走一遍 4 种 DCF 方法、周期股判定、情绪缓冲逻辑,以及最终评分系统。A walkthrough of the 4 DCF methods, cyclicality detection, sentiment buffer logic, and final scoring system.
季度年度化(默认主锚)。Q-annualized (default primary). 最新季 EPS × 4 × 目标 P/E 倍数。最当下的 run-rate 信号。倍数按分类区分 — 周期股用 8 / 14 / 20(悲观/基准/乐观),增长股用 18 / 25 / 35。Latest quarterly EPS × 4 × target P/E multiple. Most current run-rate signal. Multiples differ by classification — cyclical uses 8 / 14 / 20 (bear/base/bull), secular uses 18 / 25 / 35.
前瞻 FCF 调整。Forward FCF-adjusted. 对未来 5 年分析师 EPS 加总,乘以 FCF 转化系数(周期股约 60%,反映资本开支负担),折现后加终值。最适合没季度数据的周期股。Sums analyst EPS through FY+5, multiplied by an FCF conversion factor (~60% for cyclicals, captures capex burden), discounted, plus terminal value. Best for cyclicals without quarterly data.
前瞻纯 EPS。Forward Pure EPS. 与上类似,但将 EPS 当作可全部分配的现金对待。仅作对照参考 — 对资本密集型业务会高估。Same as above but treats EPS as fully distributable. Shown for reference — overstates for capital-intensive businesses.
历史 FCF DCF。Historical FCF DCF. 基于最近年报 FCF 的两阶段 Gordon Growth 模型。对稳定增长股有效,对周期股会系统性失真。2-stage Gordon Growth from latest annual FCF. Works for stable seculars, systematically misprices cyclicals.
一只股票被判定为 周期股 满足以下任一:A stock is classified as cyclical if:
neg_income_years ≥ 1 in 5y history
OR peak income year is not in latest 2 years
OR ≥1 income drop of −25% AND beta > 1.5
为什么不直接用行业?NVDA 属于半导体行业但盈利单调增长;MU 也属于半导体但盈利大起大落。行业标签会掩盖真实信号。Why data-driven, not industry-based? Because NVDA is in Semiconductors but has monotonic growth; MU is also in Semiconductors but has massive earnings swings. An industry label hides the real signal.
0–100 的情绪分数由价格动量代理计算:1 年收益(权重最大)、价格相对 200 日均线、1 个月动量、均线序列。A 0–100 sentiment score is computed from price-action proxies: 1-year return (heaviest weight), price relative to 200-day MA, 1-month momentum, and SMA alignment.
当情绪与基本面强烈分歧时,缓冲值调整基本面评级:When sentiment and fundamentals diverge sharply, a buffer adjusts the fundamental verdict:
sentiment ≥ 80 AND fundamentals bearish → buffer +2
sentiment ≥ 65 AND fundamentals bearish → buffer +1
sentiment ≤ 20 AND fundamentals bullish → buffer +1
原理:对一只刚涨了 +700% 的股票做纯基本面 SELL,学术上正确但操作上错误。缓冲把这种信号软化为 REDUCE(渐进减仓)。反之,恐慌出清情绪 + 强基本面 = 逆向机会。Rationale: a pure-fundamental SELL on a stock that just rallied +700% is academically right but practically wrong. The buffer softens such calls to REDUCE (gradual exit). Conversely, capitulation sentiment + strong fundamentals = contrarian opportunity.
基本面分数(通常 −5 到 +9)加总自:估值(DCF 对应现价的空间)、盈利能力(ROE、净利率)、增长(增长股的营收 CAGR)、资产负债表(净现金)、晚周期警示(价格/200 日均线、1 年涨幅、周期股的 P/S)。Fundamental score (range typically −5 to +9) sums points from valuation (DCF upside vs current price), profitability (ROE, net margin), growth (revenue CAGR for seculars), balance sheet (net cash), and late-cycle warnings (price/SMA200, 1Y return, P/S for cyclicals).
最终分数 = 基本面分数 + 情绪缓冲。Final score = fundamental + sentiment buffer.
≥ 7 → STRONG BUY
≥ 4 → BUY
≥ 1 → HOLD
≥ −2 → REDUCE
else → SELL
这些是 相对吸引力信号,不是价格目标。它们反映报告生成时刻基于公开数据 + 规则逻辑的快照。报告不会自动更新 — 重新跑流水线才能刷新。These are relative attractiveness signals, not price targets. They reflect a snapshot at the time the report was generated, using public data and rule-based logic. Reports are not updated automatically — re-run the pipeline to refresh.