Re: st: How to estimate Fixed and random effects for a long panel dataset.

[Herman Haugland <herman.haugland@gmail.com>]

Thank you for answering James!


I am trying to estimate beta(equity) from data from 5 banks.  My model
looks like this:

Just in case it helps, my model looks something like this:

beta = L.leverage L.ROA L.RWAA  i.year  D(ummy) L.LeverageD L.ROAD L.RWAAD

* Given my data, 4 of the banks have very similar characteristics,
therefore the Dummy asks: Is this bank A (0) or bank B (1)?
* RWAA = Risk-weighted assets / Average total assets

I have 5 banks, 61 quarters, and 16-1=15 year-dummies.

The original study, on which I am basing my methodology, has a model like this:

beta =  L.leverage  // The authors controlled for additional
variables, but found them not significant.

* 6 banks, 38 half-years, 19-1=18 year-dummies.


As for the clustering problem, is completely related to the fact that
I have a long panel.  In this case, vce(cluster id) just doesn't work.

As for using a VAR, there I have no expertise whatsoever.  Would you
say that it's the only, or best option given my data?


Thanks.





Med vennlig hilsen / Best regards,

Hernan Aros

________________

Contact Information:

Tel: +47 930 289 69
E-mail: herman.haugland@gmail.com
LinkedIn: /in/hermanhaugland


On Sat, Aug 3, 2013 at 8:43 PM, James Bailey <feanor1600@gmail.com> wrote:
> Herman,
>
> First, I wonder what controls you are using- this could be the true cause
> of the degrees of freedom issue.
>
> If it really is a clustering issue, then Stata may be trying to stop you
> for your own good. Mostly Harmless Econometrics section 8.2.3 discusses how
> clustered standard errors are biased if there are fewer than 42 clusters,
> and you are way under 42.
> They discuss possible solutions, one of which is bootstrapped errors.
>
> Finally, with such a large T and small N, it may make sense to abandon
> panel models entirely and use a time series technique like Vector
> Autoregression instead.
>
> Best,
> James Bailey
> Temple University Department of Economics
>
> On Sat, Aug 3, 2013 at 5:27 AM, Herman Haugland
> <herman.haugland@gmail.com> wrote:
>> Dear all,
>>
>> I think I have sent this e-mail before, but I don't know if it made it
>> through Majordomo.
>>
>>
>> I have a long panel dataset, meaning my N is much smaller than my T.
>> I have N = 5, T = 61. I am trying to perform OLS, Fixed-effects and
>> Random-effects analysis, using vce(cluster id).
>>
>> I tried to estimate my model using xtreg for FE and RE, but I get an
>> error related to the fact that I do not have enough degrees of freedom
>> for performing the estimation.
>>
>> This is what I get:
>>
>> Wald chi2(4)       =        .
>> Prob > chi2        =        .
>>
>>
>> Stata sends me here for help:  help j_robustsingular  // My case is
>> explained under the title "Are you using a svy estimator or did you
>> specify the vce(cluster clustvar) option?"
>>
>> So, after reading that, I have assumed that I cannot trust the output
>> of that estimation, because the errors might be biased.
>>
>> First question: Am I right on thinking that?
>>
>> In addition, in the book "Microeconometrics using Stata", the author
>> clearly indicates that the xtreg command, with the vce(cluster id)
>> option for calculating robust errors, is mostly appropriate for short
>> panels, which is not my case.
>>
>> An alternative is to use the command xtregar for estimating random and
>> fixed effects, which is based on an AR(1) process for the errors.
>> However, I tested using xtserial, and the errors do not show serial
>> autocorrelation. However, the xtregar command has the option rhof(#),
>> where # indicates the desired rho value (AR(rho)).
>>
>> My main questions are:
>>
>> 1) What is the right way to calculate Fixed and Random Effects for a
>> long panel dataset, in which the number of variables is larger than
>> the N?
>>
>> 2) Would specifying rho = 0 completely eliminate the AR(1) process for
>> the errors, and leave me with an estimation that fits my data?
>>
>>
>> Thank you for your answers.
>>
>> Best regards,
>> Herman Haugland
>> *
>> *   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/
> *
> *   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/
*
*   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/