MODEL RESEARCH / HORIZON-03
NVDA：一致经营路径与估值验证NVDA: coherent operating path and valuation evaluationNVDA：一致經營路徑與估值驗證
HORIZON-03／B，协议v0.3，未晋升的研究候选。市场截面为2026-10-05收盘$238.90，研究时点定义为纽约16:30；不是10月6日行情。每个目标日之后12个月Core EPS × B倍数。HORIZON-03 / B, protocol v0.3, an unpromoted research challenger. Market cutoff is the 2026-10-05 close of $238.90 with a16:30 New York research cutoff; this is not an October6 quote. Each target-date forward12-month Core EPS × B multiple.HORIZON-03／B，協議v0.3，未晉升的研究候選。市場截面為2026-10-05收盤$238.90，研究時點定義為紐約16:30；不是10月6日行情。每個目標日之後12個月Core EPS × B倍數。
| 期限Horizon期限 | 目标日 → 盈利窗口末日Target → earnings-window end目標日 → 盈利窗口末日 | Core EPS | 倍数Multiple倍數 | 候选最终估值 $／股Final challenger valuation $/share候選最終估值 $／股 |
|---|---|---|---|---|
| 0M | 2026-10-05 → 2027-10-05 | 13.8670 | 21.7836× | $302.07 |
| 6M | 2027-04-05 → 2028-04-05 | 16.6462 | 18.4146× | $306.53 |
| 12M | 2027-10-05 → 2028-10-05 | 18.9414 | 18.4215× | $348.93 |
盈利预测误差改善，但三个期限的B价格误差都变大。H03未通过价格开发门槛；没有正式价格Champion，不替换H02。网页发布的是研究记录，不是已验证模型上线。Earnings forecasts improve, but B price errors worsen at all three horizons. H03 fails the price development gates. There is no formal price champion, and H02 is not replaced. Publication of this record is not deployment of a validated model.盈利預測誤差改善，但三個期限的B價格誤差都變大。H03未通過價格開發門檻；沒有正式價格Champion，不替換H02。網頁發佈的是研究記錄，不是已驗證模型上線。
验证改善在哪里，失败在哪里Where evaluation improves and fails驗證改善在哪裏，失敗在哪裏
| 期限Horizon期限 | N | H02修复后EPS MALERepaired H02 EPS MALEH02修復後EPS MALE | H03 EPS MALE | 误差下降Error reduction誤差下降 | H02 → H03 中位APEH02 → H03 median APEH02 → H03 中位APE |
|---|---|---|---|---|---|
| 0M | 19 | 0.549 | 0.493 | 10.2% | 31.3% → 17.4% |
| 6M | 15 | 0.853 | 0.728 | 14.7% | 55.6% → 33.4% |
| 12M | 11 | 1.214 | 1.072 | 11.7% | 59.3% → 40.4% |
| 期限Horizon期限 | N | H02修复后B MALERepaired H02 B MALEH02修復後B MALE | H03 B MALE | 价格不变MALEUnchanged-price MALE價格不變MALE | H03中位APEH03 median APEH03中位APE |
|---|---|---|---|---|---|
| 0M | 24 | 0.451 | 0.559 | 0.000 | 46.6% |
| 6M | 20 | 0.535 | 0.700 | 0.364 | 53.5% |
| 12M | 16 | 0.924 | 0.944 | 0.642 | 51.8% |
盈利成熟样本为19／15／11，价格样本为24／20／16，不能混用。MALE=mean(|ln(预测/实际)|)，越低越好；它不是准确率。0M价格检验只能说明市场定价重建误差，现价并非独立公允价值真值。Matured EPS samples are19/15/11 and price samples24/20/16; do not mix denominators. MALE=mean(|ln(prediction/actual)|), with lower better; it is not an accuracy rate. The0M price test measures market-pricing reconstruction, not independent intrinsic-value truth.盈利成熟樣本為19／15／11，價格樣本為24／20／16，不能混用。MALE=mean(|ln(預測/實際)|)，越低越好；它不是準確率。0M價格檢驗只能説明市場定價重建誤差，現價並非獨立公允價值真值。
