Can the model update its parameters in real-time?

Yes, it includes an OnlineHawkesEstimator that performs stochastic gradient updates to parameters like baseline intensity as new fills are observed.

How does this skill improve upon standard kappa estimation?

Standard kappa often uses simple frequency-based averages. This skill incorporates exchange state, queue position, and self-excitation, leading to much more accurate real-time fill predictions.

Is this model specific to the Hyperliquid exchange?

While it includes specific extensions for Hyperliquid features like funding and open interest, the core Hawkes architecture can be adapted for any central limit order book exchange.

What is a Hawkes process in the context of trading?

A Hawkes process is a self-exciting point process where past events (trades) increase the probability of future events, effectively modeling how trades cluster together in time.

How do I validate the accuracy of the fill model?

The skill includes validation methods to compute Brier scores and calibration curves, ensuring your fill predictions align with actual market outcomes.

Hawkes Process Fill Modeling

Name: Hawkes Process Fill Modeling
Author: trudumb

bytrudumb

0•

データサイエンスとML

Models trade fill intensity using Hawkes processes to provide sophisticated liquidity and probability estimates for high-frequency trading.

This skill implements a specialized probabilistic framework for modeling market fill rates, moving beyond static kappa estimation. By leveraging self-exciting Hawkes processes, it captures the temporal clustering of trades while incorporating Hyperliquid-specific variables like funding rates and open interest. It is designed for quantitative traders and market makers who need to accurately predict fill probabilities, optimize queue positions, and dynamically adjust quoting strategies based on real-time exchange state and trade dynamics.

主な機能

01State-dependent baseline intensity incorporating funding and OI

02Real-time online parameter estimation and MLE batch retraining

030 GitHub stars

04Automated conversion to GLFT-compatible kappa values

05Queue-position-aware kernels for depth-sensitive fill estimation

06Self-exciting Hawkes process dynamics to capture trade clustering

ユースケース

01Predicting fill probabilities for orders placed deep in the order book

02Upgrading market-making bots from static fill-rate models to dynamic probabilistic ones

03Adjusting quoting spreads based on funding rate cycles and volatility regimes

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add trudumb/hyper_make models

For use in Claude.ai and ChatGPT