Basic Procedure for Causality Test

In empirical work, causality between two or more variables is very popular particularly to find the relationship directions. Two common examples are energy and economic growth nexus, relationship between trade and economic growth.

Basic procedure in testing causality can be summarized as follows:
Step 1: Unit root test
In this stage the order of integration for the variables needs to be checked such that the level of stationarity of the time series can be found.

Common method for this test is Augmented Dickey Fuller (ADF) test.

Step 2: Co-integration test
Co-integration is unique and it represents a long run relationship or equilibrium among tested variables. In short, we need to evaluate whether or not the linear combination is stationary. The non-stationary variables are co-integrated if their linear combination (the error-term) is stationary.

Two common methods used to check existence of cointegration are:
a. Engle and Granger cointegration test.
b. Johansen and Joselius cointegration test.

Step 3: Granger non-causality test
In this step, we determine the relationships between varibles. 
If the variables are non cointegrated then they only have short run relationships. To assess the direction of causality between the variables, we use Standard Granger test.

On the other hand, if the variables are cointegrated then the variables have not only short run but also long run relationships. We usually use vector error correction model (VECM) to check the causality direction.

In addition, Toda and Yamamoto developed a procedure to deal with the problems that exist in VECM approach.

Here is very good reference:
http://davegiles.blogspot.com/2011/04/testing-for-granger-causality.html