RESEARCH VALIDATION / HORIZON-02
NVDA 0M／6M／12M研究验证NVDA 0M / 6M / 12M research evaluationNVDA 0M／6M／12M研究驗證
研究记录：2026-10-06。市场输入截至2026-10-05收盘；H02-B raw v0.1，研究条件估值，未部署。Research record: 2026-10-06. Market inputs through the 2026-10-05 close; H02-B raw v0.1, research conditional valuations, not deployed.研究記錄：2026-10-06。市場輸入截至2026-10-05收盤；H02-B raw v0.1，研究條件估值，未部署。
| 期限Horizon期限 | 估值目标日Valuation date估值目標日 | Core EPS | 倍数Multiple倍數 | 最终计算估值 $／股Final calculated valuation $/share最終計算估值 $／股 |
|---|---|---|---|---|
| 0M | 2026-10-05 | 12.8047 | 21.7858× | $278.96 |
| 6M | 2027-04-05 | 15.4985 | 18.4176× | $285.45 |
| 12M | 2027-10-05 | 17.7089 | 18.4281× | $326.34 |
三个期限统一使用目标日后12个月Core EPS × 对应期限B倍数。Core采用15%正常化税率。主表选择是一致的报告口径，尚无正式价格Champion。All horizons use the target-date forward 12-month Core EPS × the corresponding B multiple. Core uses a normalized 15% tax rate. This is a consistent reporting basis; there is no formal price champion.三個期限統一使用目標日後12個月Core EPS × 對應期限B倍數。Core採用15%正常化税率。主表選擇是一致的報告口徑，尚無正式價格Champion。
验证结果与最终估值分别解释Evaluation results and valuations have distinct meanings驗證結果與最終估值分別解釋
本轮主表的$278.96／$285.45／$326.34来自未校正B。B在6M和12M价格检验中落后价格不变基准。12M误差改善11.9%属于CAL-C，不属于主表B，也不证明$326.34更准确。The $278.96 / $285.45 / $326.34 headline values come from uncalibrated B. B lags the unchanged-price baseline in the 6M and 12M price evaluations. The 11.9% improvement belongs to CAL-C, not headline B, and does not establish that $326.34 is more accurate.本輪主表的$278.96／$285.45／$326.34來自未校正B。B在6M和12M價格檢驗中落後價格不變基準。12M誤差改善11.9%屬於CAL-C，不屬於主表B，也不證明$326.34更準確。
| 期限Horizon期限 | 分支Branch分支 | N | MALE | 中位APEMedian APE中位APE | P90 APE | 解释Interpretation解釋 |
|---|---|---|---|---|---|---|
| 0M | B | 24 | 0.451 | 31.5% | 66.8% | 市场定价诊断Market-pricing diagnostic市場定價診斷 |
| 0M | CAL-A | 24 | 0.377 | 33.7% | 89.2% | 市场定价诊断Market-pricing diagnostic市場定價診斷 |
| 6M | 价格不变基准Unchanged-price baseline價格不變基準 | 20 | 0.364 | 28.2% | 57.0% | 同样本基准Same-sample baseline同樣本基準 |
| 6M | B | 20 | 0.535 | 35.6% | 72.5% | 落后基准Lags baseline落後基準 |
| 6M | CAL-C | 20 | 0.376 | 27.5% | 62.1% | 落后基准Lags baseline落後基準 |
| 12M | 价格不变基准Unchanged-price baseline價格不變基準 | 16 | 0.642 | 45.2% | 70.7% | 同样本基准Same-sample baseline同樣本基準 |
| 12M | B | 16 | 0.923 | 48.3% | 90.7% | 落后基准Lags baseline落後基準 |
| 12M | CAL-C | 16 | 0.566 | 38.4% | 70.0% | 仅开发候选Development candidate only僅開發候選 |
MALE = mean(|ln(预测/实际)|)，是对数误差，不是准确率或百分比。APE = |预测−实际|/|实际|。每个期限只比较相同origin、相同价格标签的样本；0M的实际市场价格不是独立公允价值真值。MALE = mean(|ln(prediction/actual)|), a log error, not an accuracy rate or percentage. APE = |prediction−actual|/|actual|. Each horizon uses matched origins and price labels. The 0M market price is not independent intrinsic-value truth.MALE = mean(|ln(預測/實際)|)，是對數誤差，不是準確率或百分比。APE = |預測−實際|/|實際|。每個期限只比較相同origin、相同價格標籤的樣本；0M的實際市場價格不是獨立公允價值真值。
