In the first part of this paper (sections 2 and 3), the components of Hachemeister's (1975) regression model are interpreted in Hilbert spaces of random vectors. These interpretations, already transparent in Norberg (1979), are more satisfactory than those in the Hilbert space of squareintegrable random variables considered by De Vylder (1976). The title of this note comes from its second part (sections 4-8) wherein credibility estimators are derived under constraints on the credibility matrices.