Around the Corner (Solution): a Multiple Discrete-Continuous approach to Food Choice Modeling

Abstract Demand modeling has long been used to inform fiscal policies, particularly for frequently purchased goods such as food, alcohol, and cigarettes. Policy simulations have traditionally relied on demand systems. In recent years, however, discrete choice models have become more common, as they are better suited to sparse datasets like consumer panel scan data, where non-purchases account for a large share of observations. These modeling approaches can be integrated into discrete-continuous specifications that accommodate multiple discreteness, allowing utility-maximizing consumers to select combinations of several goods from a potential basket. In this paper, we demonstrate how the Multiple Discrete-Continuous Extreme Value (MDCEV) model -- commonly used in transportation research -- can yield valuable behavioural insights when applied to scanner data on food purchases. In particular, we show how the model can capture key behavioural parameters, such as the satiation power of different food and drink categories, and account for unobserved heterogeneity in baseline preferences across households. We also propose an empirical adjustment to the model that improves both in-sample fit and out-of-sample predictive accuracy, while enhancing computational efficiency. Finally, we apply the model to simulate a fiscal policy aimed at promoting more sustainable food choices, highlighting its ability to capture behavioural responses and substitution patterns under budget constraints. The empirical analysis is based on a consumer panel of 3,954 Italian households, covering all weekly food purchases between January 2023 and April 2024.