谁知道3SLS 用于 PANEL DATA的方法

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

有谁接触过3 STAGE LEAST SQUARE 来分析PANEL DATA的ECONOMETRICS 的方法.

我现在有几个PANEL DATA的变量需要分析,但之间有SIMULTANEOUS的问题存在,而且这些变量又是ENDOGENOUS,所以3SLS分析很好,可惜不能用在PANEL DATA 里面. 现在唯一有的好象是BALTAGI介绍过的EC3SLS,但出来的结果很奇怪. 和我也没搞清楚的GMM,

有谁遇到过类似的问题或者知道有什么解决办法的,希望能给予帮助.

谢谢

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏6 回帖

关键词：panel data Panel Data pane SIMULTANEOUS Data Panel

Dear all,

we want to consider an equation discrete choice system in the following
way,

y1=f(x1, y2) and
y2=f(x2, y1) where

x1 and x2 are vectors of exogenous characteristics and y1 and y2 are
multinomial and conditional logit equations.
After examining several literature, it seems to be necessary to set up a
3SLS regression. Since the discrete choice framework has to be
considered, pure 3SLS in STATA might lead to wrong results, even more it
is not appropriate.

Is there somebody who did overcome that problem?

Thanks,

Stephan

It seems likely that this would be best approached by maximum likelihood estimation using the -ml- command?

-Timothy

[此贴子已经被作者于2006-4-21 4:49:46编辑过]

sureg+IV=3SLS! If you suspect non-homoskedastic error terms and/or
serila correlation, then GMM would be the right thing to do. stata can
do the former, but not the latter.

Pierre

----------------------------------------------------------------------------------
Pierre Azoulay
Associate Professor of Management
Columbia University
Phone: (212) 854-9684
Graduate School of Business
Fax: Don't send any
3022 Broadway, Uris Hall 704
New York, NY 10023

Thanks for pointing this out. Two endogenous variables that I have are not
coming as depvars in any equations in the system. I specify the regression
as:

reg3 (depvar1 varlist1) (depvar2 varlist2) ... (depvarN varlistN),
exog(iv1 iv2) endog(endogvar1 endogvar2) constraints(numlist) sure

will this be a SUR with IV

or should I be specifying -3sls- option instead of -sure- option in the
end. What I want is to estimate the SUR with IV.

Thanking you in advance.

Rijo John.

Hi,

Just to follow up with the earlier mail;

The three models

Model 1:

reg3 (depvar1 varlist1) (depvar2 varlist2) ... (depvarN varlistN),

exog(iv1 iv2) endog(endogvar1 endogvar2) sure

Model 2:

reg3 (depvar1 varlist1) (depvar2 varlist2) ... (depvarN varlistN), sure

Model 3:

sureg (depvar1 varlist1) (depvar2 varlist2) ... (depvarN varlistN)

are giving me the exact same results.

Whereas

Model 4:

reg3 (depvar1 varlist1) (depvar2 varlist2) ... (depvarN varlistN),

exog(iv1 iv2) endog(endogvar1 endogvar2) 3sls

gives a result totally different. Model 1-3 gives results which are very
much similar to simple OLS.

I am now a bit confused whether "sure" option with -reg3- is actually
doing an IV when I estimate it with endog() and exog() options.

Kindly help me with this. Thanking you in advance.

I am using Stata 9 and am estimating a system of simultaneous equations using 3sls. I see it is possible to compute iterative 3sls , however, I am not sure what is to be gained from doing so. I only have an older version of Greene at home (i.e. 1993, second edition) but on page 612 it states:

"It is possible to iterate the 3SLS computation. However, unlike the seemingly unrelated regressions estimator, this does not provide the maximum likelihood estimator, nor does it improve the asymptotic efficiency."

This is all my version of Greene states on iterative 3SLS, therefore, I'm not sure what is to be gained from using it. What are the advantages and disadvantages from using 'iterative 3SLS' over '3SLS'.

Thanks in advance for any assistance.

Kevin

Dear all,

I would very much appreciate some advice on applying the control function approach

I have 3 equations:

1. y1 = dZ + e

where y1 is a dicrete choice variable taking the value 1 if an MBO, 2 if an MBI, and zero is neither MBO or MBI. This is estimated by multinomial logit. Z the independent variable set.

My second equation:

2. wages = eemplyment + bX + aMBO + cMBI + v
3. employment = fW + u
a, b, c, d, e, and f are coefficients.

Equations 2 and 3 are estimated by 3SLS

My questions:
1. Can I use the multinomial logit in a control function approach?
2. If the answer to (1) is yes, can I use the fitted values from the multinomial logit as explanatory variables in equation (2) in order to test for endogeneity and correct the standard errors for bias?
3. If the answer to (2) is no, is it possible to obtain the choice specific residuals from the multinomial logit? Wooldridge (2002) suggests using the residuals but I cannot find a way of obtaining them after multinomial logit estimation.

Any help on this is much appreciated.

Regards,
Kevin

Vadim Grinevich

As you know when you run a constrained regression, the output statistics does not include R squared. However, I do need at least some indication of it in order to compare my results with those from other studies using similar sort of constraints. To sum up, is there any way to get indication of R squared for constrained regression (cnsreg command)

How can I get an R-squared value when a Stata command does not supply one?
http://www.stata.com/support/faqs/stat/rsquared.html