A maximum likelihood based approach to estimating a vector error correction model with several endogenous variables involving nonstationary as well as stationary variables. The procedure includes testing for and estimating multiple cointegration relationships among the variables (cointegrating vectors).