Introduction

Learn about NanoARB, a production-grade high-frequency trading engine for CME futures markets

What is NanoARB?

NanoARB is a nanosecond-level high-frequency trading engine designed for CME futures markets (ES, NQ). It combines cutting-edge machine learning models with ultra-low-latency Rust infrastructure to achieve institutional-grade performance for statistical arbitrage and market-making strategies.

Built entirely in Rust with zero Python at runtime, NanoARB delivers sub-microsecond tick-to-trade latency while maintaining production-grade reliability and realistic backtesting capabilities.

Key features

Sub-microsecond inference

< 800ns end-to-end tick-to-trade latency with median total latency of 780ns

Production Rust codebase

Zero Python at runtime, built for maximum performance and reliability

State-of-the-art ML

Mamba State Space Models for 10-50x faster inference than Transformers

Realistic backtesting

Latency simulation, queue position modeling, and adverse selection

Performance benchmarks

Measured on AMD EPYC 7763 (AWS c6a.8xlarge):

Operation	Median	P95	P99
LOB Update	45ns	62ns	78ns
Feature Extraction	120ns	145ns	168ns
Model Inference	580ns	720ns	890ns
Total Tick-to-Trade	780ns	950ns	1.2μs

Trading capabilities

Data Pipeline

• CME MDP 3.0 binary protocol parser with SBE encoding
• Zero-copy message parsing using nom
• Synthetic data generator for development and testing
• Historical replay support

Order Book Engine

• 20-level price aggregation with O(log n) updates using BTreeMap
• Real-time feature extraction: Microprice, OFI, VPIN, Book Imbalance
• Tensor serialization for ML inference

ML Models

• Mamba-LOB: State Space Model for sequence modeling (~500K parameters, <800ns inference)
• Decision Transformer: Offline RL for market-making
• IQL: Implicit Q-Learning with expectile regression
• ONNX export for Rust inference via ort

Backtesting

• Event-driven architecture with configurable latency models
• Realistic fill simulation with queue position tracking
• Walk-forward and purged cross-validation

Architecture overview

NanoARB is built as a modular Rust workspace with the following crates:

Crate	Description
nano-core	Core types, traits, error handling
nano-feed	CME MDP 3.0 parser, synthetic data generator
nano-lob	Order book reconstruction, feature extraction
nano-model	ONNX inference, signal generation
nano-backtest	Event-driven backtesting engine
nano-strategy	Trading strategies, RL environment
nano-gateway	Entry point, metrics, configuration

┌─────────────────────────────────────────────────────────────────────┐
│                         NanoARB Architecture                        │
├─────────────────────────────────────────────────────────────────────┤
│                                                                     │
│  ┌──────────────┐    ┌──────────────┐    ┌──────────────┐           │
│  │  CME MDP 3.0 │───▶│  nano-feed   │───▶│   nano-lob   │           │
│  │  Market Data │    │   Parser     │    │  Order Book  │           │
│  └──────────────┘    └──────────────┘    └──────┬───────┘           │
│                                                  │                  │
│                                                  ▼                  │
│  ┌──────────────┐    ┌──────────────┐    ┌──────────────┐           │
│  │  nano-model  │◀───│   Features   │◀───│   Snapshot   │           │
│  │ ONNX Infer.  │    │  Extraction  │    │ Ring Buffer  │           │
│  └──────┬───────┘    └──────────────┘    └──────────────┘           │
│         │                                                           │
│         ▼                                                           │
│  ┌──────────────┐    ┌──────────────┐    ┌──────────────┐           │
│  │nano-strategy │───▶│nano-backtest │───▶│ nano-gateway │           │
│  │  MM / Signal │    │   Engine     │    │   Metrics    │           │
│  └──────────────┘    └──────────────┘    └──────────────┘           │
│                                                                     │
└─────────────────────────────────────────────────────────────────────┘

Who should use NanoARB?

NanoARB is designed for:

• Quantitative researchers developing high-frequency trading strategies
• Algorithmic traders who need realistic backtesting with sub-microsecond precision
• HFT developers building low-latency trading systems in Rust
• Academic researchers studying market microstructure and ML for trading
• Trading firms evaluating market-making and statistical arbitrage strategies

Educational and research purposes only

This software is for educational and research purposes only. It is not financial advice. Trading futures involves substantial risk of loss. Past performance does not indicate future results.

Next steps

Quickstart

Get NanoARB up and running in minutes

Installation

Detailed installation instructions and requirements