The exact insensitivity of market budget shares and the "balancing effect".

We reformulate Grandmont's and its successors' notion of behavioral heterogeneity such as to get the exact insensitivity of the aggregate budget share function with respect to changes in prices and income, instead of a mere approximate insensitivity. We propose a non parametric set-up such that, if the population is distributed according to some "uniform" probability measure, the aggregate budget share function is constant. The important contribution is that this exact insensitivity is not explained by any insensitivity at the microeconomic level but rather by an exact "balancing effect". We give illustrative examples of populations that fulfill our requirements.