stock-valuation

What this skill produces

A structured valuation report with:

1. Market context selection — China / US / global parameter set and why
2. Industry classification — determines which methods apply
3. Data inventory — what inputs are available vs missing
4. Multi-method valuation — 2–4 methods suited to the industry, each producing a value range
5. Triangulation — weighted synthesis across methods with an explicit rationale
6. Sensitivity table — how the conclusion changes under bull / base / bear assumptions
7. Market context — recent news, sentiment, and how the market appears to be pricing the stock
8. Disclaimer — mandatory; see Step 8

---

Step 1 — Select the market context first

Read references/market-context.md and identify the valuation market context before selecting methods or assumptions.

Record:

• Primary listing market and ticker suffix
• Reporting currency
• Primary investor base / peer set
• Primary operating market (for long-run growth assumptions)

Rules:

• Use the listing market / investor base to choose risk-free rate, ERP, comparable universe, and default multiple ranges.
• Use the primary operating market to anchor terminal growth and industry policy context.
• If the company is cross-listed or has mixed exposure, state which market you are anchoring to and why.
• Do not reuse China A-share parameters for a US-listed stock unless you explicitly justify it.

Step 2 — Classify the industry and select methods

Read references/industry-metrics.md to identify the company's sector and the recommended valuation method set. Do not skip this step — applying PE to a bank or DCF to an early-stage biotech produces meaningless output.

Quick reference (full detail in references/industry-metrics.md):

Sector	Primary methods	Avoid
Technology / AI chips	PS, EV/Revenue, EV/EBITDA, PEG	Single-year PE if unprofitable
Internet / platform	EV/EBITDA, PS, DCF (terminal value heavy)	PB (asset-light)
Banks / insurance	PB, ROE-based DDM, P-PPOP	PE distorted by provisions
Consumer / retail	PE, EV/EBITDA, DCF	PS (margins matter here)
Pharma / biotech	Pipeline NPV, EV/Revenue (if pre-profit), PE (if profitable)	DCF alone (binary outcomes)
Energy / utilities	EV/EBITDA, EV/DACF, DDM	PE (capex distorts)
Real estate (non-REIT)	NAV, P/NAV, PE	DCF (illiquid assets)
Manufacturing	PE, EV/EBITDA, PB	PS
Pre-revenue / startup	EV/Revenue, comparable transaction multiples	PE, DCF

---

Step 3 — Gather and validate inputs

Collect from the filing, conversation context, or web search. Tag each item as [filing], [market] (real-time price data), [derived], or [estimated]. Never present [estimated] figures without labeling them.

Before computing anything, start a provenance ledger using references/provenance-template.md.

Minimum columns:

• field
• period
• value
• tag
• source_file_or_url
• page_or_capture_date
• section
• derivation_note

Rules:

• Every published figure, multiple, and assumption must have a ledger row.
• Market data rows must include the exact capture date.
• Comparable-company multiples must include source and date; they are time-sensitive.
• For [estimated] inputs, state whether the estimate comes from consensus, management guidance, or analyst inference.

From the financial filing

Input	Required for
Revenue (3yr)	PS, EV/Revenue, DCF
EBITDA (3yr)	EV/EBITDA
Net profit / EPS (3yr)	PE, PEG
扣非净利润 (3yr)	PE quality check
Free cash flow (3yr)	DCF
Book value per share	PB
Dividends per share	DDM
Net debt / cash	EV calculation for all EV-based methods
Shares outstanding	Market cap, EV
Revenue growth guidance or consensus	PEG, DCF terminal growth

From current market data (use web search)

Input	Notes
Current share price	As of today; note date
52-week high / low	Context for where price sits in range
Market cap	Verify: price × shares
Recent analyst consensus / target prices	Note source and date
Recent news (last 30–90 days)	Earnings surprises, policy, macro, sector events

Industry comparable multiples (use web search or analyst reports)

Input	Notes
Sector median PE / PS / EV-EBITDA	Use current-year forward multiples where possible
Key direct peers' multiples	3–5 closest competitors in the chosen market context
Historical range for this company's own multiples	3–5yr range if available

---

Step 4 — Run each selected method

For each method, produce:

• Formula used (written out explicitly)
• Inputs plugged in (with source tags)
• Central estimate (base case)
• Range (bull / base / bear) driven by the key assumption that varies most

Read references/valuation-methods.md for full computation guidance on each method.

