Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

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

Re: st: unit-root test

From   L P <>
Subject   Re: st: unit-root test
Date   Wed, 22 Aug 2012 16:29:58 +0200

Dear Muhammad,

Many thanks for your suggestion. If I can take some more minutes of
your time, I would like to share what follows:

1. as I said I am using STATA 9. For this reason when I typed the
command . ssc install xtunitroot I got the message
ssc install: "xtunitroot" not found at SSC, type -findit xtunitroot-
(To find all packages at SSC that start with x, type -ssc describe x-)

2. as suggested in the error message I typed . findit xtunitroot and
STATA opened a page with the link to four unit-root tests packages,
which I installed all. Among them, I found the XTFISHER test (which is
referred to be suitable for unbalanced data). I have tried to run the
test and it looks like things are working in the right way now. For
example, I run

xtfisher  Ln_MKTopn_2, lag(1)

and the result is

Fisher Test for panel unit root using an augmented Dickey-Fuller test (1 lags)

Ho: unit root

         chi2(60)     =  119.9346
         Prob > chi2  =      0.0000

My questions:

1. what is the right interpretation of this result? I think I have to
reject the null hyothesis because the p-value is < or = to 0.05.
Hence, I can conclude that the variable is stationary and its use in
the specification model is valid. Am I right?

2. what should be the right ammount of lags to be considered in the lag option?

3. does this test must be run for each single independent variable I
consider in my model specification?

4. shall I run the test for the dependent variable as well?

Thanks again for your really helpful support.

2012/8/22 Muhammad Anees <>:
> Hello,
> You can find more information on theoritical part from the references
> given in the helpfile.
> Moreoever, alternative sources of information is available from:
> -xtunitroot- Check from Stata's command line if -help xtunitroot-
> gives you these and if not then install it from -ssc install
> xtunitroot- or find it using -xtunitroot-.
> The help file contains more information but there are some points
> which can be used to answer your questions:
> 1. You can not use varlist, so it means you have to run the command
> for varname. It means may be using it for one variable at a time or
> for each variable seperately.
> 2. You can not use -xtunitroot- for data with gaps or what we can say
> some observations on some series are missing. The data should be
> strongly balanced.
> 3. Once confirmed, you can use the first difference to overcome the
> issue of non-stationarity or this can be confirmed that the series are
> stationary in first difference.
> I hope more discussion will help us learn more on these issues. Also I
> hope this helps you somehow.
> --
> Best
> ---------------------------
> Muhammad Anees
> Assistant Professor/Programme Coordinator
> COMSATS Institute of Information Technology
> Attock 43600, Pakistan
> On Wed, Aug 22, 2012 at 6:13 PM, L P <> wrote:
>> Hi there,
>> I would be grateful if I could receive your help since I am new to
>> econometrics and STATA.
>> I am using STATA 9.0 an I am working on a panel data based on
>> observations for 30 countries (id) and 25 years from 1981 to 2005
>> (time). The database is unbalanced since it contains gaps in id and
>> time dimensions. The model specification looks like
>> Ln Emissions[it] = a + b1 Ln GDP[it] + b2 LnGDP^2[it] + b3 Ln
>> Trade(lag-1)[id] + ... + e
>> According to what I read in statalist, I am trying to test the
>> variables of my model specification for stationarity with
>> Levin-Lin-Chu test and the use of the followwing STATA commands:
>> . tsset id year
>> . levinlin variable name, lag(1)
>> My questions:
>> 1. shall I run the test for each single dependent variable? What about
>> the independent variable?
>> 2. how can I overtake the problem of the gaps to allow the test for
>> those variables characterised by gaps in the databse?
>> Furthermore, if I find a p-value > or = 0.05, I have to accept the
>> null hypothesis (that is the panels contain unit-roots). With the aim
>> of overtaking this problem, is it enough to build first differences of
>> the variable performing in this way?
>> Thanks in advance.
>> Lino
>> *
>> *   For searches and help try:
>> *
>> *
>> *
> *
> *   For searches and help try:
> *
> *
> *
*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index