摘要：This paper develops a nonparametric test for quantile predictability with persistent regressors. The test uses a kernel-weighted U -statistic constructed from restricted-null quantile scores and accommodates linear and nonlinear predictive relationships at an individual quantile or over a prespecified quantile interval, without estimating the unrestricted regression function. For a scalar local-to-unity predictor, we derive a mixed-normal null limit involving self-intersection local time and establish consistency against shrinking alternatives. We also develop null inference for two local-to-unity predictors with full-rank stochastic trends, obtaining a Gaussian limit with logarithmic normalization. Despite these different limits, a common self-normalization yields standard normal inference in both cases under the stated conditions, allowing embedded endogeneity. A null-imposed wild bootstrap provides finite-sample calibration. Monte Carlo simulations illustrate the test’s power against linear and nonlinear alternatives and show that bootstrap calibration improves size control, particularly at tail quantiles. An application to U.S. stock returns uncovers predictability concentrated in the right tail of the conditional distribution.

时间：11月9日下午14：00

地点：经济学院6-210会议室

嘉宾简介：

吴吉林：男，浙江安吉人，中国民建党党员， 2010年博士毕业于厦门大学王亚南经济研究院，现为计量经济学教育部重点实验室（厦门大学）副主任，《计量经济学报》助理主编，厦门大学经济学院金融系、王亚南经济研究院、邹至庄经济研究院教授，博士生导师，“闽江学者奖励计划”特聘教授，曾任山东大学教授，博士生导师，并曾入选山东大学齐鲁青年学者。主要研究方向为金融计量理论与应用、金融风险管理以及机器学习理论。到目前为止，以第一作者或通讯作者在Journal of Econometrics、Econometric Theory, Journal of Business & Economic Statistics、Econometrics Journal、Statistica Sinica、Journal of Time Series Analysis、《经济学（季刊）》、《世界经济》、《数量经济技术经济研究》、《统计研究》、《管理科学学报》、《系统工程理论与实践》等国内外重要期刊上发表论文近40篇。担任国家自然基金匿名评审专家，国家统计局统计规划项目匿名评审专家，并还担任Econometric Theory、Journal of Business & Economic Statistics、Journal of Applied Econometrics、Econometric Review、《经济学(季刊)》《管理科学学报》以及《统计研究》等杂志的匿名审稿人。作为项目负责人主持了包括国家自然科学基金青年项目、面上项目、教育部人文社科基金等在内的多项课题。联系方式：手机：18959207219；电子邮箱：rainforest1061@126.com。