Asymptotically Constant Risk Estimator of Long-run Variance in Time Series

题目:  Asymptotically Constant Risk Estimator of Long-run Variance in Time Series

报告人：Professor Chun-Yip Yau, The Chinese University of Hong Kong

时间： 2018年12月11日（周二）15:00

地点：浙江大学工商管理楼200-9

摘要：Estimation of long-run variance is important for inference, however, estimating it non-parametrically in time series is difficult because its performance relies on the ability of picking a good problem-specific bandwidth. For all existing estimators, bandwidth selection is an ill-posed circular problem: picking a good bandwidth requires the long-run variance, the original quantity of interest. This paper proposes a non-parametric estimator for the long-run variance. It does not require estimation of the optimal bandwidth.We achieve this by introducing a novel concept of converging kernel, which can automatically balance variance and squared bias. We prove that (i) the resulting optimal bandwidth is universal instead of problem-specific, and (ii) the estimator exhibits a constant risk behaviour asymptotically.

欢迎大家参加！

联系人：张荣茂（rmzhang@zju.edu.cn）