0M：CAL-A主误差较A改善，但中位APE30.6%→33.7%，盈利层也变差，只保留诊断。6M：CAL-C MALE 0.376仍差于基准0.364。12M：CAL-C MALE 0.566对基准0.642，中位APE45.2%→38.4%，P90 70.7%→70.0%，4/4财季面板改善。0M: CAL-A improves MALE versus A, but median APE worsens from 30.6% to 33.7% and earnings errors also worsen; retain it as a diagnostic. 6M: CAL-C MALE of 0.376 still lags the 0.364 baseline. 12M: CAL-C has MALE 0.566 versus 0.642, median APE 45.2%→38.4%, P90 70.7%→70.0%, with improvement in all four fiscal-quarter panels.0M：CAL-A主誤差較A改善，但中位APE30.6%→33.7%，盈利層也變差，只保留診斷。6M：CAL-C MALE 0.376仍差於基準0.364。12M：CAL-C MALE 0.566對基準0.642，中位APE45.2%→38.4%，P90 70.7%→70.0%，4/4財季面板改善。
为什么尚不能晋升Why promotion remains unsupported為什麼尚不能晉升
12M只有16个成熟价格窗口，且高度重叠。采用连续四个季度的移动时间块、5000次抽样，配对MALE差95%区间为-0.197至+0.037，仍包含零。开发门槛通过并不代表独立样本外成功。There are only 16 mature 12M price windows, with substantial overlap. A paired moving-block bootstrap using four consecutive quarters and 5,000 draws gives a 95% MALE-difference interval of -0.197 to +0.037, which includes zero. Passing development gates does not establish independent out-of-sample success.12M只有16個成熟價格窗口，且高度重疊。採用連續四個季度的移動時間塊、5000次抽樣，配對MALE差95%區間為-0.197至+0.037，仍包含零。開發門檻通過並不代表獨立樣本外成功。
校正阶段是在看到首轮盈利低估之后选择的顺序开发诊断。当前12M校正因子3.31会把Core EPS放大至58.62，缺少独立经营支持。其CAL-C诊断价$508.02被拒绝正式采用，不能当作本轮12M最终估值。所有独立晋升状态仍为false。The calibration stage was selected after observing first-stage earnings underprediction, so it is sequential development. The live 12M factor of 3.31 raises Core EPS to 58.62 without independent operating support. Its $508.02 CAL-C diagnostic is rejected for formal use and is not this round’s final 12M valuation. All independent-promotion flags remain false.校正階段是在看到首輪盈利低估之後選擇的順序開發診斷。當前12M校正因子3.31會把Core EPS放大至58.62，缺少獨立經營支持。其CAL-C診斷價$508.02被拒絕正式採用，不能當作本輪12M最終估值。所有獨立晉升狀態仍為false。
本轮模型如何计算How this research model calculates values本輪模型如何計算
H02-OP-GUIDE085是新的公开数据实现。原R1／G0O1／LIVE-02B工作簿未取得，因此本轮不称为原模型复现，也不合并旧17-origin成绩。H02-OP-GUIDE085 is a new public-data implementation. The original R1 / G0O1 / LIVE-02B workbook is unavailable, so this is not a reproduction and its results are not merged with the earlier 17-origin study.H02-OP-GUIDE085是新的公開數據實現。原R1／G0O1／LIVE-02B工作簿未取得，因此本輪不稱為原模型復現，也不合並舊17-origin成績。
| 步骤Step步驟 | 固定规则Fixed rule固定規則 |
|---|---|
| 经营预测Operating forecast經營預測 | 下一季收入指引 × 最近最多8次已成熟指引偏差中位因子；下一季YoY随后按固定0.85逐季衰减。Next-quarter revenue guidance × the median bias from up to eight matured guidance outcomes; subsequent YoY growth decays quarterly by a fixed 0.85.下一季收入指引 × 最近最多8次已成熟指引偏差中位因子；下一季YoY隨後按固定0.85逐季衰減。 |
