Grahn, (1995) introduced the Conditional Least Squares estimators for the class (I) of bilinear models. Such estimators have a variance which is difficulty to derive analytically. The aim of the present paper is to consider ...