[Limdep Nlogit List] FRONTIER command : Battese & Coelli 95 model

Zhi Shen Z.Shen at lboro.ac.uk
Sat Oct 3 02:10:42 EST 2009


Hi Barry,

I didn't use
create;grpsize=_groupti$
calc; g=Max(grpsize)$ ( this part is option)

I use frontier; lhs=dependent variables; rhs=independent variables;pds=_groupti;... directly. I don't know whether it makes any difference with yours. Worth find out. 

Sorry for my second response. I do want to say (Frontier;........;pds=groupsize;model=bc;rh2=one,t, Zs..) rather than model=t; since I know model=t; works fine but I doubt model=bc; would provide these parameter estimates. Sorry for confusion caused.

Zhi

-----Original Message-----
From: limdep-bounces at limdep.itls.usyd.edu.au [mailto:limdep-bounces at limdep.itls.usyd.edu.au] On Behalf Of Barry Quinn
Sent: 02 October 2009 17:03
To: Limdep and Nlogit Mailing List
Subject: Re: [Limdep Nlogit List] FRONTIER command : Battese & Coelli 95 model

Hi Shi, 

thanks for reply , lots of good information.

To response to your queries 

Yes i did create an additional variable called group size using a unique firm identifier as follows

regress;lhs=one;rhs=one;str=firm;panel$
create;grpsize=_groupti$
calc; g=Max(grpsize)$ ( this part is option)

However this produced a variable which counted the number of observations of each firm over the period ( not the total number of firms in each cross section as you seem to indicate in your email)  for eg

   FIRM         t               Iit(observations)             groupsize
    A              1               1.2                                       3
    A              2               1.4                                       3
    A              3                -4                                        3
    B              1               4                                           2
    B              2               0.5                                        2
    C              1               0.7                                        3
    C              2               0.1                                       3
    C              3                1.0                                     3


Is this the type of variable that you meant?

And to your second question when i run the model (Frontier;........;pds=groupsize;model=t;rh2=one,t, Zs..)
I do get parameter estimates which are significant... perhaps this may be due to the way i calculated the previous panel variable???

many thanks 


Barry  
________________________________________
From: limdep-bounces at limdep.itls.usyd.edu.au [limdep-bounces at limdep.itls.usyd.edu.au] On Behalf Of Shen Zhi [Z.Shen at lboro.ac.uk]
Sent: Friday, 2 October 2009 3:54 PM
To: Limdep and Nlogit Mailing List
Subject: Re: [Limdep Nlogit List] FRONTIER command : Battese & Coelli   95      model

Hi Barry,

I may not be the right person to answer your first concern but I think
in Limdep, BC(95) model do capture the panel data nature. I've been
advised that Frontier 4.1 programme treats BC(95) model as a pooled
data rather than panel data. This is as far as I know.

To your second concern about the pds command. I don't think it will be
overridden by model=t command. You do have a large data set with 480
firms over 6 year. But I know Professor Greene run a US bank study
with 500 banks over 5 years period. I have a data set with 280 banks
over 8 years (although not that many as yours). So you should be fine
with your data. There is one more thing I would like to check with
you. Did you first  set up your panel data set before you run the
BC(95) model? The command is:

regress;lhs=dependent variables; rhs=independent variables; panel;
str=firms $

where firms are the number of firm identification that is set up in
your data (as in my data, I set the first column for firm ID number).
This will create a new variable named _groupti, which denotes the
group size (as in your case, should be 480). Then you can use
pds=_groupti; in your following panel data estimation. If your data
are unbalanced, pds=_groupti is the only way you can set up your panel
estimation.

Finally, for your first model:
Frontier;........;pds=groupsize;model=t;rh2=one,t, Zs..;
I know Limdep will allow you to run it and you can get results. But
interestingly, in your results, did you find any parameter estimates
for the z variables? I didn't find any with my data. so very curious
to see whether it works for you.

