This skill is the framework specification for Phase 4 (valuation) of the company-research SOP, and also applies to any discussion of "is this company expensive". Principles:Numbers come from calc, interpretation comes from this manual, conclusions stop at "scenario / decision point"(AGENTS.md §0 items 2 and 3).
0. Four iron rules
- Every PE / CAGR / PEG / year count must be the output of one
calc/cli.pycall (with calculation_id and input evidence id); if there is no corresponding evidence, write "not obtained: reason", and do not substitute memory or a different framework. - Do not output price anchors(target price / fair price / buy price / stop-loss price). The "anchor" for digestion years is the PE multiple scenario, not price; the report only gives a "scenario → years" table and decision points.
- Low PEG ≠ safe: the denominator of PEG (forward CAGR) is the only forecast in the entire calculation; when the cycle turns, E gets revised down and PEG jumps. Any PEG interpretation must simultaneously give "whether the forward vs TTM facts match up".
- Units and reporting periods are carried into calc as-is from the evidence, and calc normalizes them; do not convert ten-thousand yuan / hundred-million yuan in the prompt, and do not annualize yourself.
1. Numerator: the three PE frameworks (report side by side, primarily use the first)
| Framework | Formula (calc function) | Purpose | Known pitfalls |
|---|---|---|---|
| Non-recurring ×4 annualized PE(primary) | Total market cap ÷ (latest single-quarter non-recurring net profit × 4) → pe_deducted_annualized(total_market_cap, cap_unit, latest_quarter_deducted_profit, profit_unit) |
Captures the "current run rate", less lagging than TTM for fast-growing stocks | Must use non-recurring, not attributable to parent(single quarter ×4 amplifies one-off gains/losses such as investment income / subsidies / impairments by 4x);Seasonality: off-season single quarter ×4 overestimates PE, peak-season single quarter ×4 underestimates it—the direction of peak/off-season can only be judged fromthe company's own single-quarter series(earnings-analysis §3), and industry impressions must not be applied; when interpreting, you must state which quarter the latest single quarter is and the directional bias |
| Forward PE | Current price ÷ consensus EPS (FY T mean) → forward_pe(price, eps_forecast) |
Analysts have already handled seasonality; compare with non-recurring ×4 to see "which year the market is pricing" | Depends on consensus quality (number of institutions ≥ 3; see §3) |
| TTM PE + historical percentile | Total market cap ÷ sum of net profit for the last 4 quarters → pe_ttm_from_parts(...), and cross-check with the data source pe_ttm; percentile percentile_rank(history={"history_csv": {"raw_ref": <PE 历史 raw 文件>, "column": "peTTM", "where": {"tradestatus": "1"}, "date_column": "date"}}, current=pe_ttm)(for the CLI series input form see calc/SPEC.md §3; do not treat history_csv / where as top-level parameters) |
Historical position reference; percentile < 20 valid samples → not_meaningful | For fast-growing stockslagging and too high(includes old quarters with low profit); use only as a reference column, not as the primary judgment |
Definition of T: T = current fiscal year (the year of the current date in Asia/Shanghai).
2. Denominator: sustainable growth rate (primarily use forward CAGR, mandatory cross-validation)
- Forward CAGR = (consensus FY T+2 EPS ÷ FY T EPS)^(1/2) − 1 →
forward_cagr(eps_t, eps_t_plus_n, years=2). The smoothest, with base effects and seasonality removed; but it isthe softest link: sell-side is systematically optimistic about hot tracks, far-year samples are thin, dispersion is wide, and it will be continuously revised.must report min / max together with the number of institutions(consensus_dispersion(low, mean, high), look at max ÷ min and (max − min) ÷ mean). - TTM YoY(fact, seasonality- and single-quarter base-adjusted) = net profit sum of the last 4 quarters ÷ net profit sum of the prior 4 quarters − 1 →
ttm_yoy(single_quarters, end_period, unit, money=true);mainly use the attributable-to-parent basis(same basis as consensus EPS), with the non-recurring-inclusive basis presented alongside for cross-check. - Cross-validation interpretation →
forward_vs_ttm_judgement(forward_cagr_value, ttm_yoy_value, tolerance_pp=10):approx(forward ≈ TTM: growth already realized, forward is credible) /forward_below(forward far below TTM: the market is betting on a sharp slowdown, ask why) /forward_above(forward far above TTM: analysts are painting a rosy picture, highest risk of PEG distortion). - Two prohibited items: ❌ single-quarter YoY as the denominator (last year's low base inflates it artificially); ❌ QoQ as the denominator (seasonality makes a growth company look negative, and annualizing is even more explosive). ✅ QoQ (
qoq) only as a "turning point / momentum" thermometer; label it in the report as a signal, not as a growth rate.
