Indian Institute of Management Shillong
Question
Asked 5th Aug, 2016
VAR or VECM when the dependent variable is I(0)?
I intend to run a VAR on a set of variables, but when testing for stationarity, I found that my key variables of interest were I(0) and some others were I(1). In that event, is it possible to take the first differences of those other variables and run a VAR? Or should I consider running a VECM (although I think it is not possible to do so considering the fact that the key variables of interest are stationary at level)?
Any feedback would be greatly appreciated.
Most recent answer
Even after taking first difference of the variables they are integrated not of the same order, better to do ARDL Bound testing approach. But all these issues will arise only when you see the variables are co integrated or not?
All Answers (4)
Indian Institute of Management Shillong
Taking first difference of the variables and running a Vecm would do, provided all the variables in the first difference form are integrated of the same order. If no cointegration exists you can do a VAR model also, no need of VECM
1 Recommendation
University of KwaZulu-Natal
You might also want to consider using an ARDL model which allows you to test for cointegration even when you have both I(0) and I(1) variables.
The Central Bank of Sri Lanka
Thank you Mousumi and Barry for your responses.
@Mousumi, like I said, i have a mix of I(0) and I(1) variables. Do you suggest that I should take the first difference of the non-stationary variables to make those stationary and then run a VECM? But then, what would be the case when the dependent variable is I(0)? Then there cannot be a cointegrating relationship right?
@Barry, I saw a similar suggestion for queries on the same issue but was wondering how it would differ from a VAR/VECM as those too take into consideration the lags (although the same no. of lags for all variables unlike in ARDL).
Would very much appreciate your response.
Similar questions and discussions
Related Publications
The primary aim of the paper is to place current methodological discussions in macroeconometric modeling contrasting the ‘theory first’ versus the ‘data first’ perspectives in the context of a broader methodological framework with a view to constructively appraise them. In particular, the paper focuses on Colander’s argument in his paper “Economist...
This paper explores the existence of long-run relationship between government expenditure and GDP in Bangladesh using Penn World Table annual data from 1973-2000. The modern time series econometric approach is undertaken to identify the relationship using various forms of theories behind this. Firstly, we find the existence of long-run relationsip...
In order to estimate causal relations, the time series econometrics has to be aware of spurious correlation, a problem first mentioned by Yule [21]. To solve the problem, one can work with differenced series or use multivariate models like VAR or VEC models. In this case, the analysed series are going to present a long run relation i.e. a cointegra...