How it works and its limits
- Inputs
- The latest 5000 market trades of one venue (or all venues) from ArcDelta: time in milliseconds, aggressor side and size.
- Outputs
- Branching ratio n (total, buy to buy, buy to sell), memory half-life, current intensity against the baseline, expected trades over 10 s to 5 min, and the fit check: time-change residuals with a KS test and QQ plot, out-of-sample likelihood against Poisson, and trades per 2 s against Hawkes and Poisson.
- Evidence
- On simulated flow with known parameters the fit recovers the branching ratio within 0.15, the residuals pass the KS test and a plain Poisson fails it. A live check on real trades appears on this page for every run.
- Limits
- Fitted on trades only: limit orders and cancels are not stored any more, so they are not modelled. The window is the latest 5000 trades (minutes for BTC), the daily profile comes from that window, and a regime change inside it blurs n. Descriptive metric, not a trading signal.
- Minimum data
- 400 recent trades.
- API
GET /labs/hawkes?symbol=BTCUSDT&exchange=binance