Forecasting Shares for Multiple Purchases: Towards a Test of the New Individual-level Multinomial Probit

Abstract

We report on a field test of a new Individual-level Multinomial Probit (IMNP) model that applies to stated preference data gathered at the level of the individual consumer. Instead of inferring a covariance matrix of brand alternatives from a large sample of revealed preferences at the group level, this approach derives a covariance matrix from brand attribute similarities, weighted by the attribute importances for each consumer. The proposed IMNP simulator is not subject to the problem of Independence from Irrelevant Alternatives (IIA). Results from a Monte Carlo simulation show that the simulator performs as well as or better than the traditional Maximum Utility Rule (MUR) in ordinary situations, but is particularly useful with small samples. The simulator lends itself to forecasting shares for product assortments and frequently repurchased product classes. The paper outlines the new IMNP and reports on preliminary steps of a field test of the model, testing forecasting accuracy of students’ preferences of confectionary in product assortment selection and repeat purchase situations.