# Error Correction Model (ECM) procedure

I am getting familiar with ECM so I would like to ask for help in order to understand it. My goal is: performing a ECM in a multivariate context.

I was reading ECM is part of a Cointegration analisis. This Cointegration follows two main steps:

1. Estimate a long term equilibrium (using levels or log_price), then
2. Perform ECM (using returns).

I also read that before performing the first step I need to verify if the variables (using levels) are following a unit root process. In case they don´t, their difference needs to be verified.

My questions are:

• Am I missing something in the procedure?
• let´s suppose I find all varibles are integrated order two I(2) expect one which is I(1). Should I use it for the long term equilibrium (step 1)? How to use this information (I(2)) in the Cointegration procedure (maybe it needs to be excluded)?
• Can I use some variables for step one, and add some others for the ECM? If yes, what should I need to consider (e.g. do I need to perform more unit root test over those new variables?)

Thx a lot.

• Are you estimating ECM in a bivariate context or do you have more than two variables? – dlnB Mar 18 at 22:05
• @dlnB multivariate – Newbie Mar 18 at 22:28
• For cointegration, all variables must be $I(1)$. – dlnB Mar 18 at 22:30