报告题目：Estimation and model selection in general spatial dynamic panel data models

报告人： Wu Yuehua， 加拿大约克大学教授

报告时间：2024年8月9日，16：00- 17：00

报告地点：数学楼423

摘要：

In this talk, I will introduce a novel inverse method to validate transmissibility assumptions and to explore overfitting issues. Mechanistic models have proven to be successful in predicting the short-term future and providing insight into the disease dynamics. However, they are based on our prior understanding of the world and hence, are only as "good" as that prior understanding, and do not extend to situations where the underlying mechanisms are unknown. Simple to advanced machine-learning models are developed fully from data and without incorporating prior human expert knowledge. Some of these models have shown an exceptional forecasting power; however, they often provide no intuition about the dynamics – the reason why they are often questioned and even avoided by mathematicians. A natural bridging between the two approaches would be to take a mechanistic modelling approach for those compartments of the disease spread whose governing dynamics are well-understood and a machine-learning approach for those other yet not-well understood compartments. If time allows, I will briefly discuss our new hybrid approach for disease forecasting and mechanism understanding. Common methods for estimating parameters of spatial dynamic panel data models include two-stage least squares, quasi-maximum likelihood, and generalized moments. In this talk, we present a method that uses the eigenvalues and eigenvectors of a spatial weight matrix to directly construct consistent least squares estimators of parameters of general spatial dynamic panel data models for both undirected and directed networks. Our method is conceptually simple and effective, and easy to implement. Results show that our parameter estimators are consistent and asymptotically normally distributed under mild conditions. We demonstrate the superior performance of our method through extensive simulation studies. We also provide a real data example.

Joint work with Hou, Jin and C.R. Rao

报告人简介： 吴月华，加拿大约克大学数学与统计系教授。她师从世界著名统计学家C. R. Rao，于1989年获得美国匹兹堡大学统计学博士学位。目前，她从事高维数据分析、模型选择、变点分析、时空建模、环境统计和统计金融等多领域研究， 是国际统计学会的当选会员。她在Proceeding of National Academy Science, USA, Biometrika, Journal of Economics等期刊上发表了140余篇学术论文，目前承担加拿大国家自然科学基金科研项目。