题　目：A Semiparametric Approach to Discrete Choice Demand Estimation: Theory and Empirical Evidence

嘉　宾：Qinxin Chen（陈沁歆），Ph.D. Candidate，Washington University in St. Louis

主持人：左思 助理教授 上海交通大学安泰经济与管理学院

时　间：2026年10月14日（周三）13:30-15:00

地　点：上海交通大学安泰经济与管理学院包兆龙图书馆A303

内容简介:

Discrete choice models are central to demand estimation and counterfactual policy analysis, but existing approaches face a core tradeoff: parametric specifications preserve economic interpretability while imposing functional-form assumptions that real-world substitution patterns often violate, whereas flexible machine-learning methods relax these restrictions but forfeit valid root-n inference on policy-relevant demand objects. We resolve this tension by Semiparametric Debiased Multinomial Logit (SPDML), under which a low-dimensional parametric component captures the policy lever of interest, such as price, commission, or promotion, preserving its economic interpretation, while another choice-set-aware deep neural network learns the remaining utility components from own attributes, rival attributes, and choice context,including unstructured inputs such as text and images, so the researcher need not specify baseline utility, substitution patterns,or behavioral mechanisms ex ante. Neyman-orthogonal debiasing then yields root-n consistent, asymptotically normal estimators of own and cross demand responses, elasticities, and counterfactual policy effects. It encompasses a broad class of logit data-generating processes, including nonlinear utility, complementarity, and behavioral mechanisms such as decoy effects, choice overload, and reference dependence, within the logit framework. Through a mixed-logit bridge, it can approximate general random-utility models with bounded errors. Monte Carlo experiments show accurate recovery of policy effects where misspecified parametric models fail. In an application to salesperson commissions in Chinese retail pharmacies, SPDML aligns closely with quasi-experimental synthetic difference-in-differences benchmarks and outperforms leading structural and nonparametric alternatives.

演讲人简介：

Qinxin Chen is a Ph.D. Candidate in Quantitative Marketing at the Olin Business School, Washington University in St. Louis. Her research integrates AI/ML into choice models to recover causal effects of marketing policies, particularly in pricing strategies, healthcare markets, and digital platforms. Qinxin’s work leverages causal inference to address complex empirical questions in business and economics.

欢迎广大师生参加!