CMStatistics 2022: Start Registration
View Submission - CFE
A1905
Title: Machines do not go for lunch: A new diurnal adjustment for trade durations Authors:  Markus Belfrage - Hanken School of Economics (Finland) [presenting]
Abstract: A new diurnal adjustment method is proposed for stock trade durations. A well-known feature of stock markets is the diurnal seasonality in the intensity of trading. Trade durations are often modelled by the class of Autoregressive Conditional Duration (ACD) models, where it is assumed that the seasonality factor acts multiplicatively on all durations. We show that this assumption is violated for ultra-high precision trade data when a large portion of the trades are executed by computer algorithms. A two-component mixture model with features that vary nonparametrically over time of the day is developed as a response to the heterogeneity in the diurnal seasonality caused by the mix of machines and regular traders. Furthermore, an estimation algorithm much in the flavor of expectation maximization is proposed and applied to a large set of Apple (AAPL) trades using data from Nasdaq Historical TotalView ITCH.