Re: st: Cross-section regression with fixed effects

[Austin Nichols <austinnichols@gmail.com>]

Cecilia Dassatti <cecilia.dassatti@gmail.com>
That should work fine, as would:

xtset deudor
xtreg depvar indepvars, fe vce(cluster bank)

and the VCE will be adjusted automatically for the #clusters in bank.

But you are apparently selecting on the depvar:
"change in the log of strictly positive loans"
which is not allowed.  You should be including zeros in loans, and
negative values if they exist in the data.

You can still work on the log scale if you change your
assumptions--see page 3 of
http://www.stata.com/meeting/boston10/boston10_nichols.pdf
To get to the change in the log, you can model log(y) at t as a
function of log(y) at t-1, i.e. put the lag on the RHS (if it has a
coef of one, then this is the same as the regression with the
difference in logs as the outcome). Then you are allowed to select on
the RHS, so you can run a GLM of y on the log of the lag, dropping
cases where the lag is zero or negative.

On Mon, Sep 9, 2013 at 12:39 PM, Cecilia Dassatti
<cecilia.dassatti@gmail.com> wrote:
> i want to include firm fixed effects, and i that that the syntax would be
> xtreg depvar indepvars, i(deudor) fe vce(cluster bank)
>
> where deudor==firm
>
>> Am 08.09.2013 21:24, schrieb Cecilia Dassatti:
>>
>>> Hi everybody,
>>>
>>> I am working with a panel dataset in which the id is given by pairs of
>>> bank-firm relationships and the time variable are months.
>>>
>>> My LHS variable is the change in the log of strictly positive loans
>>> given by one bank to a firm in t+1, while my RHS variable of interest
>>> is the liquidity of the bank in t-1 (there's a change of policy in t).
>>> I want to include firm fixed effects. I tried first with xtreg, but
>>> since my panel id are bank-firm pairs, I can't find the way for
>>>
>>> xtreg depvar indepvars, i(deudor) fe vce(cluster bank)
>>>
>>> to work.
>>>
>>> I thought about trying with areg o manually demeaning all my variables
>>> at the firm level, but then I saw that i need to do a degrees of
>>> freedom adjustment since I am using clusters for the standar errors. I
>>> can do these other options, but i wanted to know if I am doing
>>> something wrong with xtreg.
>>>
>>> Thanks!
>>>
>>> Cecilia
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/