3. Credibility of consensus expectations
| Scenario | Handling |
|---|---|
| number of institutions ≥ 3, min / max both present | normally goes into forward_cagr, report lists "number of institutions · range" |
| number of institutions < 3 | mark as "consensus expectations unreliable", can still be calculated but interpretation is downgraded, write "await more coverage" in the verdict points |
| only per-report forecasts (East Money backup source) | can only be called "per-report forecasts",must not impersonate consensus expectations, must not go into forward_cagr |
| dispersion max ÷ min ≥ 2 | state explicitly "far-year EPS dispersion ≥ 2x; PEG / years range should be calculated once each at min–max rather than only looking at the mean" |
4. Interpretation table
| PEG (non-recurring × 4 ÷ forward CAGR) | Initial judgment | must be overlaid with |
|---|---|---|
| < 1 | growth can digest the valuation | "cheap" only counts when forward vs TTM = approx; when forward_above, write "cheap is what the forecast gives it" |
| ≈ 1 | Reasonable | same as above |
| > 1.5–2 | Expensive | see whether "the harder the bottleneck, the more expensive" (irreplaceability already priced in), see industry-chain |
PEG is a high-incidence "value trap" for growth-cyclical stocks: PE is lowest just before the cycle tops (E is at the peak). Any conclusion of PEG < 1 must be accompanied by a counterargument (catalyst-risk §2).
5. Four-anchor PE digestion years and the "30x anchor three iron rules"
pe_digestion_years(pe, cagr, anchor) = ln(PE ÷ anchor) ÷ ln(1 + CAGR); PE ≤ anchor → 0 years (details.below_anchor).pe_digestion_scenarios(pe, cagr) calculate four anchors at once:boom continues 30 / neutral deceleration 25 / cycle re-rating upper bound 22 / lower bound 18(x).
- the four anchors arescenarios, not forecasts: the report must give all four together, and state the verdict data corresponding to each scenario (e.g. downstream capex guidance, price / rent thermometer, monthly production scheduling data).
- 30x anchor three iron rules: ① 30x only multiplies "locked / nearly locked" profit (earnings previews, orders in hand for the year),never multiply far-forward (T+2) consensus to make a judgment——That is the mathematical form of "paying a perpetual premium for a perpetual shortfall"; ② Anchor and adjudication data switching: use 30 for continued boom, 25 for deceleration, and 18–22 for re-rating; write the switching conditions into the adjudication point; ③ State conclusions separately by scenario: "a conclusion holds only in the 30x scenario / a conclusion also holds in the 25x scenario"—only describe the conditions under which the scenario holds,do not translate into position sizing or buy/sell recommendations。
- Historical central tendency: this skill does not give any PE central-tendency values for any company or industry; whenever "historical position" is needed, always use
fetch_pe_history's measured series processed throughpercentile_rankto derive the percentile, and state the sample period.
6. Standard output columns (consistent with SOP §1 valuation stage)
扣非×4 PE | 前瞻 PE | TTM PE(分位) | ★PEG(扣非×4 ÷ 前瞻 CAGR) | 前瞻 CAGR(机构数 · 区间) | TTM 同比(交叉验证) | 环比(拐点) | 四锚消化年数 | 前瞻 vs TTM 判读
Each cell: calculation_id, or "not obtained: reason" (follow the structured gaps in SOP §2), or not_meaningful (state the meaningless domain as-is: PE ≤ 0 / CAGR ≤ 0 / insufficient sample).
7. Cyclical growth stock calibration (when the cycle turns, E changes before P)
- A high PE is not necessarily expensive (E is depressed at the cycle bottom), and a low PE is not necessarily cheap (E is at its peak at the cycle top)—first judge the cycle position (industry capacity / prices / downstream capital expenditure), then look at PE.
- Forward CAGR is systematically optimistic near the cycle top; use TTM year-over-year + quarter-over-quarter inflection points + monthly industry data together to see "acceleration or deceleration."
- The "valuation" section of the report must clearly state: which single quarter's data the current PE uses, the seasonal direction, the number of institutions and range for the forward CAGR, the interpretive conclusion and the premises under which it holds.
8. Common error checklist (self-check)
- Using net profit attributable to the parent for a single quarter ×4 as non-recurring profit ×4 (one-off gains/losses are magnified).
- Using single-quarter year-over-year / quarter-over-quarter as the PEG denominator.
- Looking only at the consensus mean without looking at the range and number of institutions.
- Treating the forecasts from individual research reports as consensus expectations.
- Giving only one anchor for the digestion years; or multiplying the forward consensus by 30x.
- Giving a price anchor, target price, or "reasonable market cap."
- Converting units or annualizing yourself (should be left to calc).
- Reporting a percentile when the percentile sample is < 20.