Day/ Date/Time/ Venue: Wednesday, 13th December 2023, from 4:00 PM – 5:00 PM in AB2-CR102.
Title of the Talk: Models for High Frequency Time Series.
Speaker: Prof. Nalini Ravishanker, Professor, Department of Statistics, University of Connecticut.
Abstract: We describe modeling high-frequency time series such as intra-day transaction level asset prices which are irregularly spaced in time within a trading day. The size of the gaps between successive transactions can vary depending on the asset features such as liquidity. We discuss univariate and multivariate stochastic volatility models for irregularly-spaced time series, distinguishing between scenarios where the gaps between transactions (events) are assumed to be either fixed or random. We then describe a hierarchical model where the random gaps follow an autoregressive conditional duration (ACD) model, and the log-returns follow an irregular stochastic volatility (SV) model. This IR-SV-ACD model is useful for estimating and forecasting inter-transaction gaps as well as the volatility of log-returns. For model fitting in the Bayesian framework, we employ the Hamiltonian Monte Carlo (HMC) algorithm through the R package cmdstanr. We demonstrate the accuracy and computing speed on simulated data and healthcare stocks traded on the NYSE. We also show how we can extend this approach to model multiple stocks using IR-MSV-ACD models that we have developed.
This is joint work with Sreeram Anantharaman (Statistics, UConn), Sumanta Basu (Statistics and Data Science, Cornell), and Chiranjit Dutta (eBay).
Speaker’s webpage: https://nalini-ravishanker.scholar.uconn.edu/