A Generalised Nested-Logit Model of the Demand for Automobile Variants

This paper estimates the demand for car model variants instead of looking only at demand for models in terms of the 'baseline' variant of each model as done in the literature. The data has sex and age of the buyer for every car sold in Norway 2000-2004, in addition to characteristics of the cars. The demand model uses this information to estimate taste coefficients which depend on demographic characteristics. A nested logit model and a generalised nested logit model are used to induce correlation in the logit error between products with observable and unobservable similarities. Results indicate that it may be problematic to have different logit errors for every product when the number of products is very high, even when allowing for flexible correlation patterns