Statistical Methods in Finance 2023

December 19 to 23, 2023



Modeling Irregularly-spaced High Frequency Financial Time Series

By Sreeram Anantharaman
University of Connecticut, USA

Stochastic volatility (SV) models are widely used for volatility modeling in regularly spaced time series, such as daily stock returns. Usually, high-frequency financial data arrive at irregular time intervals, leading to unequal durations between consecutive data points. The data might be quickly spaced with short durations between arrivals or have slower arrivals with longer durations between data points. To effectively model such irregular data, we propose to consider the irregularity/gaps as random variables that are modeled using a autoregressive conditional duration (ACD) model. This forms the basis for building a hierarchical irregular stochastic volatility autoregressive conditional duration (IR-SV-ACD) model for estimating and forecasting inter-transaction gaps and volatility of log-returns. For multiple stocks, we extend the IR-SV-ACD model to the IR-MSV-ACD model and describe estimation and forecasting. To illustrate this approach, we use intraday prices available at microseconds level of health stocks traded on the NYSE. The log-returns and gaps are calculated for the stocks and are used for modeling in both univariate and multivariate settings. The parameter estimation is carried out using Bayesian analysis with Hamiltonian Monte Carlo in R using cmdstanr package. This is joint work with Nalini Ravishanker (Statistics, UConn) and Sumanta Basu (Statistics and Data Science, Cornell).