Unit Root Problem

[Murry Siyasiya]

Dear All,

I have my 5 variable model where the dependent variable is I(2) while 2 of the independent variables are I(0) and the other 2 independent variables are I(1). What do I do?

[Dirk Enzmann]

I don't understand your question. What do you mean by I(0), I(1), and I(2)? How does your model look like (what kind of model)? How do your data look like? How is this related to the "unit root problem"? Why do you think so?

Please note that your question could be answered much more easily if you provide us with (a) the data you are using (example data would be sufficient), (b) with an explanation what your initial analysis (or research) question is, (c) the Stata commands exactly as you used them (if you did), and (d) what you did expect or in which way the results are not as expected.

As a first step I would advise you to carefully read the Stata Forum's FAQ completely (!) before trying to improve your question.

Further pondering about your question it seems that your are investigating time series and have a problem with non-stationarity (or stationarity). It would have been helpful if you had mentioned this from the start: There are many statistical procedures available in Stata and readers not familiar with time series analysis won't understand your problem if they don't identify I(0), I(1), etc. as indicating that you are running unit root tests (if I understand this correctly).

[Andrew Musau]

I have my 5 variable model where the dependent variable is I(2) while 2 of the independent variables are I(0) and the other 2 independent variables are I(1). What do I do?

For #2, \(I(0),\; I(1),\) and \(I(2)\) refer to the order of integration and is a standard way to express this. A series that is \(I(0)\) is stationary in levels; a series that is \(I(1)\) becomes stationary after first differencing; and a series that is \(I(2)\) becomes stationary only after second differencing.

For the OP, what are the dimensions of your data? For the most part, we assume that economic variables are either \(I(0)\) or \(I(1)\). With short \(T\), unit root and cointegration testing is very unreliable. The small-sample distributions of most tests are wide, power is weak, and distinguishing between highly persistent \(I(0)\) and genuine \(I(1)\)/\(I(2)\) is basically impossible when you only have, say, 10–20 periods or less.