Trade Arrival Clustering

Article FMZ digest · Author: 小草

Summary

This analysis studies aggregated trades for a newly listed crypto perpetual contract, focusing on buy-trade arrival intervals and counts within short time buckets. It compares observed behavior with a Poisson process, whose assumptions imply a stable event rate and independent arrivals. The observed interval and count distributions differ from that model: longer gaps and unusual bursts occur more often than the Poisson fit suggests. The author points to changing activity rates and interactions between orders as possible explanations, including clustered trading and periodic activity that may reflect automated execution.

The proposed practical response is to update market parameters instead of relying on fixed assumptions. The article plots rolling order-count and traded-quantity averages, then evaluates forecasts against subsequent observations using absolute residuals. In the reported sample, an exponentially weighted estimate improved on a simple mean, and a short recent history performed better than the baseline. This is an exploratory single-market example; it does not establish a profitable trading rule, and the author presents the forecasting method as a simple starting point for more detailed time-series and volatility-clustering analysis.

Key ideas

A Poisson arrival model assumes a stable event rate and independent events, assumptions that may not fit real trade data.
The examined trade intervals and short-window counts show clustering and more extreme activity than the Poisson comparison predicts.
Order frequency and traded quantity can vary together, so trading parameters may need continual updating.
Rolling and exponentially weighted averages offer simple ways to adapt estimates to recent market activity.
Forecast residuals can compare candidate estimates, but this exploratory analysis does not demonstrate strategy profitability.

Trade Arrival Clustering and Adaptive Parameters in Crypto Markets

Summary

Key ideas

Tags