How informative is high-frequency data for tail risk estimation and forecasting? An intrinsic time perspectice

This paper proposes a novel and simple approach to compute daily Value at Risk (VaR) and Expected Shortfall (ES) directly from high-frequency data. It assumes that financial logarithm prices are subordinated unifractal processes in the intrinsic time, which stochastically transforms the clock time in accordance with the markets activity. This is a very general assumption that allows for a simple computation of daily VaR and ES by scaling up their intraday counterparts computed from data sampled in intrinsic time. In the empirical exercise, we discuss the statistical and dynamic properties of the resulting daily VaR and ES estimates and show that our method outperforms standard ones in accurately estimating and forecasting VaR and ES.