Unit root testing in Time series data
1. Trend of drift in unit root testing?
Professor Muili Adebayo Hamid explained as such > I once read a journal that said you
can equally estimate a univariate model with the variable of concern adding trend to know
which is better. for instance, GDP. estimate In eviews. ... gdp can @trend .. if the pvalue
of the trend is significant then you choose with trends and intercept ,if only intercept is
significant then choose only intercept but no trend and if none is significant you can
choose none. although I can't vividly remember which journal coz its been long.
2. None, intercept, intercept and trend - Which one should be chosen in ADF (unit
Professor Moulana Naykrasyvishyy Cholovik explained as such> 'Dolado procedures
tells tht you should start from ''with trend and intercept'' then 'with intercept but no trend''
then 'no interecpt and no trend'' and stop where you find the series is stationary. you can
learn more details in ENDERS APPLIED TIME SERIES.
3. Lag selection in ADF test
Professor Moulana Naykrasyvishyy Cholovik commented as such>> You can follow for
optimum lag length for unit root test 1. general to specific: choose a maximum lag length
like 8, 12 etc. based on size of observation. Then start removing higher lags if the
coefficient of lag is not significant in estimated ADF model. 2. use information criteria AIC,
SBC etc. eviews and other softwarss has option for this. 3. choose any lag where you find
no serial correlation. Basically including longer lags is to avoid serial correlation. So You
can choose any lag even no lag if ADF estimated model is not suffering from serial
4. None, intercept, intercept and trend - Which one should be chosen in ADF (unit
Professor Shishir Shakya has made a video as such>
5. Durbin Watson test and panel data
Professor Olasehinde Timilehin commneted that > Durbin Watson can be applied in Panel
data for autocorrelation testing as per Baltagi.
6. Should not have lags
Professor Noman Arshed commented> Durbin watson test is applicable for time series
autocorrelation testing but it has one limitation that if there are lags in model, we cannot
7. Disadvanatages of Durbin-Watson test
Professor Nicat Gasim commented as such> Durbin Watson test has three
disadvantages. 1. If models comtain lagged variables DW test cannot use for detecting
autocorrelation. 2.It has indesicion area. 3. DW test derermine only first order
8. Whether any unit root?
Professor Imran Rajan posted the following figure for interpretation.
Professor Vahe Nazaryan commented > Since Prob<0.05, this means that you can reject
the null of non stationary (has a unit root = non stationary) and conclude that your series
are stationary. I think yes you can Like · Reply · November 19, 2016 at 4:11pm
Professor Keith Araneo-Yowell commented> you have to check the values of the Dickey-
Fuller distribution for the kind of Dickey-Fuller test you are doing. The critical tau value is
in the dickey-fuller table. It will be different based on the number of observations and
whether you are testing for a Unit root, unit root with a drift, or a unit root with a drift and a
Professor Festus Nkwo commented> It is ok to reject the null at 5% and 10% and
conclude that the sequence is stationary. However, it is not save to accept the alternate at
1% in the first difference level.
Meo School of Research
Unit root testing in time series data
Hossain Academy Note