Speaker: Chencheng Fang（方陈承），University of Bonn

Time/Date: 2024年3月6日下午3：00

TENCENT meeting: 275346230

PW: 361944

Abstract:

Non linear models are widely applied in empirical researches of Economics and Management. In this seminar, I focus on three classes of commonly used non-linear Econometric models: count data, censored data and binary data models. For each class, canonical and tailored models are introduced. Most importantly, I will suggest some common misconceptions that empirical researchers may have on non-linear models. For example, it is usually believed that whenever there is a tested overdispersion in count data, negative binomial regression should be preferred over Poisson regression. However, on one hand, Poisson model is actually consistent only under a single conditional mean assumption without any restriction on the data distribution. It can also be efficient when we use sandwich covariance instead. On the other hand, negative binomial model can often be non-convergent, especially in panel data case.

Bio:

Chencheng Fang is working as a Research Assistant at Institute of Finance and Statistics, University of Bonn, Germany. He is also pursuing a PhD in Economics at the same institute. He is interested in application oriented Econometric/Statistical models that concentrate on real data problems. Also, he has strong enthusiasm in contributing to platforms such as GitHub that advocate open collaboration for production of academic researches and data softwares. Currently, he is working on a research that spans over functional data analysis and difference-in-differences. He has publications on journals such as Production and Operations Management, Computer in Human Behavior.