同一批历史已用于开发。盈利MALE相对H02修复版下降10.2%／14.7%／11.7%，通过预设盈利开发筛选；配对四季度移动块、5000次抽样的三个95%误差差区间都包含零。改善尚不稳定，也未经过独立样本外确认。These histories have already been used in development. EPS MALE falls10.2%/14.7%/11.7% versus repaired H02 and passes the predefined earnings development screens. All three paired95% loss-difference intervals from four-quarter moving blocks and5,000 draws include zero. Improvements are unstable and lack independent out-of-sample confirmation.同一批歷史已用於開發。盈利MALE相對H02修復版下降10.2%／14.7%／11.7%，通過預設盈利開發篩選；配對四季度移動塊、5000次抽樣的三個95%誤差差區間都包含零。改善尚不穩定，也未經過獨立樣本外確認。
即使只看盈利已成熟的同样本，0M、6M的B价格MALE仍变差：0.517→0.593、0.590→0.673；12M子集则由1.096降至0.964。完整12M价格样本却由0.924升至0.944，说明尚未成熟的盈利窗口和样本构成也会影响结论。Even within identical matured-EPS samples,0M and6M B price MALE worsen:0.517→0.593 and0.590→0.673. The12M subset improves from1.096 to0.964, while the full12M price sample worsens from0.924 to0.944. EPS-label maturity and sample composition therefore matter.即使只看盈利已成熟的同樣本，0M、6M的B價格MALE仍變差：0.517→0.593、0.590→0.673；12M子集則由1.096降至0.964。完整12M價格樣本卻由0.924升至0.944，説明尚未成熟的盈利窗口和樣本構成也會影響結論。
价格对数误差=盈利对数误差+倍数对数误差。H02部分盈利误差曾与倍数误差互相抵消；盈利更接近真实值，并不保证价格更准确。C价格收缩分支在6M／12M也未超过价格不变基准，不能替代主表条件估值。Price log error equals EPS log error plus multiple log error. Some H02 earnings and multiple errors offset each other; better EPS alone does not guarantee better prices. C price-shrinkage branches also fail to beat unchanged-price benchmarks at6M/12M and cannot replace headline conditional valuations.價格對數誤差=盈利對數誤差+倍數對數誤差。H02部分盈利誤差曾與倍數誤差互相抵消；盈利更接近真實值，並不保證價格更準確。C價格收縮分支在6M／12M也未超過價格不變基準，不能替代主表條件估值。
经营路径如何改进How the operating path changes經營路徑如何改進
先保存原H02，再做数据修复对照，最后只改变收入递推。下一季仍使用已公布收入指引×历史成熟指引偏差中位数。其后把收入换成日均收入，剔除财季季节性，再让季度增长动量按四季度半衰期逐步衰减。Preserve H02, add a data-repair control, then change only revenue recurrence. The next quarter still uses published revenue guidance × the median matured guidance-bias factor. Thereafter, use daily revenue, remove fiscal-quarter seasonality and damp quarterly momentum with a fixed four-quarter half-life.先保存原H02，再做數據修復對照，最後只改變收入遞推。下一季仍使用已公佈收入指引×歷史成熟指引偏差中位數。其後把收入換成日均收入，剔除財季季節性，再讓季度增長動量按四季度半衰期逐步衰減。
| 环节Step環節 | 固定方法Fixed method固定方法 |
|---|---|
| 季节性Seasonality季節性 | 最近12个连续且当时已披露季度：ln(收入/季度天数)=常数+时间趋势+财季效应。四季系数中心化；不足12季用零效应并标记。Latest12 consecutive then-available quarters:ln(revenue/quarter days)=intercept+time trend+quarter effects. Center the four seasonal coefficients; use flagged zero effects with insufficient history.最近12個連續且當時已披露季度：ln(收入/季度天數)=常數+時間趨勢+財季效應。四季係數中心化；不足12季用零效應並標記。 |
| 动量与递推Momentum and recurrence動量與遞推 | m=ln(下一季日收入/最新实际日收入)−季节变化。第j季日收入=上季日收入×exp(季节变化+m×2^(-(j−1)/4))，j=2..10。再乘季度天数。m=ln(next-quarter daily revenue/latest actual daily revenue)−seasonal change. Quarterj daily revenue=previous daily revenue×exp(seasonal change+m×2^(-(j−1)/4)),j=2..10. Multiply by quarter days.m=ln(下一季日收入/最新實際日收入)−季節變化。第j季日收入=上季日收入×exp(季節變化+m×2^(-(j−1)/4))，j=2..10。再乘季度天數。 |
