ABIDES Rust

A high-performance, Rust-only implementation of an agent-based market simulator for market microstructure research, execution strategy testing, and privacy-preserving federated learning experiments.

Project Summary

This project focuses on building a production-grade simulation core in Rust with deterministic behavior, reproducible experiments, and configurable large-scale multi-agent scenarios.

It is designed to showcase:

• Systems engineering for event-driven simulation
• Quant/market microstructure modeling
• Multi-agent architecture and message protocols
• Research-oriented experiment design and benchmarking

Why This Is Portfolio-Ready

• End-to-end discrete-event kernel with deterministic scheduling
• Rich library of trading and market agents (exchange, MM, execution, RL, ETF arbitrage, PPFL)
• Realistic simulation scenarios from toy demos to stress-style market environments
• Reproducible runs with seed control and environment-based configuration
• Test suite and data/report artifacts for analysis workflows

Tech Stack

• Rust 2021
• Key crates: chrono, clap, serde, serde_json, csv, rand, rand_distr, bzip2, sha2, x25519-dalek, serde-pickle

Architecture

• src/main.rs: CLI entrypoint and simulation orchestration
• src/kernel.rs: event queue, simulation clock, message dispatch
• src/message.rs: inter-agent message protocol
• src/order.rs, src/order_book.rs: order model and matching logic
• src/trading_agent.rs: shared trading-agent behavior
• src/config.rs: simulation builders and environment parsing
• src/oracle.rs: fundamental value oracles (sparse/mean/data/external)
• src/*_agent.rs: strategy-specific agents

Getting Started

Prerequisites

• Rust toolchain (rustup, cargo)
• Latest stable Rust recommended

Setup

git clone <your-repository-url>
cd <your-repository-folder>
cargo check
cargo test

CLI Usage

cargo run -- --help
cargo run -- --config-help

Main options:

• -c, --config <NAME>: simulation configuration
• -s, --seed <U64>: global seed
• -l, --log_dir <DIR>: output log directory
• -v, --verbose: verbose runtime logs
• --config-help: list available configs

Available Configurations

Core / Demo

• sum
• lob_demo
• value_noise
• value_noise_mean
• value_noise_data
• sparse_zi_100
• sparse_zi_1000

Market Simulation

• random_fund_value
• random_fund_diverse
• execution
• exp_agent_demo
• impact
• obi_rmsc02
• loop_obi
• qlearning
• marketreplay
• parallel

Research Scenarios

• rmsc01
• rmsc02
• rmsc03
• twoSymbols
• hist_fund_value
• hist_fund_diverse
• ppfl_template
• ppfl_icaif20

Quick Run Examples

# Kernel sanity check
cargo run -- -c sum

# Execution scenario (TWAP default)
ABIDES_EXECUTION_TRADE=true cargo run -- -c execution

# Execution scenario with VWAP
ABIDES_EXECUTION_STRATEGY=VWAP ABIDES_EXECUTION_TRADE=true cargo run -- -c execution

# Historical fundamentals from external .bz2 source
ABIDES_HIST_FUND_FILE=data/synthetic_fundamental.bz2 cargo run -- -c hist_fund_value

# PPFL template scenario
cargo run -- -c ppfl_template

Environment Configuration

Scenarios are controlled through ABIDES_* environment variables.

Common groups:

• ABIDES_TICKER
• ABIDES_HIST_FUND_FILE
• ABIDES_EXECUTION_*
• ABIDES_EXP_AGENT_*
• ABIDES_RMSC01_*, ABIDES_RMSC02_*, ABIDES_RMSC03_*
• ABIDES_PPFL_*
• ABIDES_QLEARNING_*
• ABIDES_TWOSYMBOLS_*

To extract the complete, source-of-truth list from code:

rg -o "ABIDES_[A-Z0-9_]+" src/config.rs src/main.rs src/ppfl_client_agent.rs src/ppfl_service_agent.rs | sort -u

Output Artifacts

Primary outputs:

• Run logs under log/ (or --log_dir)
• Analysis/result files under results/

PPFL artifacts:

• results/all_results.csv
• results/timing_log.csv
• results/collusion_weights.csv (when enabled)
• results/collusion_selected_noise.csv (when enabled)
• results/collusion_consensus.csv (when enabled)

Data / Oracle Support

ExternalFileOracle supports:

• Native CSV
• Native .bz2 CSV
• Native .bz2 pandas-Series pickle decoding

Current limitation:

• pandas DataFrame pickle support is not yet available in Rust-only mode.

Quality & Validation

cargo fmt
cargo check
cargo test

Reproducibility

• Explicit run seeds via --seed
• Deterministic-friendly configuration model
• Runtime parameters centralized in src/config.rs

Current Limitations

• Some scenarios remain compatibility-first relative to historical implementations.
• External pickle support is currently focused on pandas Series.

Roadmap

• Add pandas DataFrame pickle support in the external oracle loader
• Expand benchmark and performance reporting pipeline
• Increase test coverage for long-horizon, high-agent-count scenarios

License

BSD-3-Clause. See LICENSE.txt.