xtabond2 test endogenous variable

[Meng Song]

Dear all,
I'm a bit desperate here but could anyone help, please? I want to test whether or not a variable should be treated as endogenous or not when using xtabond2. In both situations when this variable is treated as exogenous in model 1 and endogenous in model 2, AR(1) is significant while AR(2) is insignificant; and Sargan and Hansen tests in both models show instruments are valid. Here is how I tested the endogeneity of this variable, could anyone tell me if this is right or wrong and correct me, please? Thank you very much.

/* Model 1: treat X1 as exogenous */
xi: xtabond2 y l.y X1 X2 X3, gmm(y, lag(5 6)) iv(X1 X2 X3 ) twostep nol nom artests(3)
/* obtain Hansen test of overid restrictions chi2() statistics: S0 */

/* Model 2: treat X1 as endogenous */
xi: xtabond2 y l.y X1 X2 X3, gmm(y X1, lag(5 6)) iv(X2 X3 ) twostep nol nom artests(3)
/* obtain Hansen test of overid restrictions chi2() statistics: S1 */
The endogeneity test is a test of S0-S1 (or should this be S1-S0 I'm not entirely sure); The test statistic (S0-S1 or S1-S0) should be distributed as chi-square in the number of regressors being tested for endogeneity, so I'm testing 1 variable as endogenous or not, the dof should be 1. Does this method looks correct? Many thanks. Best wishes, Meng

[Meng Song]

Any help, please?

[daejung yang]

Dear Meng Song
I have same question.
Do you have any solution about endogeous test with xtabond2 ?
Thanks in advance.

[Sebastian Kripfganz]

You might find my 2019 London Stata Conference presentation slides helpful:

• Kripfganz, S. (2019). Generalized method of moments estimation of linear dynamic panel data models. Proceedings of the 2019 London Stata Conference.

Slides 90 onwards illustrate an approach for the classification of variables.

[daejung yang]

Dear professor
thank you .
I have studied your paper. And it is very helpful to me.
I have one more question about estimate of lagged dependent. Bond(2001) paper wrote that
ols estimate have upward bias and fe estimate have downward bias. In my case, difference gmm estimate have below fe estimate. So i choose systemm gmm. Do you think systemm gmm estimate should be between ols and fe estimate.
Thanks in advance.

[Sebastian Kripfganz]

In models with multiple endogenous or predetermined regressors, these simple rules of thumb often do not apply any more. I would not use them as the basis for any modeling decisions.

[daejung yang]

[QUOTE=Sebastian Kripfganz;n1551266]In models with multiple endogenous or predetermined regressors, these simple rules of thumb often do not apply any more. I would not use them as the basis for any modeling decisions.[/QUOT

[daejung yang]

THANK YOU FOR YOUR PRECIOUS ADVICE.
AND I ESTIMATE MY SYSTEM GMM AFTER OMMITTNG CONSTANT TERM AGAIN. THE RESULT IS SAME AS THUMB OF RULE OF BOND. IN MY OPINION , THIS MEANS THE THUMEB OF RULE SHOULD BE USED WITH VERY CAREFULL ATTENTION such as percise simulation model
. Thanks a lot.

In models with multiple endogenous or predetermined regressors, these simple rules of thumb often do not apply any more. I would not use them as the basis for any modeling decisions.