Affiliations: [a] Department of Statistics, Carnegie Mellon University, Pittsburgh, PA, USA | [b] Savvysherpa, Inc., Minneapolis, MN, USA
Corresponding author: Grant B. Weller, 6200 Shingle Creek Pkwy, Suite 400, Minneapolis, MN 55430, USA. Tel.: +1 3202487380; E-mail:[email protected]
Abstract: Quantifying the nature of extreme value dependence in high-frequency fluctuations of asset prices is an important yet difficult problem.
In this work, we propose a two-stage estimation procedure for conditional joint distribution of high-frequency extremes, given past information on returns.
The model combines an intraday volatility component and GARCH model for marginal time dependence with a tail dependence model for extreme values which is based on the framework of regular variation.
Examining 15-second returns of four banking sector securities, we find that there exists tail dependence in the detrended residuals.
The proposed model outperforms a benchmark Gaussian model in predicting conditional value-at-risk and expected shortfall, as well as in predicting the probability of jointly extreme returns.
Keywords: Extreme value theory, heavy tails, high-frequency returns, portfolio risk, tail dependence, time series