Marcel-Dagenais Econometrics Seminar 2017-2018
joint with the Département de sciences économiques, Université de Montréal
room C-6070-9 (U. of Montreal, Pavillon Lionel-Groulx, 3150, rue Jean-Brillant)
Organizer : Marine Carrasco (U. of Montreal)
Abstract
This paper proposes a framework to test for the presence of jumps in high frequency returns. We consider a framework where the efficient log-price follows a semi-martingale diffusion that is contaminated with a dependent microstructure noise and a finite activity jump process. The proposed test is designed to automatically select the nominal size that minimizes the overall probability of error associated with the decision rule. The test procedure requires a first-step estimator of the integrated variance and the autocovariances of the noise, which we construct by exploiting the moments of standard realized measures. Monte Carlo simulations show that out test detects fewer and fewer jumps as the variance of the microstructure noise increases. Nevertheless, our procedure detects a much larger number of jumps than existing tests, both in simulations and in empirical applications.
Keywords : Bipower Variation, Jump Detection Test, Method-of-Moment, Microstructure Noise, Quadratic Variation, Realized Kernels
JEL Classification : C12, C58
