Guo Yan - University of Melbourne

Abstract

Heteroskedastic binary choice models (HBCMs), proposed by Manski (1975, 1985), are known to pose both practical and theoretical challenges: (i) conventional policy parameters, such as partial effects, are not naturally identified in the model, and (ii) estimators converge at rates slower than √n and require nonstandard inference. Exploiting the fact that HBCMs admit an exact nonparametric interpretation in terms of level curves of the conditional choice probability (CCP), we propose a new estimator that corresponds to constant marginal rates of substitution along a CCP level curve. These rates quantify how one variable must adjust to offset changes in another while holding the CCP fixed. We derive a linearization of the estimator and show that the slower-than-√n convergence rate arises from estimating an irregular aggregation functional of the CCP, a function defined on ℝᵈ, along a (d − 1)-dimensional level curve. We establish the limit distribution and provide a valid inference procedure. Our inference procedure is easy to implement: it is based on self-normalization and does not require knowledge of the exact convergence rate. We also discuss extensions to nonparametric identification of HBCMs and to the estimation of heterogeneous marginal rates of substitution along level curves.

Additional information:

• Speaker: Guo Yan 
• Time: Thursday, 02.07.2026, 16:00 - 17:00
• Location: Faculty Lounge, Room 0.036
• Further links:
  • Guo Yan's webpage
• Organizer: Statistics Group
• Contact:
  • Almut Lunkenheimer
  • +49 228 73-9228
  • ifs@uni-bonn.de