japan-trading-agents 0.1.0

Multi-agent AI trading analysis for Japanese stocks — powered by the Japan Finance Data Stack

japan-trading-agents

Multi-agent AI trading analysis for Japanese stocks — powered by real government data, not just LLM reasoning.

What is this?

9 AI agents analyze Japanese stocks using real financial data from government and market data sources:

Agent	Role	Data Source
Fundamental Analyst	Financial statements, DuPont analysis	EDINET (有報)
Macro Analyst	GDP, CPI, interest rates	e-Stat + BOJ
Event Analyst	Earnings, dividends, M&A	TDNet (適時開示)
Sentiment Analyst	News sentiment scoring	Japan News RSS
Technical Analyst	Price action, volume	Yahoo Finance (yfinance)
Bull Researcher	Builds the bullish case	All analyst reports
Bear Researcher	Challenges with risks	All analyst reports
Trader	BUY/SELL/HOLD decision	Debate + analysis
Risk Manager	Risk validation	Final approval

How it works

jta analyze 7203  (Toyota)
      |
      v
  [Data Fetch] ── yfinance + 5 MCP sources in parallel
      |
      v
  [5 Analysts] ── Fundamental, Macro, Event, Sentiment, Technical (parallel)
      |
      v
  [Bull vs Bear Debate] ── Sequential argumentation
      |
      v
  [Trader Decision] ── BUY / SELL / HOLD with confidence
      |
      v
  [Risk Manager] ── Approve or reject with concerns

Quick Start

# Install with all data sources
pip install "japan-trading-agents[all-data]"

# Set your LLM API key
export OPENAI_API_KEY=sk-...

# Analyze a stock
jta analyze 7203

Use Any LLM

Powered by litellm — supports 100+ LLM providers:

# OpenAI
jta analyze 7203 --model gpt-4o

# Anthropic
jta analyze 7203 --model claude-sonnet-4-5-20250929

# Google
jta analyze 7203 --model gemini/gemini-2.0-flash

# Local (Ollama)
jta analyze 7203 --model ollama/llama3.2

# Any litellm-supported model
jta analyze 7203 --model deepseek/deepseek-chat

CLI Commands

# Full analysis
jta analyze 7203

# With EDINET code override
jta analyze 7203 --edinet-code E02144

# Multi-round debate
jta analyze 7203 --debate-rounds 2

# JSON output
jta analyze 7203 --json-output

# Check data sources
jta check

# MCP server mode
jta serve

Data Sources

Each data source is an independent MCP package. Install only what you need:

pip install japan-trading-agents                        # Core only
pip install "japan-trading-agents[edinet]"              # + EDINET
pip install "japan-trading-agents[all-data]"            # All 6 sources

Source	Package	API Key Required
Yahoo Finance (stock prices)	yfinance (bundled)	No
EDINET (financial statements)	edinet-mcp	Yes (free)
TDNet (disclosures)	tdnet-disclosure-mcp	No
e-Stat (government statistics)	estat-mcp	Yes (free)
BOJ (central bank data)	boj-mcp	No
News (financial news RSS)	japan-news-mcp	No

The system gracefully degrades — agents work with whatever sources are available.

Architecture

• No LangChain/LangGraph — pure asyncio for orchestration
• litellm for multi-provider LLM support (single interface, 100+ providers)
• Pydantic models for structured agent outputs
• Rich CLI with streaming progress
• FastMCP server for MCP integration
• All LLM calls are mocked in tests — no API keys needed to run tests

Part of the Japan Finance Data Stack

Layer	Tool	Description
Corporate Filings	edinet-mcp	XBRL financial statements
Disclosures	tdnet-disclosure-mcp	Real-time corporate disclosures
Government Statistics	estat-mcp	GDP, CPI, employment
Central Bank	boj-mcp	Interest rates, money supply
Financial News	japan-news-mcp	RSS news aggregation
Stock Prices	yfinance	Yahoo Finance (TSE coverage)
Benchmark	jfinqa	Japanese financial QA benchmark
Analysis	japan-trading-agents	Multi-agent trading analysis

See awesome-japan-finance-data for a complete list of Japanese finance data resources.

Development

git clone https://github.com/ajtgjmdjp/japan-trading-agents
cd japan-trading-agents
uv sync --extra dev
uv run pytest -v
uv run ruff check src tests
uv run mypy src

Disclaimer

This is not financial advice. For educational and research purposes only. Do not make investment decisions based on this tool's output.

License

Apache-2.0