Joint prediction of automobile ownership and mileage by a cross-section model

A previously constructed model, which explains car ownership and private car use simultaneously and which was originally estimated on the 1980 Dutch budget survey, is now applied to 1985 in order to obtain validation of this model. Two methods are used: post sample prediction and re — estimation. This exercise raises some general issues of predicting individual discrete choice and of conditional prediction in a simultaneous framework. The main result is that a model which performs rather well at the aggregate level may fail to explain and predict behaviour at the level of the individual household.