financial-statement-analysis

Before anything else — handle the data source

The quality of the analysis depends entirely on whether the underlying numbers are trustworthy. Do this first, before writing a single ratio.

If a PDF URL was provided

Attempt web_fetch on the URL. If the fetch succeeds and returns readable text, extract the core financial tables. If the fetch fails or returns empty/garbled content (common with exchange servers like SSE, SZSE, HKEX):

1. Note clearly: "PDF could not be fetched directly — using secondary source"
2. Search for the filing data via web_search using company name + year + "年度报告" + "主要财务数据"
3. Fetch a reliable secondary source (Sina Finance, 东方财富, official company IR page)
4. Clearly label every number with its source: [年报原文] or [二手来源: XXX]

Never silently blend primary and secondary sources. If a number came from a news article rather than the filing itself, say so.

Tag every data point before using it

Tag	Meaning	Rule
[直接]	Appears verbatim in the filing	Can be stated as fact
[推算]	Computed from filing numbers	Must show the formula inline
[推断]	Inferred from context, not stated	Must be flagged or dropped
[二手]	From secondary source, not filing	Must cite the source name

[推断] items must never be presented as facts. Either drop them or explicitly flag: "根据上下文推断，未经年报直接披露."

---

Step 1 — Normalize the input

• Confirm reporting scope: consolidated vs standalone, annual vs interim
• Confirm accounting standard: PRC GAAP / IFRS / US GAAP
• Confirm unit and currency — unit mismatches (万元 vs 元) are the most common arithmetic error
• Note any major M&A, disposals, or accounting policy changes that break comparability
• If source is PDF or HTML, extract core rows into a structured table before analysis

---

Step 2 — Mandatory data extraction checklist

Collect these fields before writing any analysis. Mark — if genuinely missing; do not estimate.

Income statement (3 years where available)

Field	Tag	Notes
营业收入	[直接]
营业成本	[直接]
毛利率	[直接] or [推算: (rev-cogs)/rev]
归母净利润	[直接]	Required
扣非净利润	[直接]	Required — always extract both
非经常性损益合计	[直接]	Explain gap if |np − np_ex| > 10% of np
营业利润	[直接]
研发/销售/管理/财务费用	[直接]

Cash flow statement

Field	Tag	Notes
经营活动现金流净额 (CFO) — all years	[直接]	Record trend, not just latest year
资本开支 (capex)	[直接] or [推算]	For FCF
FCF	[推算: CFO − capex]
CFO / 归母净利润	[推算]	Quality check: sustained < 1× is a concern

Balance sheet

Field	Tag
总资产	[直接]
总负债	[直接]
归属股东净资产	[直接]
货币资金	[直接]
存货	[直接]
应收账款	[直接]
有息负债（短期借款+长期借款）	[直接]
资产负债率	[直接] or [推算: 负债/总资产]

---

Step 3 — Three-statement sanity checks

Run these before drawing any conclusions:

1. Balance sheet identity: 总资产 = 总负债 + 股东权益（含少数股东）. Flag if it doesn't balance.
2. Net income linkage: 期末净资产 ≈ 期初净资产 + 归母净利润 − 分红 ± OCI. Large unexplained gaps = red flag.
3. Cash conversion: CFO vs 归母净利润 across multiple years. Sustained CFO < net income = earnings quality concern.
4. Production-sales-inventory check (where volume data is disclosed): 期初库存 + 产量 ≈ 销量 + 期末库存. Residual > ~2% of production = note and explain.

If a key statement is missing or a fetch failed, explicitly state: "三表勾稽不完整，缺少 [X 表]，以下分析基于可用数据." Do not silently skip this check.

---

Step 4 — Analysis order

1. Data quality and source transparency (Steps above)
2. Operating trend: revenue, gross margin, cost structure, product/segment/geography mix
3. Profit quality: 归母净利润 vs 扣非净利润 — explain any divergence in plain language
4. Cash flow quality: CFO conversion trend, FCF, working capital behavior
5. Balance sheet: leverage, liquidity, asset composition, notable items
6. Key ratios (see references/ratios.md)
7. Red flags: 3–7 items only, each with current value + prior-year trend + plain-language impact
8. Follow-ups: specific unanswered questions for the full filing or management

Do not jump to conclusions if data quality is uncertain.
Do not use valuation language if accounting quality is unclear.

---

Step 5 — Write the analysis

Use the structure in assets/report_template.md. Key rules for each section:

One-sentence conclusion — what changed and why it matters. Be specific.

Operating trend — always distinguish volume vs price effects where data allows (e.g. "销量+39%但均价−16%"). Segment or geographic breakdown if available.

Profit quality — present 归母净利润 and 扣非净利润 side by side in every analysis. If they diverge >10%, identify the non-recurring items and state whether they are likely to recur. This is the most commonly omitted step — do not skip it.

Cash flow quality — show CFO/net income ratio for each year available. If capex is available, compute FCF. A company with rising profits but falling CFO conversion needs an explicit explanation.

Balance sheet — leverage trend, liquidity position, working capital cycle, any outsized single items (e.g. goodwill, large inventory, high payables days).

Red flags — each flag must include:

• The current value with units [直接] or [推算: formula]
• Prior-year value for comparison (direction of trend)
• One-sentence plain-language "so what"

Follow-ups — concrete questions that require the full filing, MD&A, or earnings call to resolve.

---

Derived metric labeling rules

Any computed number must show its formula inline or in a note. No exceptions.

Metric	Required label format
净利率	推算：归母净利润 ÷ 营业收入
每GWh/吨均价	推算：收入 ÷ 销量（GWh/万吨/千升）
人均创收	推算：营业收入 ÷ 总员工数；若总员工数本身为推算，show that formula too
CFO/净利润	推算：CFO ÷ 归母净利润
净现金	推算：货币资金 + 交易性金融资产 − 有息负债
FCF	推算：CFO − capex

This labeling rule applies in ALL output formats — structured analysis, social media posts, table cells, and inline commentary.

---

Sector flags — confirm before proceeding

For these sectors, standard ratios do not apply without modification. Confirm with the user first:

• Banks / insurance / securities brokers (debt is operating input; PB and ROE-DDM replace EV/EBITDA)
• REITs (NAV-based; FFO replaces net income)
• Property developers (pre-sales recognition; NAV and cash collection rate matter more than income)
• Early-stage pre-revenue companies (no earnings; EV/Revenue or pipeline NPV)

---

Load extra files when relevant

• references/ratios.md — ratio formulas and pitfalls (read for Step 4)
• assets/report_template.md — structured output shell (use for Step 5)
• assets/example_input.csv — minimal CSV format for the compute script
• scripts/compute_ratios.py — run for repeatable ratio calculations on tabular data