Re: st: RE: difficulty in explaining GMM sargan overid

[B B <binta.sarat@yahoo.co.uk>]

Dear Mark,

Thank you very much for saving me from this situation. Its been very frustrating trying to figure out how to get out if this and to be honest its tiring. U perfectly understood my problem and have given me a breather. 

I did wonder why my chi sq was that huge and plus the fact that I have a small sample size, I didnt know what else to do.

Just wanted to clarify, are you suggesting using 2sls? And if yes, from your response it seems like when I use 2sls, I might end up with similar output to OLS, is this correct? Please help clarify this. I'm sorry it sounds really lame,but Im new to most of these things and Im trying to not get confused with them. 

Thanks a bunch

Binta

--- On Fri, 25/6/10, Schaffer, Mark E <M.E.Schaffer@hw.ac.uk> wrote:

> From: Schaffer, Mark E <M.E.Schaffer@hw.ac.uk>
> Subject: st: RE: difficulty in explaining GMM sargan overid
> To: statalist@hsphsun2.harvard.edu
> Date: Friday, 25 June, 2010, 0:58
> Binta,
> 
> If I understand your situation correctly, your Sargan test
> is basically useless, and so would a J stat.  Here's
> why.
> 
> From the output you posted it looks like:
> 
> 1. You have 209 observations.
> 
> 2. You have something like 8 or fewer endogenous regressors
> and a similar small number of exogenous regressors (it's
> hard to read the output because of formatting problems).
> 
> 3. Your Sargan test has 344 degrees of freedom.
> 
> From (2) and (3) it follows that you have something like
> 350 instruments.
> 
> Now consider the fact that you only 209 observations. 
> Say you were doing 2SLS (closely related to the GMM
> estimator you are using, and the intuition is the same).
> 
> In the first stage, you would regress each endogenous
> regressor on the handful of exogenous regressors and the 350
> instruments.  But you have ony 209 observations! 
> Your R-sq will be 100%, and your predicted values of your
> endogenous regressors will  all be ... exactly the same
> as the original values.
> 
> In the second stage of 2SLS you'd be regressing the
> dependent variable on these "predicted values" of the
> endogenous regressors plus the exogenous ones, but since
> they are just the same as the original values, your 2SLS
> would in fact be exactly the same as OLS.
> 
> This is why, I think, your overid test has no power to
> detect violations of the orthogonality conditions.  The
> overid test is, intuitively, a vector of contrast test,
> where the contrasts are between GMM estimators using
> different combinations of instruments.  But in your
> case, all the different combinations are giving you,
> essentially, slightly different versions of OLS.  Since
> the contrasts are all tiny, the chi-sq statistic is
> small.  But this is telling nothing about the validity
> of your instruments.
> 
> What you need to do is to cut down dramatically on the
> number of instruments.  In the context of the
> Arellano-Bond and related estimators, this means cutting
> down the number of lags.  A lot!  There are no
> hard and fast rules for this, but if I were you and I had
> only 209 observations, I'd be looking for a Sargan or J
> statistic with maybe 20 or so degrees of freedom.
> 
> Cheers,
> Mark
> 
> > -----Original Message-----
> > From: owner-statalist@hsphsun2.harvard.edu
> 
> > [mailto:owner-statalist@hsphsun2.harvard.edu]
> On Behalf Of B B
> > Sent: 23 June 2010 18:43
> > To: statalist@hsphsun2.harvard.edu
> > Subject: st: difficulty in explaining GMM sargan
> overid
> > 
> > Dear All,
> > 
> > I am kind of new to the GMM procedure and like a
> newbie, I am 
> > having difficulties understanding the main intution
> behind 
> > it. My main purpose of using GMM is to enable me deal
> with 
> > endogeneity problem which may arise in the analysis I
> intend 
> > to carry out. In my research, I want to examine the
> impact of 
> > financial liberalisation on financial development in
> emerging 
> > countries.
> > 
> > My sample consists of 11 countries over 28 years which
> gives 
> > a total of 308 obs. However, reading through some of
> the 
> > archives, I noticed that my chi2(344) might be too big
> and 
> > probably create a problem. I might be wrong but like
> earlier 
> > stated, I am a novice in this.
> > 
> > My depvar is FD for both bank and stock marketindvar
> includes 
> > lnpcap, bhldate, trade, infl, fdi and institutions. To
> test 
> > the RZ hypothesis I have included the interactions
> between FO 
> > and TO. My model is similar to that of Baltagi et al
> (2007) 
> > and Ito (2006). From what I understand, you would have
> to 
> > include the lag dependent variable and lag of the
> indvar as 
> > instruments in the GMM estimation, correct me if Im
> wrong.
> > My main problem now is, using the xtabond command in
> stata 9, 
> > I obtained the following:
> > 
> > Arellano-Bond dynamic panel-data estimation    
> Number of obs 
> >      =    209Group variable (i): cty
>                      
> >    Number of groups   =    11
> > Wald chi2(7)       =    1008.11
> > Time variable (t): year                    
>     Obs per 
> > group: min =    11avg
> =    19max =    23
> > One-step results
> >     D.m3wdi       Coef.   Std. Err.
>      z    P>z     [95% 
> > Conf.    Interval]   
> m3wdi LD.    .8884923    .047715    
> > 18.62   0.000     .7949727   
> .9820119bhldate D1.    1.453598 
> >   1.312559     1.11   0.268  
>  -1.118971    
> > 4.026166lnpcapwdi D1.    2.620653   3.494215    
> 0.75   0.453 
> >    -4.227882    9.469188trade D1.  
>  .0624551   .0328946     
> > 1.90   0.058    -.0020171   
> .1269274inf D1.   -.0914649   
> > .0278294    -3.29   0.001  
>  -.1460095    -.0369202fdi 
> > D1.    .2869984   .2403093     1.19   0.232  
>  -.1839991    
> > .757996icrgqog D1.   -9.449567   4.480311  
>  -2.11   0.035    
> > -18.23082    -.6683196_cons  
>  -.101707   .1329192    -0.77  
> >  0.444    -.3622238   
> .1588098    
> > 
> > Sargan test of over-identifying restrictions:  
>   chi2(344) = 
> >   193.65     Prob > chi2 = 1.0000
> > 
> > Arellano-Bond test that average autocovariance in
> residuals 
> > of order    1 is 0:H0: no
> autocorrelation   z =  -7.08   Pr 
> > > z = 0.0000
> > 
> > Arellano-Bond test that average autocovariance in
> residuals 
> > of order    2 is 0:H0: no
> autocorrelation   z =   0.56   Pr 
> > > z = 0.577538   
> .1588098    
> > 
> > From my understanding of the sargan test, the
> chi2(344) = 
> > 1.0000 should mean that I cannot reject the
> overidentifying 
> > restrictions. However, like I stated earlier,
> according to 
> > the archives, my chi2(344) might be too large, but I
> dont 
> > think I understand this reason, I am confused or maybe
> 
> > confusing myself
> > I indeed will appreciate any help to clarify this.
> > 
> > Thanks
> > Binta
> > 
> > 
> > 
> >       
> > 
> > *
> > *   For searches and help try:
> > *   http://www.stata.com/help.cgi?search
> > *   http://www.stata.com/support/statalist/faq
> > *   http://www.ats.ucla.edu/stat/stata/
> > 
> 
> 
> -- 
> Heriot-Watt University is a Scottish charity
> registered under charity number SC000278.
> 
> 
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
> 


      

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/