| 盈利和倍数Earnings and capitalization盈利和倍數 | 保留指引GAAP毛利率、首季Opex指引、后续TTM费用率、最新稀释股数、Core正常化15%税率及H02 B拟合方法。同一条10季路径用于三个期限，不加期限EPS校正因子。Retain guided GAAP margin, q1 guided Opex, later TTM expense ratio, latest shares, normalized15%Core tax and H02 B fitting. One10-quarter path serves all horizons, with no horizon EPS correction factors.保留指引GAAP毛利率、首季Opex指引、後續TTM費用率、最新稀釋股數、Core正常化15%税率及H02 B擬合方法。同一條10季路徑用於三個期限，不加期限EPS校正因子。 |
四季度半衰期与最终季调增长趋于零是本轮事先固定的研究假设，未优化。阻尼趋势理论提供设计思路，不证明这套NVIDIA模型有效。The four-quarter half-life and eventual zero adjusted growth are fixed research assumptions, not optimized parameters. Damped-trend theory informs the design; it does not establish this NVIDIA model’s performance.四季度半衰期與最終季調增長趨於零是本輪事先固定的研究假設，未優化。阻尼趨勢理論提供設計思路，不證明這套NVIDIA模型有效。
数据修复、研究时点与风险Data repair, research timing and risks數據修復、研究時點與風險
| 期限Horizon期限 | 原H02冻结估值Frozen H02 value原H02凍結估值 | 仅数据修复H02估值Data-repaired H02 value僅數據修復H02估值 | H03候选估值H03 challenger valueH03候選估值 |
|---|---|---|---|
| 0M | $278.96 | $278.93 | $302.07 |
| 6M | $285.45 | $285.40 | $306.53 |
| 12M | $326.34 | $326.22 | $348.93 |
12个Q4改用原始公告的当季GAAP报表列，保留原可得日期。FY2017 Q4原推导拆股后股数28.60B，官方季度值26.40B；修复使该季Core EPS增加8.3%。这属于测量修复，不能算模型预测能力提高。原H02文件及发布成绩保持原样。Repair12 Q4 records using original-release quarterly GAAP columns, preserving availability dates. FY2017 Q4 shares were inferred as28.60B after splits versus the official26.40B; repaired Core EPS rises8.3%. This is measurement repair, not improved forecasting skill. Original H02 files and published scores remain unchanged.12個Q4改用原始公告的當季GAAP報表列，保留原可得日期。FY2017 Q4原推導拆股後股數28.60B，官方季度值26.40B；修復使該季Core EPS增加8.3%。這屬於測量修復，不能算模型預測能力提高。原H02文件及發佈成績保持原樣。
53周财年的额外7天按日均收入处理；FY2027 Q4为98天。Core经营利润已扣GAAP SBC，不重复扣除；15%是正常化研究税率，区别于公司16–18%的全年指引。把所有标签与预测税率同时改为17%并重拟合倍数，估值基本抵消，不是新的准确率提升。Daily revenue handles the extra7 days in a53-week year; FY2027 Q4 has98 days. GAAP operating profit already deducts SBC. The15%tax rate is normalized research policy, distinct from16–18%full-year guidance. Scaling all labels and predictions to17% and refitting multiples cancels in valuations, not a new accuracy gain.53周財年的額外7天按日均收入處理；FY2027 Q4為98天。Core經營利潤已扣GAAP SBC，不重複扣除；15%是正常化研究税率，區別於公司16–18%的全年指引。把所有標籤與預測税率同時改為17%並重擬合倍數，估值基本抵消，不是新的準確率提升。
利率输入为2026-10-02的5.28%，超过各期限成熟训练样本上限4.61%。B在证据范围之外外推，当前估值存在重要倍数风险。本轮只披露该风险，未加入新的利率裁剪或重新选择倍数。The rate input is5.28% from2026-10-02, exceeding every horizon’s mature-training maximum of4.61%. B extrapolates outside its evidence support, creating material capitalization risk. This round discloses it without new yield clipping or multiple selection.利率輸入為2026-10-02的5.28%，超過各期限成熟訓練樣本上限4.61%。B在證據範圍之外外推，當前估值存在重要倍數風險。本輪只披露該風險，未加入新的利率裁剪或重新選擇倍數。
