Separate Noise and Jumps From Tick Data: An Endogenous Thresholding Approach

Abstract

We study the problem of jump detection for ultra-high-frequency tick-by-tick data. We propose a novel easy-to-implement procedure that can separate the contribution of microstructure noise and that of finite activity price jumps from the price process, which may have interesting implications on asset pricing and forecasting problems. We provide theoretical grounds of our approach, and suggests practical guidelines for determining the tuning parameter. Making a comparison with the “star performers” in a recent comprehensive review for jump detection methods by Maneesoonthorn et al. (2020) as well as a test based on Christensen et al. (2014) on tick data, we show that our method performs admirably well via extensive simulation and rich empirical illustration.