ldy26098
/
BMAD-METHOD
mirror of https://github.com/bmad-code-org/BMAD-METHOD.git
1
0
Fork 0
Code Packages Projects Releases Wiki Activity
BMAD-METHOD/README.md

7.6 KiB
Raw Blame History

Quant Method

v1.0.0-alpha.1 -- Early access. APIs and workflows are evolving.

AI-Driven Investment Quant Development and Research Framework -- Specialized AI agents and structured workflows for systematic strategy research, backtesting, risk management, and production deployment.

Built on the BMAD Method, adapted for quantitative finance.

Why Quant Method?

Quantitative investment research demands rigor at every stage -- from hypothesis formation through statistical validation to live monitoring. Traditional tools leave gaps between research notebooks and production systems. Quant Method fills those gaps with AI agents that act as expert collaborators across the entire quant lifecycle.

Key Capabilities

  • Bias Prevention -- Agents enforce point-in-time data correctness, flag look-ahead bias in vectorized operations, and require out-of-sample validation before any strategy advances
  • Statistical Rigor -- Walk-forward analysis, significance testing, and reproducibility requirements (random seeds, data versioning, parameter logging) are built into every workflow
  • Structured Research Process -- Guided workflows grounded in quantitative finance best practices across research, strategy design, validation, and production
  • Quant-Adaptive Depth -- Adjusts rigor based on strategy complexity: a simple momentum factor needs different treatment than a multi-asset statistical arbitrage system
  • Full Lifecycle Coverage -- From alpha research and factor discovery through backtesting, risk decomposition, and live performance attribution

Quick Start

Prerequisites: Node.js v20+

npx quant-method install

Follow the installer prompts, then open your AI IDE (Claude Code, Cursor, Windsurf, etc.) in the project folder.

Workflow Paths

Rapid Strategy Path (Quick Flow)

For quick hypothesis testing, single-factor strategies, and clear signals:

  1. /strategy-spec -- Analyze your data and produce a strategy specification with implementation tasks
  2. /dev-strategy -- Implement each task (signals, backtest, risk checks)
  3. /strategy-review -- Validate statistical rigor and code quality

Full Research Path (Quant Method)

For multi-factor strategies, portfolio-level research, and production deployment:

Step Command What Happens
1 /research-brief Define investment thesis, universe, and data requirements
2 /create-strategy-design Full spec: signal definitions, risk constraints, performance targets
3 /create-architecture Technical infrastructure: data pipelines, execution, monitoring
4 /create-research-plan Break work into prioritized research and implementation tasks
5 /research-planning Initialize research tracking and sequencing
6 Per task: /create-task Prepare individual research or implementation tasks
7 Per task: /dev-task Execute tasks (signal implementation, backtesting, risk checks)
8 Per task: /task-review Validate statistical correctness and code quality

Research Lifecycle

Phase 1: Research

  • Market and academic literature review
  • Factor discovery and screening
  • Data exploration and alternative data evaluation
  • Investment thesis development

Phase 2: Strategy Design

  • Signal specification and universe selection
  • Risk constraint definition (drawdown limits, VaR/CVaR targets)
  • Performance target setting (Sharpe, information ratio, alpha decay)
  • Model specification -- statistical, ML, or rules-based

Phase 3: Validation

  • Backtesting with walk-forward analysis
  • Out-of-sample testing (mandatory, not optional)
  • Statistical significance validation
  • Transaction cost, slippage, and capacity analysis
  • Risk decomposition and stress testing
  • Regime-conditional performance review

Phase 4: Production

  • Deployment and system integration
  • Live monitoring and alerting
  • Performance attribution
  • Research retrospective and strategy refinement

Specialized Agents

Each agent brings domain-specific expertise to the research process:

Agent Persona Role Focus
Quant Researcher Elena Alpha Research + Factor Analysis Signal discovery, literature review, statistical analysis
Portfolio Manager Catherine Portfolio Construction + Allocation Position sizing, rebalancing, benchmark-aware optimization
Quant Architect Victor Systems Design + Infrastructure Data pipelines, execution systems, backtesting frameworks
Quant Developer Marcus Strategy Implementation Signal code, backtest harnesses, production adapters
Data Engineer Priya Market Data + Alternative Data Data pipelines, quality validation, feature engineering
Risk Analyst Quinn Risk Management + Model Validation Drawdown analysis, stress testing, regime detection
Research Director David Research Process + Coordination Pipeline management, prioritization, tracking
Research Documentarian Paige Research Reports + Documentation Strategy docs, research logs, compliance records
Strategy Developer Barry Rapid Prototyping Quick hypothesis testing, single-factor research

Built-In Safeguards

Agents enforce quant best practices throughout the workflow:

  • No look-ahead bias -- Vectorized operations must respect temporal ordering
  • Point-in-time data correctness -- Non-negotiable for backtesting pipelines
  • Reproducibility -- Random seeds, data versioning, and parameter logging required
  • Out-of-sample validation -- Every strategy must pass before advancing
  • Realistic assumptions -- Transaction costs, slippage, and capacity constraints must be modeled
  • Risk metrics -- Drawdown, VaR, CVaR, and regime-conditional analysis required

What You Can Build

  • Single-factor momentum or value strategies using the Quick Flow path
  • Multi-factor equity models with statistical validation and risk decomposition
  • Statistical arbitrage systems with cointegration testing and regime detection
  • Alternative data signal pipelines with feature engineering and quality validation
  • Portfolio construction frameworks with position sizing and rebalancing logic
  • Backtesting infrastructure with walk-forward analysis and transaction cost modeling

Project Structure

src/
  bmm/              Quant Method module (agents, workflows, data)
    agents/          9 specialized quant agents
    workflows/       4-phase research lifecycle + quick flow
    teams/           Agent team configurations
    data/            Templates and context files
  core/              Core framework (shared agents, resources)
  utility/           Shared utilities and components

tools/
  cli/               Command-line interface
  schema/            Agent schema validation
  flattener/         Document processing tools

docs/                Tutorials, guides, and reference

Documentation

Resource Description
Getting Started Installation and first strategy walkthrough
Quick Flow Guide Rapid prototyping methodology
Workflow Map Complete visual workflow reference
Customization Adapting agents and workflows

Contributing

We welcome contributions. See CONTRIBUTING.md for guidelines.

License

MIT License -- see LICENSE for details.