Hi,

You need to read about the difference between VAR models and VEC models, and the long run equilibrium. If you have access for this book will help you in this direction:

https://www.amazon.com/Econometrics-Example-Damodar-Gujarati/dp/0230290396

Hi Ayman.

Thanks for replying, I don't have it, sorry. Would you sum up? so this is wrong then?

You are asking "What are they doing", yet you are running analysis. You have to study and understand the concept of stationarity and cointegration before embarking on this journey. The first step isn’t grabbing Eview for analysis.

Stationarity

To understand cointegration, first, the concept of stationarity should be learnt. A time series is called stationary if it doesn’t wander off to infinity or stays around the mean. In simple terms, a price series which doesn’t have much price movement is called stationary.

Cointegration:

No worries if the price series is not stationary.

If we have two non-stationary time series X and Y that become stationary when differenced in (I(1)), such that the linear combination of X and Y is stationary (aka, I(0)), then we say that X and Y are cointegrated. In other words, while neither X nor Y alone hovers around a constant value, some combination of them does, so we can think of cointegration as describing a particular kind of long-run equilibrium relationship.

Your Problem(s)

Since you are having different levels of stationarity I(0), I(1), I(2), etc, In this case you should be using ARDL approach to cointegration as popolarized by Pesaran et al. (2001). This method has the advantage of using it with a mix of variables that are integrated of different degrees.

*

*

*

Goodluck

Hi Osarumwense,

So you mean that I can do ARDL in EViews, and then see the cointegration and then run the VAR-model? (or VECM?)

None of your graphs look essentially I(2). Your graphs suggest the possibility of an unusual episode impacting on all series around one third of the way into the sample. This can cause standard tests to over-estimate the degree of integration. It may be worthwhile to re-run your stationarity tests with this episode excluded.

Hi Melina, let me see if I understand at least one of your concerns. It seems to me that you are asking why should we bother looking for cointegration relationships if just by taking the first dfference of your series you achieve stationarity. Is that right?

Well , if that is the question I think is a good one. I will give you two reasons:

1. Sometimes we want to ask questions about GPD, a stock price or an exchange rate. These series are typically I(1). So, if we take the first difference of them we obtain stationary series and we can work with them easily. The problem with that is that we will not be responding questions about GDP but about the growth rate. See the difference? Sometimes we care about the price of an asset, because we want to buy it, so we really want to understand what lies behind the evolution of the price. If you take first differences, you will be modelling returns, not the price.

2. Granger showed that if you have a set of M series that are cointegrated, the first difference of them depends on the cointegrating relationship. Therefore, if you just model the first difference without including the cointegrating relationship you will have a misspecification problem. How seriuos is that problem? It depends, sometimes is not seriuos at all, in other occasions it can play an important role.

Hope it helps

Best

Pablo

First may I say that it would be easier to understand your question if the attachments were supplied as text or pdf files.

How did you arrive at the conclusions that your series were I(0), or I(1) or I(2)?

Your Johansen tests appear to be of two first differences. the results indicate that these first differences are stationary. (I have not used eviews for many years and may not understand the output.

In working with unit value tests, stationarity and non-stationarity it is important to keep the economics of your problem in mind. WhaT does economic theory indicate about the stationarity of the processes that you are examining?

Dear Melina, if I understand your question, you want to know the connection between stationarity test and cointegration test? OK! This is a time series phenomenon.. Before the economic researchers became conscious of the time series characteristics of time series data, economic policies have relied on level state OLS regression with the assumption that the underlying time series are stationary. A stationary time series is one whose mean, variance and autocovariance are time invariant.

However evidence have shown that most economic time series are non-stationary and OLS regression involving non-stationary time series may lead to spurious results. Hence the need for stationarity test. From stationarity test, we find out whether a variable is stationary or not. If it's non-stationary, we apply differencing to make it stationary. Thus, in order to overcome the problem of spurious results, economic researchers now rely on differenced stationary time series. However when we difference time series, we inevitably incur loss of important longrun information. This is another problem on its own. Hence the need to carry out cointegration test. The aim of cointegration is to find out if a linear combination of non-stationary variables is stationary. If cointegration exists between two variables that share similar non-stationary properties, then regression can proceed without generating spurious results. The implication of cointegration is that deviation may occur among the variables but that's temporary as equilibrium holds in the longrun for them. Thus the presence of cointegration between two variables that share similar non-stationary properties implies that consistent estimate of longrun coefficient is evident. Thus there is need to specify and estimate an error correction model in order to reconcile the longrun behavior of variables with their shortrun responses.