Describing the separability properties of empirical demand systems

In empirical demand analysis separability restrictions are usually imposed as maintained hypotheses. The separability inflexibility of most functional forms makes parametric tests of these restrictions appear problematic. We follow Pudney (1981) in stressing the usefulness of a description of the separable structure of preferences underlying an estimated model, and propose a non-iterative method for the production of such a description. An illustration with different models estimated for nine categories of goods on French aggregate time-series data is presented.