然后出来得会是
: first stage
: x1=z1 z2 x2
: second stage
: y=x1 x2
: 你可以随便找个dataset看看是不是这个情况
|
m********5
发帖数: 619 | 7 对了
xiaotian你怎么看第三个censored data问题
volatility我还可以理解用tobit model来算，如果做dependent variable的话
如果是unsigned forecast error呢
forecast error = abs (analyst forecast - actual reported earnings)
也是bounded by 0
我从来没见人用tobit model, dependent variable or independent variable
should
exclusion
【在 x********4 的大作中提到】
: it is ok... in the first stage, x2 is the instrument for itself... it does
: not bise your coefficient of x1 in the second stage... actually, you should
: include all exogenous variables x2, x3.... x119 in your 1st stage. exclusion
: restriction only asks z1 and z2 not be in the second stage...
|
m********5
发帖数: 619 | 8 恩，我基础不好
多教育
那你们一般报结果是报还是不报
【在 t****g 的大作中提到】
: Yes, you are right. But the way I said above is the correct way to do this.
: If you do not see why, open your 1st year metrics notes.
|
x********4
发帖数: 405 | 9 i report all coefficient estimates in the first stage... you only need to
make sure
1. coefficients of z1 and z2 are significant and have the right sign. (
consistent with your instrument construction)
2. test joint significance for weak instrument concerns. (see stock and yogo
2005)
2.5 do hausman test and convince people that if iv is not used, endogeneity
will bias the coefficient.
3. if you wish, since you system is over-identified, you may want to do run
sargan's overidentification test and
【在 m********5 的大作中提到】
: 那做table的时候是报还是不报coefficients on x2 in the first stage
: 报的话，没有理论支持
: 不报的话，我觉得好像在cheating....
:
: should
: exclusion
|
m********5
发帖数: 619 | 10 恩谢谢
email回信箱慢慢看
yogo
endogeneity
run
【在 x********4 的大作中提到】
: i report all coefficient estimates in the first stage... you only need to
: make sure
: 1. coefficients of z1 and z2 are significant and have the right sign. (
: consistent with your instrument construction)
: 2. test joint significance for weak instrument concerns. (see stock and yogo
: 2005)
: 2.5 do hausman test and convince people that if iv is not used, endogeneity
: will bias the coefficient.
: 3. if you wish, since you system is over-identified, you may want to do run
: sargan's overidentification test and
|
|
|
t****g
发帖数: 715 | 11 报的话，没有理论支持...
This is a standard textbook problem in 1st year. No need to report x2 in the
table, but a F test on other coefficients.
However, 2SLS is now a bad method to use. It has been proved that 2SLS is
outperformed by several other methods. We can give up 2SLS from now on. If
you are still using it, make sure your F >10.
【在 m********5 的大作中提到】
: 那做table的时候是报还是不报coefficients on x2 in the first stage
: 报的话，没有理论支持
: 不报的话，我觉得好像在cheating....
:
: should
: exclusion
|
x********4
发帖数: 405 | 12 i am not sure for this point... if i were you, i do both (ols and tobit) for
robustness...
【在 m********5 的大作中提到】
: 对了
: xiaotian你怎么看第三个censored data问题
: volatility我还可以理解用tobit model来算，如果做dependent variable的话
: 如果是unsigned forecast error呢
: forecast error = abs (analyst forecast - actual reported earnings)
: 也是bounded by 0
: 我从来没见人用tobit model, dependent variable or independent variable
:
: should
: exclusion
|
x********4
发帖数: 405 | 13 ke qi le...
i report and have a separate table for the 1st stage regression to check
validity of my IVs. this is because iv is a big deal in my paper. If it is a
small part of your paper, no need. just coefficeints of z1 and z2 are fine.
..
【在 m********5 的大作中提到】
: 恩，我基础不好
: 多教育
: 那你们一般报结果是报还是不报
|
m********5
发帖数: 619 | 14 en
just a robustness check
but worry that people might ask
a
fine.
【在 x********4 的大作中提到】
: ke qi le...
