Welfare Analysis for Discrete Choice with Interval-data on Income

---

• 研討會日期 : 2016-01-19
• 時間 : 14:30
• 主講人 : Dr. Ying-Ying Lee
• 主持人 : Professor Yu-Chin Hsu
• 地點 : Conference Room B110
• 演講者簡介 : Dr. Lee received her Ph.D. in Economics from University of Wisconsin-Madison in 2013. She is currently a Postdoctoral Research Fellow in Nuffield College, University of Oxford. Her research fields are Econometric Theory, Empirical Microeconomics. She is applying for a position of the Institute of Economics, Academia Sinica now.
• 演講摘要 : This paper develops a new method of welfare estimation in binary choice models with general unobserved heterogeneity and income observed in intervals. Bhattacharya (2015) has shown that the marginal distribution of the equivalent/compensating variation (EV/CV) resulting from a hypothetical price-change can be expressed as a closed-form transformation of the choice probabilities. However, since income is only interval-observed, the conditional choice probabilities and hence the distributions of EV/CV are not point-identified. We adopt a probit/logit model that is a best parametric approximation to the conditional choice probabilities. This specification leads to a simple, valid inference procedure for the partially identified distributional features of EV/CV, subject to a revealed preference restriction pre- scribed by economic theory. Partial identification from the interval-observed income along with uncertainty concerning which restriction from theory binds yields non-differentiability of the welfare/policy objects of interest and poses a key challenge to inference in this setting. Our best parametric approximation facilitates a solution to this inference problem. In particular, we show that our estimator is directionally differentiable so that recently developed bootstrap methods (Fang and Santos, 2014, Hong and Li, 2015) can be applied. Our approach extends to more general settings where a class of set identified functions are subject to linear inequality restrictions. We impose a best parametric approximation and show how to obtain valid inference on general features of the partially identified functions. Our methods are illustrated with an application to demand for insecticide-treated bednets in Kenya.