Best wishes,
Zhi


On 2009-10-2, at 下午2:59, Barry Quinn wrote:

> Hi Zhi,
>
> Thanks for your reply that is very helpful.  I am concerned though
> about whether the BC(95) method you mention (model=t; rh2=one,)
> isn't capturing the panel data nature of the data!( as the manual
> alludes - bottom of  Page E33-44 in new manuals)
>
> Limdep does allow me to run these models
>
> Frontier;........;pds=groupsize;model=t;rh2=one,t, Zs..;
> Frontier;........;pds=groupsize;model=bc;rh2=one,t, Zs..;
>
> May concern still is whether or not the pds command is being
> overridden by the model=t command in the first model above??
>
> many thanks
>
>
> Barry
> ________________________________________
> From: limdep-bounces at limdep.itls.usyd.edu.au [limdep-bounces at limdep.itls.usyd.edu.au
> ] On Behalf Of Zhi Shen [Z.Shen at lboro.ac.uk]
> Sent: Friday, 2 October 2009 1:41 PM
> To: Limdep and Nlogit Mailing List
> Subject: Re: [Limdep Nlogit List] FRONTIER command : Battese &
> Coelli 95        model
>
> Dear Barry,
>
> In Limdep, model=bc is the command for modeling Battese and Coelli
> (1992) time varying inefficiency model in which a monotonic time
> trend is specified to inefficiency term, written as u(i,t)=exp(eta(t-
> T))|U(i)|. If you want to use Battese and Coelli (1995) model to
> examine the impact of firm specific factors on mean of inefficiency
> instead, u have use model=t; rh2=one, time trend,... and other
> factors. I don't think the model command for Battese and Coelli
> (1992) model will recognize the command rh2=one, time trend, ..., at
> least not in my sample experience. You can have a try.
>
> To go further, if you also want to consider the possible impact of
> these firm specific factors in the variances of inefficiency
> (heteroscedasticity), you have to use Battese and Coelli (1995)
> model. The basic commands are
>
> frontier;(cost);lhs=dependent variables;rhs=independent variables;
>    ;model=truncation;rh2=one,z variables             ? z variables
> in mean of uit
>    ;het;hfu=one, z variables                        ? z variables in
> variance of inefficiency uit
>    ;pds=_groupti;eff=uibc $
>
> If you want to consider heteroscedasticity in random error term, you
> can add in commands
>    ;hfv=z variables                               ? z variables in
> variance of random error vit
>
> Hope it helps. Good luck!
>
> Zhi Shen
> Research student
> Department of Economics
> Loughborough University
> LE11 3TU
> UK
>
> -----Original Message-----
> From: limdep-bounces at limdep.itls.usyd.edu.au [mailto:limdep-bounces at limdep.itls.usyd.edu.au
> ] On Behalf Of Barry Quinn
> Sent: 02 October 2009 12:24
> To: limdep at limdep.itls.usyd.edu.au
> Subject: [Limdep Nlogit List] FRONTIER command : Battese & Coelli 95
> model
>
> Hi
>
> I am attempting to recreate the Battesse & Coelli 1995 model in
> which they model the Technical inefficiency using firm specific
> factors in a panel data setting.
> To date i have only been using the Ceolli DOS program in which i
> have been including a time trend variable in the Technical
> inefficiency model to control for temporal effects.
> In Limdep  i use the options
> pds=_groupsize ;model=bc ; Rh2= one, time trend ,......and other
> firm specific factors
>
> But my concern is that in the manual the BC model seems to be the
> 1988 model specification which specifically accounts for time
> variation using the eta term
> Also the manual notes that " Battesse and Coelli (1995) ..not
> withstanding the assertion in the paper, the latter is not a panel
> data treatment as observations are still assumed to be independent".
>
> Would this mean that i should not use the Model=BC option at all,
> but instead use model=T and Rh2= one, time trend, ......other firm
> specific factors, and control for the panel using firm dummies ?
>
> My panel is has a large number of firms(480) over a short time
> period(6yrs) so i am concern this amount of dummies my cause
> estimation issues
>
> any advice would be much appreciated
>
> many thanks
>
> Barry Quinn
>
> Queens University Belfast
> _______________________________________________
> Limdep site list
> Limdep at limdep.itls.usyd.edu.au
> http://limdep.itls.usyd.edu.au
>
>
>
> _______________________________________________
> Limdep site list
> Limdep at limdep.itls.usyd.edu.au
> http://limdep.itls.usyd.edu.au
> _______________________________________________
> Limdep site list
> Limdep at limdep.itls.usyd.edu.au
> http://limdep.itls.usyd.edu.au

_______________________________________________
Limdep site list
Limdep at limdep.itls.usyd.edu.au
http://limdep.itls.usyd.edu.au
_______________________________________________
Limdep site list
Limdep at limdep.itls.usyd.edu.au
http://limdep.itls.usyd.edu.au





More information about the Limdep mailing list