Fuji: The Shape, Composition, and Spread of Demand Bursts on Robinhood Chain (arXiv (Cornell University))

Summary

• The marketplace operates at a usual baseline gas level of 16.1 Mgas s−1, while 656 identified burst intervals account for 6.1% of minutes and 12.2% of total gas.p.5
• Compared to matched quiet control periods, a demand burst experiences 1.05 times the pools, 1.40 times the swaps, and 1.68 times the gas usage.p.11
• Trader entry clustering within burst pools exhibits an excess autocorrelation that decays exponentially with a time constant of 23.7 seconds.p.15

Key insights

1. The median peak demand inside a burst reaches 2.5 times the usual marketplace load.p.1
2. A two-exponential model demonstrates that the slow timescale of demand build-up is about 60 times the fast timescale.p.3
3. After adjusting for the daily demand cycle, the ratio between the slow and fast burst timescales decreases to about 40.p.8
4. Trading inside burst windows displays significant asset rotation, with 43.7% of trading occurring in pools that were absent at the window start.p.10
5. On weekdays, 54.8% of bursts starting within 30 minutes of the New York stock market opening are led by equity tokens.p.10
6. Excess gas usage in bursts relative to control periods decomposes into 1.21 times senders, 1.26 times transactions per sender, and 1.10 times gas per transaction.p.11
7. The arrival intervals between consecutive bursts fit an external cascade model better than a self-excitation Hawkes process model.p.14
8. Trader entries into individual trading pools cluster beyond common trigger expectations, decaying over a 23.7 second timescale.p.15