Cointegration model and error correction model is basically meant for long-run relationship. And we use cointegration technique specially when all the variables are no-stationary. Error correction model basically corrects the fluctuation if any in the model while estimating the long run relationship.
1. Can i use Cointegration models to check the relationship between Export as Dependent Variable and Independent variables such as Industrial Production, GDP etc.
2. How Cointegration models are better to Regression analaysis?
The best advice would be to get some expert advice before you try any quantitative analysis. This particular area which tries to examine dynamic relationships among variables is fraught with issues and problems particularly for the unwary.
I am simplifying my response so that you can pick up the main thrust of the cointegration/regression issues. I am also just considering simple bi-variate relationships to make it easier.
When we wish to examine if variables have a theoretical relationship we typically specify a model and then estimate the model via empirical analysis and regression analysis. The nature of the relationship (or indeed the decision if any relationship exists) is usually hypothesised using logic, economic reasoning or empirical evidence In general, the estimated regression model contains statistical evidence (Rsq, t, F tests) which can be used to test whether a relationship exists or the nature of the relationship.
However, in the case of time series data (in many cases) certain characteristics of time series data "invalidate" the standard statistical evidence presented as output in regression models. Typically regression using time series data which is non-stationary produces mis-leading Rsq, t and F statistics. These statistics suggest that a relationship exists between the variables when no such theoretical relationship exists. These models produce poor predictions.
One of the dangers of much empirical work undertaken using certain time series data is that regression results will seem to indicate relationships exists between variables when they typically do not exist. This means the regression results are mis-leading directing researchers to make false conclusions.
You can try to avoid this issue if you try and regress variables that are stationary. This requires you to do some preliminary investigation of the variables and typically involves some variable transformation leading to a different theoretical model. Regression using these typically transformed stationary data will be free of the spurious regression issue above but may suffer from not being theoretically appropriate.
However, not all non-stationary variable relationships will lead to the spurious regression issue. Certain non-stationary variables are connected or co-integrated and typically regression of these variables will correctly determine that a relationship exists. Predictions from these models will typically be "good".
The types of variables that are likely to be co-integrated are those that are linked closely by logic and economic reasoning. Thus it is important to think carefully about the logical and economic connections between your variables and not regress just any variables of interest.
If you seek to understand whether variables are co-integrated and verify this empirically you can reformulate your regression model so that you can test for possible co-integration. There are several variants possible for this reformulation. You will need to understand and research the various tests for co-integration. Statistical software which has more sophisticated regression routines can test for co-integration.
The above explanation is very basic and is just skimming the surface. There are many more issues to consider. The nature of the relationship you seek to describe may be dynamic with lagged responses and this opens up more issues. My best advice to you would be to seek out a statistician or an econometrician with experience in time series models to help you with the task.
A vector error correction (VEC) model is designed for use with nonstationary series that are known to be cointegrated. Engle and Granger (1987) pointed out that a linear combination of two or more non-stationary series may be stationary. If such a stationary, or I(0), linear combination exists, the non-stationary (with a unit root), time series are said to be cointegrated. Please check below for ex-im model.
Its popularity also stems from the fact that co-integration of non-stationary variables is equivalent to an error correction(EC) process, and the ARDL model has a departmentalization in EC form(Engle and Granger, 1987; Hassler and Wolters, 2006). The existence of a long-run / co-integrating relationship can be tested based on the EC representation. A bounds testing procedure is available to draw conclusive inference without knowing whether the variables are integrated of order zero or one,I(0) or I(1), respectively(Pesaran, Shin, and Smith, 2001).