We propose a panel regression model with a predetermined and fixed number of classes, where each class is defined by its parameters, but any reference as to which group any observation belongs to is absent. The classes or groups are rationalized by a willingness to attribute some of the observed...