报告时间🪼：2021年11月18日

               下午14:00-15:00

报告地点：腾讯会议

（会议ID：935 694 336）

报告嘉宾🤙🏿😊：兰伟

报告主题👨🏼‍🔬：等预测能力的最大子抽样检验

报告摘要

Maximum-Subsampling Test of Equal Predictive Ability

Recently, McCracken (2019) investigated the equal predictive ability test of Diebold and Mariano (1995, DM), and had two important findings about the DM test as studied by Giacomini and White (2006) and Diebold (2015) via numerical examples. First, when the loss differentials between two competing models are weakly dependent, the DM test yields size distortions in small samples. This is because the variance of the DM test statistic depends on the autocovariances, which are difficult to estimate accurately (see McCracken, 2019). Second, the DM test fails when the loss differentials are highly correlated. For the second case, McCracken (2020) provided a counterexample in which the test statistic diverges with probability one under the null hypothesis. To alleviate the issues raised by McCracken (2019, 2020), we propose a maximum-subsampling (MS) test that does not require estimating the autocovariances. Accordingly, the MS test is applicable for comparing the accuracy of two competing forecasts when the loss differentials are weakly dependent. We demonstrate theoretically that the MS test converges to the type I extreme value distribution under proper conditions. We further apply our proposed test to the examples modified from McCracken (2019, 2020) and show that the MS test performs satisfactorily, while the DM test either exhibits size distortions or fails.

个人简介

兰伟🫳🧑🏼‍🍼，博士毕业于北京大学光华管理学院🙄，现为西南财经大学副教授，博士生导师。主要研究方向为高维数据建模、大型网络数据分析和投资组合优化🐪。主持国家自然科学基金项目和多个重点重大项目子课题🥽。在JASA,AOS, JOE, JBES🙍🏼，《金融研究》等国内国际知名期刊发表和接收中英文论文40余篇👨🏿‍⚖️。