Disentangling Autocorrelated Intraday Returns

Robust Estimation of Integrated and Spot Volatility

“…Lastly, our estimator is more robust to the choice of tuning parameters and model specifications. In terms of finite sample performances, our estimator is comparable to the Quasi-Maximum-Likelihood-Estimator (QMLE) (Da and Xiu, 2021b).…”

“…We compare the performances of four estimators developed in similar settings:The the Flattop Realized Kernels (FRK) (Varneskov, , 2017, the Pre-averaging estimator with ReMeDI bias correction (PaReMeDI), the quasi-maximum-likelihood estimator (QMLE) (Da and Xiu, 2021b), and the Two-scales Realized Kernels (TSRK) (Ikeda, 2015(Ikeda, , 2016. The next subsection will compare PaReMeDI with a class of pre-averaging estimators, including the pre-averaging estimator with local averaging correction (PaLA).…”

“…This is in contrast with the IV estimation: Many IV estimators preserve the consistency in the presence of random and irregular sampling times, though the limiting variance will be different, see Remark 4.2 in.4 For example, to perform the same estimation, our estimator usually uses less than 5% and 0.5% of the CPU time used by the estimators inJacod et al (2019) andDa and Xiu (2021b), respectively.…”

“…While Jacod et al ( , 2019 only consider noise of order O p (1), we allow noise to be shrinking as the data frequency increases. The "small noise" asymptotics are considered by Kalnina and Linton (2008), Da and Xiu (2021b), see also Chapter 7 of Aït-Sahalia and Jacod (2014) and Chapter 16 of Jacod and Protter (2011). We develop new volatility estimators using the ReMeDI estimators of the moments of noise to correct the bias of a pre-averaging type estimator.…”

“…As a consequence, it converges faster than the usual rate n 1/4 in the presence of shrinking noise. This is an advantage in dealing with recent highfrequency data where it is found that the scale of noise is much smaller, see, e.g., Da and Xiu (2021a).…”