The log-likelihood function contains a quadratic form in the disturbance v, which results in a well-behaved optimization problem. If the variance-covariance matrix for y is positively defined, the general asymptotic properties for the ML estimators will hold. In order to maintain the positive variance-covariance