Tutorial 2: Writing a Financial Research Report

Learn how to generate professional financial research reports for A-shares and global markets.

Overview

The financial report pipeline generates institutional-grade research reports with:

Executive Summary
Financial Analysis (ROE, cash flows, DuPont analysis)
Valuation (DCF, comparable multiples)
Risk Assessment
Investment Recommendation

Pipeline Architecture

User Input (stock code, topic)
    │
    ├── TushareDataAgent ──→ A-share market data
    ├── YFinanceAgent ──────→ US stock data
    ├── MacroDataAgent ─────→ GDP, CPI, interest rates
    │
    ├── ParallelAnalystOrchestrator
    │   ├── FundamentalAnalyst ──→ Financial health
    │   ├── ValuationAnalyst ────→ DCF / Multiples
    │   ├── RiskAnalyst ─────────→ Risk factors
    │   ├── EarningsAnalyst ──────→ Earnings quality
    │   ├── CompetitiveAnalyst ───→ Competitive position
    │   └── MacroAnalyst ─────────→ Macro sensitivity
    │
    └── ResearchReportAgent ──→ Final report

Using TushareDataAgent

Note: TushareDataAgent requires TUSHARE_TOKEN in your .env file (not TUSHARE_API_KEY). Register at https://tushare.pro/register.

from scripts.core.analyst_agents import TushareDataAgent

agent = TushareDataAgent(default_ts_code="000001.SZ")

# A-share daily quotes
quote = agent.get_daily_quote(
    ts_code="000001.SZ",
    start_date="20230101",
    end_date="20241231"
)

# Index data
index_data = agent.get_index_data(
    ts_code="000001.SH",   # 上证指数
    start_date="20230101",
    end_date="20241231"
)

# Financial statements
fin_data = agent.get_financial_report(
    ts_code="000001.SZ",
    report_type="income"   # income / balance / cash_flow
)

# Margin data (融资融券); data_type: margin_detail / margin / short_margin
margin = agent.get_margin_data(data_type="margin_detail")

# Stock list (获取所有股票基本信息)
stocks = agent.get_stock_basic()

# Trading calendar
calendar = agent.get_trade_calendar(start_date="20240101", end_date="20241231")

# Concept stocks (concept board)
concept = agent.get_concept_stocks(board_name="AI语料")

All methods require TUSHARE_TOKEN to be set. When the token is not available, MCP calls gracefully return mock data (marked with _mock=True).

Running the Demo

Basic Demo

python scripts/demo_research_report.py --stock 000001.SZ

With Custom Output Directory

python scripts/demo_research_report.py \
    --stock 000001.SZ \
    --output papers

--output is a directory, not a file. A TeX file will be created at papers/demo_000001_SZ.tex. There is no --format flag — PDF compilation is attempted automatically (requires TeX Live installed).

Command-line Help

python scripts/demo_research_report.py --help

Report Structure

1. Executive Summary

One-page overview with: - Investment thesis (一句话结论) - Key financial highlights - Valuation and recommendation

2. Financial Analysis

Metric	Formula	Value
ROE	Net Income / Equity	12.5%
Net Profit Margin	Net Income / Revenue	35.2%
Asset Turnover	Revenue / Assets	0.45x
Financial Leverage	Assets / Equity	2.8x

3. Valuation

Three methods integrated: - DCF: Discounted cash flow with scenario analysis - Comparables: P/E, P/B, EV/EBITDA multiples - Dividend Discount Model: For dividend-paying stocks

4. Risk Factors

Industry-specific risks
Macroeconomic sensitivity
Policy risks
Operational risks

5. Recommendation

Rating	Criteria
Buy	Upside > 20%
Hold	-10% < Upside < 20%
Sell	Upside < -10%

Customizing Templates

Using Chinese Journal Templates

from scripts.journal_template import JournalTemplate, get_template

# 经济研究
template = get_template("经济研究")

# 金融研究
template = get_template("金融研究")

# 管理世界
template = get_template("管理世界")

List Available Templates

python scripts/journal_template.py --list

Generate Template File

python scripts/journal_template.py --generate JFE output/paper.tex

Programmatic Usage

import asyncio
from scripts.core.analyst_agents import ParallelAnalystOrchestrator
from scripts.core.llm_gateway import LLMGateway

async def analyze_stock(ticker: str):
    gateway = LLMGateway()
    orchestrator = ParallelAnalystOrchestrator(gateway=gateway)

    # Run parallel analysis (6 analysts concurrently)
    context = {
        "financial_data": {...},   # pre-fetched financial data
        "market_data": {...},      # pre-fetched market data
    }
    result = await orchestrator.run_parallel_analysis(
        ticker=ticker,
        context=context,
        analyst_types=["fundamental_financial", "valuation", "risk"],
    )

    print(f"Consensus: {result.consensus_view}")
    print(f"Confidence: {result.confidence:.2f}")
    return result

# Execute
result = asyncio.run(analyze_stock("000001.SZ"))

The run_parallel_analysis() method is async. Use asyncio.run() or await in an async context. Do not use orchestrator.analyze() or orchestrator.generate_report() — these methods do not exist.

Data Sources

Source	Data Type	API Key Required
Tushare	A-share quotes, financials, margin	Yes (`TUSHARE_TOKEN`)
akshare	A-share free data	No
yfinance	US stocks	No
EODHD	Macro indicators	Yes (`EODHD_API_KEY`)