Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: RE: Forward orthogonal deviations with unbalanced panel data


From   Rodolphe Desbordes <rodolphe.desbordes@strath.ac.uk>
To   "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu>
Subject   st: RE: Forward orthogonal deviations with unbalanced panel data
Date   Mon, 8 Jun 2009 19:15:28 +0100

Dear all,

I have found the problem. I now get the same estimates and standard errors. My very crude code is below. However, computations are quite slow...


webuse grunfeld, clear
egen i=group(company)
egen t=group(year)
tsset i t

forvalues v=1(25)200 {
replace kstock=. in `v'
}

capture drop orth*
qui foreach var of varlist  invest mvalue kstock  {
capture drop v
gen v=(invest<.&mvalue<.&kstock<.)
capture drop i
capture drop t
egen i=group(company)
egen t=group(year)
tsset i t
gen orth`var'=.
forvalues x =1/10{
capture drop qs
capture drop e2*
capture drop e3*
egen double e2=max(t) if i==`x'
egen double e3=min(t) if i==`x'
egen double e22=mean(e2)
egen double e33=mean(e3)
replace e2=e22
replace e3=e33
local a=e3
while `a'<e2 {
capture drop sqrt*
capture drop w*
capture drop dev*
capture drop fin*
capture drop ss
capture drop ss2
capture drop `var'x
capture drop ds
capture gen double `var'x=f1.`var'  if i==`x'  &f1.v==1
capture egen double w`var'=total(`var'x) if t>=(`a') &  i==`x'
capture egen ss2=count(`var'x) if t>=(`a')  &  i==`x'
capture gen double w`var'2=w`var'/ss2
capture gen double dev`var'=`var'-w`var'2
capture gen double sqrt=sqrt(ss2/(ss2+1))
capture gen double fin`var'`a'=dev`var'*sqrt
capture gen double fin`var'`a'2=l1.fin`var'`a'
capture replace  orth`var'=fin`var'`a'2 if t==(`a'+1) &  i==`x' & v==1
local a=`a'+1
capture drop fin*
}
}
}


. reg  orthinvest orthmvalue orthkstock,nocons

 obs= 176
------------------------------------------------------------------------------
  orthinvest |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
  orthmvalue |   .1137694   .0120434     9.45   0.000     .0899995    .1375393
  orthkstock |   .3126414   .0173844    17.98   0.000       .27833    .3469528
------------------------------------------------------------------------------

. xtabond2  invest mvalue kstock,iv(mvalue kstock)  nolev orth small


obs=176
------------------------------------------------------------------------------
  orthinvest |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
  orthmvalue |   .1137694   .0120434     9.45   0.000     .0899995    .1375393
  orthkstock |   .3126414   .0173844    17.98   0.000       .27833    .3469528
------------------------------------------------------------------------------



Rodolphe

PS: a better title for this thread would have been "Forward orthogonal deviations with gaps in the data".

________________________________________
From: owner-statalist@hsphsun2.harvard.edu [owner-statalist@hsphsun2.harvard.edu] On Behalf Of Rodolphe Desbordes [rodolphe.desbordes@strath.ac.uk]
Sent: 07 June 2009 18:27
To: statalist@hsphsun2.harvard.edu
Subject: st: Forward orthogonal deviations with unbalanced panel data

Dear all,

I am trying to calculate "by hand" forward orthogonal deviations. This transformation consists in subtracting the mean of the remaining future observations available in the
sample from the current observation. A weighting is introduced to equalise the variances. A more detailed explanation can be found at http://www.tinbergen.nl/~buhai/papers/publications/panel_data_econometrics.pdf , p.6. I am not sure how to handle missing values.

A) I have written the following code:

webuse grunfeld, clear
egen i=group(company)
egen t=group(year)
tsset i t
capture drop orth*
qui foreach var of varlist  invest mvalue kstock{
gen orth`var'=.
forvalues x =1/10{
capture drop e2*
capture drop e3*
egen double e2=max(t) if i==`x'
egen double e3=min(t) if i==`x'
egen double e22=mean(e2)
egen double e33=mean(e3)
replace e2=e22
replace e3=e33
local a=e3
while `a'<e2 {
capture drop sqrt*
capture drop w*
capture drop dev*
capture drop fin*
capture drop ss
capture drop ss2
capture drop `var'x
capture gen double `var'x=f1.`var'  if i==`x'
capture egen double w`var'=sum(`var'x) if t>=(`a') &  i==`x'
capture gen double w`var'2=w`var'/(e2-(`a'))
capture gen double dev`var'=`var'-w`var'2
capture gen double sqrt=sqrt((e2-`a')/(e2-`a'+1))
capture gen double fin`var'`a'=dev`var'*sqrt
capture gen double fin`var'`a'2=l1.fin`var'`a'
capture replace  orth`var'=fin`var'`a'2 if fin`var'`a'2<.
local a=`a'+1
capture drop fin*
}
}
}

And I obtain the following results:

. reg  orthinvest orthmvalue orthkstock,nocons

obs 190
------------------------------------------------------------------------------
  orthinvest |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
  orthmvalue |   .1101238   .0118567     9.29   0.000     .0867345    .1335131
  orthkstock |   .3100653   .0173545    17.87   0.000     .2758308    .3442999


The estimates are equivalent to those obtained with the user-written command -xtabond2- :


. xtabond2  invest mvalue kstock,iv(mvalue kstock)  nolev orth

obs 190
-------------+----------------------------------------------------------------
      mvalue |   .1101238   .0117941     9.34   0.000     .0870077    .1332399
      kstock |   .3100653   .0172629    17.96   0.000     .2762306       .3439
------------------------------------------------------------------------------


B) However, with missing values:

forvalues v=1(25)200 {
replace kstock=. in `v'
}

I have tried to replace the following lines

***
capture gen double w`var'2=w`var'/(e2-(`a'))
capture gen double dev`var'=`var'-w`var'2
capture gen double sqrt=sqrt((e2-`a')/(e2-`a'+1))
***

by

***
capture gen ss=(`var'==.) if t>=(`a') &  i==`x'
capture egen ss2=sum(ss) if t>=(`a') &  i==`x'
capture gen double w`var'2=w`var'/(e2-(`a')-ss2)
capture gen double dev`var'=`var'-w`var'2
capture gen double sqrt=sqrt((e2-(`a')-ss2)/(e2-(`a')+1-ss2))
***


But I obtain different results (different number of observations and different estimates):



. reg  orthinvest orthmvalue orthkstock,nocons

obs 182

------------------------------------------------------------------------------
  orthinvest |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
  orthmvalue |   .1139964   .0123395     9.24   0.000     .0896476    .1383452
  orthkstock |   .3100502   .0177373    17.48   0.000     .2750506    .3450499
------------------------------------------------------------------------------

. xtabond2  invest mvalue kstock,iv(mvalue kstock)  nolev orth


obs 176

------------------------------------------------------------------------------
      invest |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
      mvalue |   .1137694   .0119747     9.50   0.000     .0902993    .1372395
      kstock |   .3126414   .0172853    18.09   0.000     .2787627      .34652
------------------------------------------------------------------------------


Obviously, I am doing something wrong. I would greatly appreciate if someone could point me in the right direction.

Yours sincerely,

Rodolphe Desbordes


*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index