Taming impulsive high-frequency data using optimal sampling periods

Optimal sampling period selection for high-frequency data is at the core of financial instruments based on algorithmic trading. The unique features of such data, absent in data measured at lower frequencies, raise significant challenges to their statistical analysis and econometric modelling, especially in the case of heavy-tailed data exhibiting outliers and rare events much more frequently. To address this problem, this paper proposes a new methodology for optimal sampling period selection, which better adapts to heavy-tailed statistics of high-frequency financial data. In particular, the novel concept of the degree of impulsiveness (DoI) is introduced first based on alpha-stable distributions, as an alternative source of information for characterising a broad range of impulsive behaviours. Then, a DoI-based generalised volatility signature plot is defined, which is further employed for determining the optimal sampling period. The performance of our method is evaluated in the case of risk quantification for high-frequency indexes, demonstrating a significantly improved accuracy when compared against the well-established volatility-based approach.