Nonparametric mixed logit model with market-level parameters estimated from market share data

2023-09-22

Transportation Research Part B: Methodological

AI Summary

This paper addresses the limitation of market-level choice models—which rely on individual-level microdata and thus struggle to accurately identify preference heterogeneity—by proposing the first nonparametric mixed logit estimation framework using only market-level share data. Methodologically, it integrates nonparametric density estimation, moment matching, and generalized least squares optimization, while explicitly incorporating market equilibrium constraints to invert individual preference distributions purely from macroscopic aggregate data. Unlike conventional approaches requiring micro-level choice observations, our method eliminates the need for individual-level data, substantially improving parameter identifiability. Across multiple simulation studies and empirical market applications, it achieves high fidelity in recovering true preference distributions and reduces out-of-sample prediction error by an average of 32%. The framework establishes a novel paradigm for market structure analysis and counterfactual inference grounded exclusively in aggregate market shares.

Problem

• Estimates market-level parameters from share data
• Improves accuracy over traditional logit models
• Facilitates supply-demand integration in transportation

Innovation

• Nonparametric mixed logit model with market-level parameters
• Multiagent inverse utility maximization problem solving
• Improved accuracy and faster computation than benchmarks