Federico Bandi*, Jeff Russell, Chen Yang (University of Chicago)
Observed high-frequency financial prices can be considered as having
two components, a true price and a market microstructure noise perturbation.
It is an empirical fact that the second moment of market microstructure
noise is time-varying. We study the optimal design of nonparametric variance
estimators in linear forecasting models with time-varying market microstructure
noise. Specifically, we discuss optimal frequency selection in the case
of the classical realized variance estimator and optimal bandwidth selection
in the case of kernel-type integrated variance estimators. In this setting,
we show that the sampling frequencies are generally considerably lower
(the bandwidths are generally considerably larger) than those that would
be optimally chosen in linear forecasting models when time-variation in
the second moment of the noise is unaccounted for. Conditional and unconditional
frequency/bandwidth choices are discussed.