Fed H15现行发布时间是纽约16:15，因此本轮明确16:30研究时点；收盘价只是参考，不能声称在同一收盘成交。历史精确发布时刻和利率vintage未完整取得，仍不是完整PIT回测。Current Fed H15 publication is16:15 New York, so this round defines a16:30 research cutoff. The closing price is a reference, with no claim of execution at that close. Historical exact release timestamps and rate vintages remain incomplete; this is not a complete PIT backtest.Fed H15現行發佈時間是紐約16:15，因此本輪明確16:30研究時點；收盤價只是參考，不能聲稱在同一收盤成交。歷史精確發佈時刻和利率vintage未完整取得，仍不是完整PIT回測。
Fed H15发布时间Fed H15 release timingFed H15發佈時間 · FRED利率来源FRED rate sourceFRED利率來源 · 公司最新核验公告Latest verified company release公司最新核驗公告
复算检查与可追溯记录Reconciliation and traceable records復算檢查與可追溯記錄
实现检查全部通过：原H02纯规则复现、未来数据扰动、季节拟合时点、季度天数、首季指引相同、共同盈利路径、成熟训练日期、价格误差分解和同口径税率不变性。实现正确与预测有效性分别评价。Implementation checks pass: pure H02-rule replication, future-data perturbations, as-of seasonal fitting, quarter days, unchanged first-quarter guidance, shared earnings paths, mature training dates, price-error decomposition and same-basis tax invariance. Implementation correctness and predictive validity are separate.實現檢查全部通過：原H02純規則復現、未來數據擾動、季節擬合時點、季度天數、首季指引相同、共同盈利路徑、成熟訓練日期、價格誤差分解和同口徑税率不變性。實現正確與預測有效性分別評價。
初始协议、Q4修复补充及最终v0.3时点补充均保留，发生在首次H03输出之前；此后没有调半衰期、季节窗口或增设校正候选。研究日期纽约2026-10-06，捕获日期UTC2026-10-07；网页保存派生估值、方法、聚合误差，原始证据与Pro数据继续保留内部。Initial protocol, Q4-repair amendment and finalv0.3 timing amendment are retained, all before the first H03 output. No subsequent half-life, seasonal-window or calibration tuning occurred. Research date is2026-10-06 New York; captures are2026-10-07 UTC. The page stores derived values, methods and aggregate errors; raw evidence and Pro data remain internal.初始協議、Q4修復補充及最終v0.3時點補充均保留，發生在首次H03輸出之前；此後沒有調半衰期、季節窗口或增設校正候選。研究日期紐約2026-10-06，捕獲日期UTC2026-10-07；網頁保存派生估值、方法、聚合誤差，原始證據與Pro數據繼續保留內部。
Protocol SHA256 c9ea0fc37c54d047b51348251ea6a30cce7404f371dc337f6ed4eb07bb178e9f
Record SHA256 0ca7b16e644a4b7480e44505b581fd2cd47a6b4a780708fb907865f63ea39b8c
下一轮优先研究Priority for the next round下一輪優先研究
保留H03收入路径作为盈利研究候选，不替换估值模型。下一轮优先研究倍数层：利率训练范围以外的稳健处理，以及盈利成长、现金流和商业阶段是否带来可验证增量；同时检查毛利率、费用率和股数桥。先锁规则，再计算，再检验新origin，停止在相同历史上反复选优。Retain H03 as an earnings-research candidate, without replacing the valuation model. Prioritize robust capitalization beyond rate-training support, and whether earnings growth, cash flow and business stages add verified value; also examine margin, expense and share bridges. Lock rules before evaluation, test new origins and stop repeated selection on the same history.保留H03收入路徑作為盈利研究候選，不替換估值模型。下一輪優先研究倍數層：利率訓練範圍以外的穩健處理，以及盈利成長、現金流和商業階段是否帶來可驗證增量；同時檢查毛利率、費用率和股數橋。先鎖規則，再計算，再檢驗新origin，停止在相同歷史上反覆選優。