Instability in mixed logit demand models. (June 2022)

• Record Type: Journal Article
• Authors: McFadden, Daniel
• Is Part Of: Journal of choice modelling. Volume 43(2022)
• Publication Date: 2022-06
• Subjects: Mixed logit -- Market simulation; Decision making -- Periodicals; Social choice -- Periodicals; Decision making; Social choice; Periodicals; 302.13
• DOI: 10.1016/j.jocm.2022.100353
• Languages: English
• ISSNs: 1755-5345

Abstract

Abstract: When a mixed logit demand model is used to estimate market-clearing prices, the high influence of the left tail of the random price coefficient can lead to numerical and statistical instability. I show this is an issue with any mixing distribution whose price coefficient is not bounded away from zero. I give conditions under which market equilibrium prices implied by mixed logit models exist, and show they are satisfied for lognormal mixing and some cases of truncated normal mixing. However, even when market equilibria exist, these models can be unstable and produce statistically unreliable and economically implausible conclusions. Highlights: Mixed logit demand models can be numerically and statistically unstable. Problem: Price coefficients not bounded away from zero - truncated, log normal. Model hypersensitive to "yea-saying" in data. Jumps and hypersensitivity in profit-maximizing prices impede estimation. Stabilizers: Income-constrained demand, bound price coefficients away from zero.