Title: Robust volatility estimation to semimartingale violations in high-frequency financial data
Authors: Torben Andersen - Northwestern University (United States)
Yingying Li - Hong Kong University of Science and Technology (Hong Kong)
Viktor Todorov - Northwestern University (United States)
Bo Zhou - Durham University Business School (United Kingdom) [presenting]
Abstract: The semimartingale assumption for the price process of an asset, traded in a frictionless market, is effectively a no-arbitrage condition. Recently, more and more evidence confirms the existence of violations of the semimartingale property for intraday periods of non-trivial duration. Such violations include, but are not limited to, gradual jumps and bursts in the volatility or drift component. We develop a volatility estimator of integrated volatility (IV), named differenced-increments volatility (DV), robust to these semimartingale violations, while other commonly-used estimators relying on the semimartingale assumption suffer non-trivial finite-sample biases. We document the reliability of our DV estimator in finite samples through a comprehensive Monte Carlo study. In our empirical application, we employ the DV estimator as the predictor to forecast realized volatility (RV). In an application for the S\&P 500 index and individual equities, we find that our DV-based Heterogeneous Autoregressive (HAR) model dominates popular competitors according to standard out-of-sample MSE and QLIKE criteria.