| 利润与股数Profit and shares利潤與股數 | GAAP毛利率使用下一季指引；首季Opex用指引，后续用当时TTM费用率；稀释股数固定于最新披露值。Core EPS=经营利润×85%/稀释股数。Use next-quarter guided GAAP gross margin, guided first-quarter Opex and the then-available TTM expense ratio thereafter. Hold shares at the latest disclosure. Core EPS=operating profit×85%/diluted shares.GAAP毛利率使用下一季指引；首季Opex用指引，後續用當時TTM費用率；稀釋股數固定於最新披露值。Core EPS=經營利潤×85%/稀釋股數。 |
| 期限盈利窗口Horizon earnings windows期限盈利窗口 | 按照52／53周财年和季度日期重叠，计算每个目标日之后12个月Core EPS，保留季度均匀分摊近似。Use 52/53-week fiscal calendars and quarter-date overlap to calculate the forward 12-month Core EPS at each target date; uniform intra-quarter allocation remains an approximation.按照52／53周財年和季度日期重疊，計算每個目標日之後12個月Core EPS，保留季度均勻分攤近似。 |
| A / B | A用最近8个成熟标签倍数的中位数；B用至少12个成熟标签，逐期限扩展拟合ln(M)=α+β×origin滞后十年利率。A uses the median multiple of eight matured labels. B requires at least 12 matured labels and fits ln(M)=α+β×lagged origin 10Y yield separately by horizon.A用最近8個成熟標籤倍數的中位數；B用至少12個成熟標籤，逐期限擴展擬合ln(M)=α+β×origin滯後十年利率。 |
| C / CAL | C=P0×(B/P0)^k；k6=.25、k12=.50固定。CAL因子来自最多8个完全成熟的盈利残差，至少4个。C按校正B重新收缩，不直接乘EPS因子。C=P0×(B/P0)^k with fixed k6=.25 and k12=.50. CAL uses up to eight fully matured earnings residuals, with at least four required. Recompute C from calibrated B; do not multiply C directly by the EPS factor.C=P0×(B/P0)^k；k6=.25、k12=.50固定。CAL因子來自最多8個完全成熟的盈利殘差，至少4個。C按校正B重新收縮，不直接乘EPS因子。 |
估值与价格路径分开：EPS×倍数生成条件估值；C向当前市场价收缩，研究价格路径。后者不能无说明地替代公允价值或主表估值。盈利标签须等覆盖窗口最后一季披露后才成熟，0M／6M／12M盈利层约需12／18／24个月。Separate valuation from price paths: EPS×multiple gives a conditional valuation; C shrinks toward the origin price to study the path. It cannot silently replace fair value or headline valuation. Earnings labels mature only after the last covered quarter is disclosed, taking roughly 12/18/24 months for 0M/6M/12M.估值與價格路徑分開：EPS×倍數生成條件估值；C向當前市場價收縮，研究價格路徑。後者不能無説明地替代公允價值或主表估值。盈利標籤須等覆蓋窗口最後一季披露後才成熟，0M／6M／12M盈利層約需12／18／24個月。
数据覆盖与实现检查Data coverage and implementation checks數據覆蓋與實現檢查
50个2014年起的财报季度、50份NVIDIA指引公告、2525个日价格记录。历史origin共40个，范围2016-11-23至2026-08-27；价格共同样本0M／6M／12M分别24／20／16，盈利成熟子集分别19／15／11，不能混用分母。Inputs comprise 50 financial quarters from 2014 onward, 50 NVIDIA guidance releases and 2,525 daily price records. There are 40 historical origins from 2016-11-23 through 2026-08-27. Matched price samples are 24/20/16 for 0M/6M/12M; matured earnings subsets are 19/15/11. Do not mix denominators.50個2014年起的財報季度、50份NVIDIA指引公告、2525個日價格記錄。歷史origin共40個，範圍2016-11-23至2026-08-27；價格共同樣本0M／6M／12M分別24／20／16，盈利成熟子集分別19／15／11，不能混用分母。
