Classifying Human and High-Frequency Orders by Replacement Time

Summary

This study proposes distinguishing human-originated orders from high-frequency algorithmic orders using the time taken to modify an order before execution. Orders with a minimum or average replacement time below a selected threshold are labeled algorithmic; those above it are labeled human. The approach is evaluated against historical tick-by-tick records that already identify orders as algorithmic or non-algorithmic, with a confusion matrix used to assess classification.

The reported pattern is that orders labeled algorithmic are more often identified as such when average replacement time is under one second, with that share declining as the time grows. The human-order classification rate is described as comparatively steady, between 36% and 47%, and the article notes a small change around a ten-second threshold. These findings suggest that slower algorithms can be mistaken for human activity. The author cautions that the observations are not final and may change as market behavior adapts; the document does not provide broader validation across markets or datasets.

Key ideas

• Order replacement time is used as a heuristic to separate human orders from high-frequency algorithmic orders.
• The study compares threshold-based labels with historical tick data carrying preexisting order classifications.
• Algorithmic orders are more often identified at short replacement times, with the rate declining as replacement time increases.
• The reported human-order classification share stays within a stated range, suggesting some algorithms operate on slower timescales.
• The results are preliminary and may change with market conditions or new data.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.