Autoregressive conditional duration modelling of high frequency data

TL;DR

This paper investigates duration modeling of high-frequency financial data using the ACD framework, finding Gamma distributions best fit SPY durations and improving understanding of duration dynamics.

Contribution

It evaluates different distribution assumptions within the ACD model, identifying Gamma distribution as most suitable for SPY high-frequency durations.

Findings

Gamma distribution best fits SPY durations

ACD with Gamma innovations outperforms other distributions

Unconditional durations tend toward Gamma distribution

Abstract

This paper explores the duration dynamics modelling under the Autoregressive Conditional Durations (ACD) framework (Engle and Russell 1998). I test different distributions assumptions for the durations. The empirical results suggest unconditional durations approach the Gamma distributions. Moreover, compared with exponential distributions and Weibull distributions, the ACD model with Gamma distributed innovations provide the best fit of SPY durations.