Re: st: RE: Panel unit root tests

[Neesha Harnam <neesha.harnam@gmail.com>]

Dear Nick

Thank you for your response. I have added the Stata commands I used below:

a. Is the Fisher-ADF test valid when Statalist generates the message
"Stata could not compute test for panels 6, 12, 15, etc.?" I am using
the following code for the unemp variable:

-xtunitroot fisher unemp, dfuller trend lags(1)-

b. What does it mean when one gets the following error message
"performing unit-root test on first panel using the syntax..." and
returns "error code 2000?"  The command I used precisely (for the
inequality variable) is as follows:

-xtunitroot fisher inequality, dfuller trend demean lags(1)-

c. I understand that demeaning is used when cross-sectional dependence
is thought to occur in the data, but is there any way to test for
cross-sectional dependence? Likewise, is there any way to test for a
time trend, or is it based on visual inspection of plots / empirical
evidence? If not, what is the convention?

d. Which p-value of the Fisher-ADF test is valid for finite panels?
Stata generates p-values for the inverse chi-squared, inverse normal,
inverse logit, and modified inverse chi-squared when using the command
for the gdppc variable: -xtunitroot fisher gdppc, dfuller trend
lags(1)-, and I would like to know which would be appropriate to use
given that I have a finite sample.

e. Is it possible to run Fisher-ADF in Stata using AIC-selected lag
lengths? When running the Im-Pesaran-Shin unit root tests this is
possible using the aic specification as follows: xtunitroot ips gdppc,
trend lags(aic). However, when I attempt to run a similar command for
the Fisher test (xtunitroot fisher gdppc, dfuller trend lags(aic)) I
get error code 198. I believe that using the AIC to determine lag
length for the Fisher test is possible in EViews, and would like to
know how I could do this in Stata.

Thanks again,
Neesha



On Mon, Oct 3, 2011 at 6:11 PM, Nick Cox <n.j.cox@durham.ac.uk> wrote:
> Your thread was hijacked, true, but your question was visible nevertheless. A bigger problem is that there is absolutely no detail here on precisely what commands you used and precisely what you typed in Stata.
>
> Error 2000 does not mean "no data"; it means "no data with which to do precisely what you asked". That could arise because of missing values, -if- or -in- restrictions, string values, or requests which are contradictory and are thus satisfied by no observations in your data. I can't say what is biting in your case.
>
> Nick
> n.j.cox@durham.ac.uk
>
>
> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Neesha Harnam
> Sent: 03 October 2011 11:01
> To: statalist
> Subject: st: Panel unit root tests
>
> Dear all,
>
> Apologies for reposting, but I did not receive any responses to my
> earlier query and thought it may have been because someone hijacked my
> thread with a question of their own. I have also rephrased some of my
> questions in the event that they were not clear initially, as
> suggested by the Statalist FAQ:
>
> I have a set of variables for which I would like to check for
> non-stationarity (these are in panel data form, with 31 years and 70
> countries worth of data). As there are some unbalanced panels in my
> dataset I am using IPS and Fisher (ADF) tests to conduct these checks.
> I have looked at the individual line plots for each country/variable
> to determine if there is a time trend, and have run these tests both
> with and without cross-sectional demeaning. I am planning on running
> country-fixed effects regressions on these data.  My questions are as
> follows:
>
> a. Is the Fisher-ADF test valid when Statalist generates the message
> "Stata could not compute test for panels 6, 12, 15, etc.?"
>
> b. What does it mean when one gets the following error message
> "performing unit-root test on first panel using the syntax..." and
> returns "error code 2000?"  A google search revealed that this occurs
> when one has no observations, but I do have observations so am unclear
> as to what is happening behind the scenes.
>
> c. I understand that demeaning is used when cross-sectional dependence
> is thought to occur in the data, but is there any way to test for
> cross-sectional dependence? Likewise, is there any way to test for a
> time trend, or is it based on visual inspection of plots / empirical
> evidence?
>
> d. Which p-value of the Fisher-ADF test is valid for finite panels?
> Stata generates p-values for the inverse chi-squared, inverse normal,
> inverse logit, and modified inverse chi-squared.
>
> e. Is it possible to run Fisher-ADF in Stata using AIC-selected lag lengths?
>
>
> Thank you very much,
> Neesha
> *
> *   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/
>

*
*   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/