A0348
Title: On predicting the likelihood of high-frequency extreme price movements
Authors: Julien Hambuckers - University of Liege (Belgium) [presenting]
Philippe Hubner - HEC Liege, University of Liege (Belgium)
Abstract: In finance, the ability to better understand and predict high-frequency extreme price movements (EPM) is crucial for market stability and investor confidence. This is, however, a challenging task, due to the complexity surrounding EPM identification and the numerous predictors available. Extreme value theory is exploited to bypass some of these issues: Rather than focusing on the EPM, the time-varying characteristics and determinants of their distribution are studied. To do so, it is assumed that the distribution for the maxima of high-frequency negative returns over small time intervals can well approximate it. Thus, these maxima are specified for: A parametric generalized autoregressive score model with a generalized extreme value response distribution. To perform covariate selection, a L0-penalized likelihood procedure is introduced. The properties of the estimation and selection procedures are investigated in an extensive Monte Carlo simulation, highlighting good finite-sample performance. Empirically, limit order book data is used from 2009 to 2019 for 50 NASDAQ-traded stocks and a large set of liquidity measures. It is shown that the model performs better than other benchmark specifications, mainly due to the incorporation of important conditioning variables. From the variable selection procedure, it is shown that liquidity measures play a significant role in determining the distribution of future EPM, along with daily uncertainty indicators.