: i report and have a separate table for the 1st stage regression to check
: validity of my IVs. this is because iv is a big deal in my paper. If it is a
: small part of your paper, no need. just coefficeints of z1 and z2 are fine.
: ..
|
i*******e
发帖数: 349 | 15 请问一下有哪些方法比2SLS好，是finite sample还是large sample的比较？另外能告
知是哪篇文章就更
好。谢谢～～
the
【在 t****g 的大作中提到】
: 报的话，没有理论支持...
: This is a standard textbook problem in 1st year. No need to report x2 in the
: table, but a F test on other coefficients.
: However, 2SLS is now a bad method to use. It has been proved that 2SLS is
: outperformed by several other methods. We can give up 2SLS from now on. If
: you are still using it, make sure your F >10.
|
t****g
发帖数: 715 | 16 In case of strong intruments, for example, Ander-Rubin test can do as well a
s 2SLS; while in case of weak instruments, AR, KLM,JKLM,CLR outperforms 2SLS
, as 2SLS simply does not work under this circumstance. There is a nice surv
ey paper by Stock, Wright and Yogo in JBES 2002.
【在 i*******e 的大作中提到】
: 请问一下有哪些方法比2SLS好，是finite sample还是large sample的比较？另外能告
: 知是哪篇文章就更
: 好。谢谢～～
:
: the
|
v******a
发帖数: 45075 | 17 用过RATS没有？有人用过吗？ anybody??? where to download @_@
【在 t****g 的大作中提到】
: First stage
: x1=z1 z2 (z1 z2 are instrumental variables)
: Second stage
: y=x1 x2
: both estimated using OLS
: 第一个问题，我用SAS proc model也好，Stata ivreg也好,都会把x2也算成
: explanatory variables for first stage。如果我直接用predicted value of x1来
: run second stage regression，就要自己调Murphy and Topel correction. 有没有有
: 经验的啊
: ivreg y x2 (x1=z1,z2), here you go. No need to manually fix anything.
|
t****g
发帖数: 715 | 18 No idea of RATS, as I use matlab even with time series data. You can purchas
e RATS on line, but I think you are looking for a free version...
【在 v******a 的大作中提到】
: 用过RATS没有？有人用过吗？ anybody??? where to download @_@
|
i*******e
发帖数: 349 | 19 Thanks a lot! It's good to know.
a
2SLS
surv
【在 t****g 的大作中提到】
: In case of strong intruments, for example, Ander-Rubin test can do as well a
: s 2SLS; while in case of weak instruments, AR, KLM,JKLM,CLR outperforms 2SLS
: , as 2SLS simply does not work under this circumstance. There is a nice surv
: ey paper by Stock, Wright and Yogo in JBES 2002.
|
m********5
发帖数: 619 | 20 赞
trying搞计量的？
a
2SLS
surv
【在 t****g 的大作中提到】
: In case of strong intruments, for example, Ander-Rubin test can do as well a
: s 2SLS; while in case of weak instruments, AR, KLM,JKLM,CLR outperforms 2SLS
: , as 2SLS simply does not work under this circumstance. There is a nice surv
: ey paper by Stock, Wright and Yogo in JBES 2002.
|
v******a
发帖数: 45075 | 21 of course free version, never ever pay for any software ^o^
purchas
【在 t****g 的大作中提到】
: No idea of RATS, as I use matlab even with time series data. You can purchas
: e RATS on line, but I think you are looking for a free version...
|
o****o
发帖数: 8077 | 22 if both OLS, this is a generalized Berkson Error model, you can fit it as a
Mixed Model in SAS using ML since the likelihood function is easy to write.
If ML is available, don't even bother to use IV.
【在 m********5 的大作中提到】
: First stage
: x1=z1 z2 (z1 z2 are instrumental variables)
: Second stage
: y=x1 x2
: both estimated using OLS
: 第一个问题，我用SAS proc model也好，Stata ivreg也好,都会把x2也算成
: explanatory variables for first stage。如果我直接用predicted value of x1来
: run second stage regression，就要自己调Murphy and Topel correction. 有没有有
: 经验的啊
: 第二个问题是如果first stage estimated using OLS and second stage estimated
|