参考收盘价$238.90；严格滞后利率使用2026-10-02十年利率5.28个百分点。资料为现在重建的历史可得性，尚无完整原始PIT快照或ALFRED利率vintage；中途指引修改未纳入，Q4推导与拆股股数归一化保留近似限制。Reference close: $238.90. The strictly lagged yield input is 5.28 percentage points from 2026-10-02. This reconstructs historical availability today, without complete original PIT snapshots or ALFRED yield vintages. Interim guidance revisions are excluded; Q4 derivation and split-adjusted shares retain approximation limits.參考收盤價$238.90；嚴格滯後利率使用2026-10-02十年利率5.28個百分點。資料為現在重建的歷史可得性，尚無完整原始PIT快照或ALFRED利率vintage；中途指引修改未納入，Q4推導與拆股股數歸一化保留近似限制。
七项实现检查通过：未来标签扰动、未来财报扰动、校正成熟日期、月末／闰年日期映射、校正收缩恒等式、早期异常股数排除、oracle同样本对照。工作簿估值乘法与盈利窗口核对一致，税率编辑能重算。实现检查不代表预测准确率。Seven implementation checks pass: future-label and future-financial perturbations, calibration maturity, month-end/leap-year mapping, calibration-shrinkage identity, early share-anomaly exclusion and same-sample oracle comparisons. Workbook valuations and earnings windows reconcile, and tax edits recalculate. These checks do not establish predictive accuracy.七項實現檢查通過：未來標籤擾動、未來財報擾動、校正成熟日期、月末／閏年日期映射、校正收縮恆等式、早期異常股數排除、oracle同樣本對照。工作簿估值乘法與盈利窗口核對一致，税率編輯能重算。實現檢查不代表預測準確率。
SEC申报资料SEC filingsSEC申報資料 · NVIDIA最新核验指引Latest verified NVIDIA guidanceNVIDIA最新核驗指引 · FRED DGS10FRED DGS10FRED DGS10 · Nasdaq历史价格源Nasdaq historical-price sourceNasdaq歷史價格源
公开保存原创方法、派生估值和聚合验证指标；原始证据、完整工作簿、Pro记录与账户资料保留内部。记录指纹可用于识别本次发布版本，不证明这些材料当年已公开。Publish original methods, derived valuations and aggregate evaluation metrics. Raw evidence, the complete workbook, Pro records and account information remain internal. The record fingerprint identifies this publication, without proving contemporaneous historical publication.公開保存原創方法、派生估值和聚合驗證指標；原始證據、完整工作簿、Pro記錄與賬户資料保留內部。記錄指紋可用於識別本次發佈版本，不證明這些材料當年已公開。
SHA256 10f59a4bf6279a5466ce4fc61b21085b945d1957077429a6a7a681e2225abf96
停止条件与下一研究Stop condition and next research停止條件與下一研究
本轮停止在相同历史样本上继续调参。下一步建立跨期限一致的经营路径，让盈利校正受可核验经营展望约束，再冻结新增origin。6M继续检验相对价格不变的增量优势；0M保留条件估值及独立敏感性研究。Stop further tuning on these same historical samples. Next, establish a coherent operating path across horizons, constrain earnings calibration using verifiable business outlooks, and freeze new origins. Continue testing 6M incremental value against the unchanged-price baseline; retain conditional valuation and independent sensitivity research for 0M.本輪停止在相同歷史樣本上繼續調參。下一步建立跨期限一致的經營路徑，讓盈利校正受可核驗經營展望約束，再凍結新增origin。6M繼續檢驗相對價格不變的增量優勢；0M保留條件估值及獨立敏感性研究。
以后每轮研究直接列出0M、6M、12M估值、目标日期、EPS／倍数、模型版本、输入截面和验证状态；新记录追加，旧冻结预测不覆盖。Every future research round reports 0M, 6M and 12M valuations, target dates, EPS/multiples, model version, input cutoff and evaluation status. Append new records and preserve frozen historical forecasts.以後每輪研究直接列出0M、6M、12M估值、目標日期、EPS／倍數、模型版本、輸入截面和驗證狀態；新記錄追加，舊凍結預測不覆蓋。