Mandatory for all methods

• Use forward-looking inputs (next 12 months or next FY) as the primary case; supplement with trailing for context
• When using multiples: anchor to industry median, then adjust ±20–40% for quality (growth premium, risk discount, governance)
• Never use a single comparable — use a range of 3–5 peers and take the median
• State the key assumption that most affects the output for each method

DCF-specific rules

• Use at least 3 distinct WACC assumptions (bear +1%, base, bull -1%)
• Terminal growth rate must be ≤ long-run nominal GDP growth of the company's primary operating market; use references/market-context.md for market-specific defaults
• If FCF history is <3 years, flag DCF as [low confidence] and weight it lower in triangulation

---

Step 5 — Triangulate and produce the valuation range

1. List each method's bull / base / bear output in a summary table
2. Assign a weight to each method (must sum to 100%); justify the weighting
3. Compute weighted base-case fair value and weighted range
4. Compare to current price: state the implied upside/downside
5. Flag if methods diverge by >30% — large divergence usually means one method is inappropriate or an input is wrong

Weighting principles:

• Weight methods higher when: inputs are from the filing (not estimated), the method is standard for this sector, and the company has ≥3 years of history for the relevant metric
• Weight methods lower when: FCF history is short, the company is loss-making (PE gets zero weight), or the method relies heavily on terminal value

---

Step 6 — Sensitivity table

Produce a 3×3 matrix for the most impactful variable (usually revenue growth or exit multiple):

	Bear case	Base case	Bull case
Low multiple	XX	XX	XX
Mid multiple	XX	XX (base)	XX
High multiple	XX	XX	XX

Bold the base-case cell. The range across the table is the honest uncertainty range.

---

Step 7 — Market context and news synthesis

Use web search to gather:

• Recent earnings / guidance — any surprises vs consensus?
• Sector/policy news — regulatory shifts, government support or restriction, supply chain events
• Macro backdrop — interest rate environment, currency, relevant commodity prices
• Sentiment signals — short interest, insider transactions, institutional flow if available

Summarize in 3–5 bullet points. State whether each item is a tailwind, headwind, or neutral to the valuation.

---

Step 8 — Write the output report

Structure:

# [Company] 综合估值报告
日期：YYYY-MM-DD  |  当前股价：XX [计价货币]  |  分析师声明：见文末

## 一、市场与行业定位
## 二、采用的估值方法及理由
## 三、各方法估值结果
## 四、综合估值区间
## 五、敏感性分析
## 六、市场环境与近期消息
## 七、关键假设、来源与风险提示
## 声明（必须包含）

The report must end with this disclaimer verbatim (translate to Chinese if the report is in Chinese):

声明：本报告为基于公开信息的分析性输出，不构成任何投资建议或买卖推荐。估值结果依赖于假设和历史数据，不代表未来表现。投资者应独立判断并承担相应风险。本分析由 AI 生成，未经持牌投资顾问审核。

---

Step 9 — Quality checklist

Accuracy gates:

• Market context selected correctly; parameter set matches references/market-context.md
• Industry correctly classified; method set matches references/industry-metrics.md
• Current share price sourced from web search with date stamp
• All [estimated] inputs labeled as such
• 扣非净利润 used for PE/PEG, not 归母净利润 alone
• EV computed correctly: market cap + net debt (or − net cash)
• No method uses single-comparable anchoring (minimum 3 peers)

Completeness gates:

• At least 2 methods computed (3 preferred)
• Bull / base / bear range for every method
• Sensitivity table present
• Market context section covers last 30–90 days
• Provenance ledger covers every published figure, multiple, and assumption
• Disclaimer present and unmodified

Intellectual honesty gates:

• If methods diverge >30%: flagged and explained
• If key inputs are estimated or unavailable: confidence level stated
• If DCF is used with <3yr FCF history: labeled [low confidence]
• Conclusion states a range, never a single price target

---

Reference files

• references/industry-metrics.md — sector classification and recommended method sets (read for Step 2)
• references/market-context.md — market-specific rates, peer-set rules, and terminal-growth defaults (read for Step 1)
• references/valuation-methods.md — computation formulas, input sources, common pitfalls for each method (read for Step 4)
• references/comparable-template.md — how to structure the peer comparison table
• references/provenance-template.md — ledger format for figure and assumption traceability