[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: xtabond2: Sargan test

From   "Augusto Cadenas" <[email protected]>
To   [email protected]
Subject   Re: st: xtabond2: Sargan test
Date   Mon, 18 Aug 2008 18:32:43 +0200


sorry to reply only with more questions, but...: why don't you include
the lagged dependent variable L.y in your -gmm()- option? After all,
that's the whole point about the dynamic panels literature.

If your tests reject the null, why don't you try the difference GMM
estimator instead? (option -noleveleq-) It could be that the
assumption underlying system GMM, namely that initial values be
uncorrelated to the fixed effects, is violated in your dataset.

And finally: do you have an economic rationale for the suboptions
-eq(diff)- and -eq(level)- you're using there?


On Mon, Aug 18, 2008 at 8:53 AM, sevcan yesiltas
<[email protected]> wrote:
> Dear All,
>  Using DPD, I am trying to  estimate the following equation of the general
> form ,
> yit=a(y)it-1+b(x)it-1+(n)i+(d)t+(v)it  i=1,2,....,N &  t=1,2,...,T
> I have two samples, first one is an unbalanced panel that consists of
> nearly 30.000 observations ( every firm has observations for at least 3
> consecutive years); the second one is a balanced panel that consists of
> approximately 600 firms having information for the whole period. My
> analysis stands for 1994-2006 period.
> In stata 9, via  commands recently updated by David Roodman, I use the
> following command;
> xi:xtabond2 y L.y L.x i.year,gmm(L.x,eq(diff)) iv(i.year,eq(level)) small
> robust two ( since my regressors are endogeneous and correlated with
> firm-specific effects) I run may commands in order to find valid
> instruments.( using only second lags, using older lags, collapsing all
> instruments). But none of my estimation results (for both samples) have a
> p-value grater than 0.1 for
> sargan and hansen test, indeed p values are always 0.000. In short, I am
> not able to get the wanted p values.
> Is it because of maybe persistent data I am working with, or
> multicollinearity between the regressors?
> If so, anyone knows a panel unit root test works with unbalanced data in
> stata?(levinlin, ipshin,madfuller, all work with balanced panel)
> if anyone helps, I strongly appreciate it!
> Best regards
> Sevcan
> --
> Sy
> *
> *   For searches and help try:
> *
> *
> *
*   For searches and help try:

© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index