报告题目：Nonparametric Trending Regression: Estimation and Testing

报 告 人：宋晓军

时   间：10月20日（周三）上午10:00

腾讯会议：317 882 358

报告人简介：

宋晓军，北京大学光华管理学院商务统计与经济计量系副教授，博士生导师，西班牙马德里卡洛斯三世大学经济学博士。主要研究兴趣是理论计量经济学，包括非参数/半参数方法，假设检验和自助法，以及计量经济学的应用等。论文发表在Journal of Econometrics，Econometric Theory和Journal of Business & Economic Statistics等国际期刊。主持和参加自然科学基金面上项目和国家重点专项等。目前担任Economic Modelling副主编。

摘要：

Nonparametric approach is increasingly employed to model the deterministic trend function in panel data models due to the fact that it can help reduce misspecifications of the trend function and let the data "speak for themselves". When allowing for possible dependence in time and cross-sectional dimensions, the question of efficiency improvement via utilizing dependence structure arises naturally. Robinson (2012) efficiently estimates a nonparametric trending regression with possible cross-sectional dependence and heteroskedasticity. However, he rules out temporal dependence in the error terms, and sometimes this can be restrictive and unrealistic because serial dependence is widely existing in long panel data models. In this article we propose to extend Robinson's (2012) approach to allow for the presence of autocorrelation, cross-sectional dependence and heteroskedasticity in error terms simultaneously, and shows that the autocorrelation also contains useful information for efficiency gains in nonparametric estimation of the time trend function. Lastly, we propose a nonparametric test for the common trend specifications in panel data models with fixed effects. Our test (i) is consistent against various alternatives that deviate from the null, and no prior information on the alternatives is required for our test; (2) it is general enough to allow for heteroskedasticity, cross-sectional and serial dependence in error components; (iii) unlike many tests for model speci cation in the literature, which often have nonstandard asymptotic distributions, it has an asymptotic normal distribution under the null hypothesis.

经济与金融学院

2021年10月18日