MULTIPLE RESEARCH / HORIZON-05
NVDA H05：统一预测盈利口径的P/ENVDA H05: P/E on one forecast-earnings basisNVDA H05：統一預測盈利口徑的P/E
H05-FORECAST-BASIS-STAGE-RIDGE／协议v0.2，未晋升研究候选。市场截面为2026-10-06收盘$239.24，纽约16:30为收盘后资料可得性重建约定；原始来源实际于22:19捕获。10Y观察日10月5日、5.31%，10月6日16:15发布，不是16:00执行输入或同时保存的PIT档案。H05-FORECAST-BASIS-STAGE-RIDGE / protocol v0.2, an unpromoted research candidate. Market cutoff is the 2026-10-06 close of $239.24. The 16:30 New York convention reconstructs after-close availability; raw sources were captured at 22:19. The October 5 10Y observation is 5.31%, published October 6 at 16:15, not a 16:00 executable input or contemporaneous PIT archive.H05-FORECAST-BASIS-STAGE-RIDGE／協議v0.2，未晉升研究候選。市場截面為2026-10-06收盤$239.24，紐約16:30為收盤後資料可得性重建約定；原始來源實際於22:19捕獲。10Y觀察日10月5日、5.31%，10月6日16:15發佈，不是16:00執行輸入或同時保存的PIT檔案。
| 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 | P/E | 最终条件估值 $／股Final conditional value $/share最終條件估值 $／股 |
|---|---|---|---|---|
| 0M | 2026-10-06 → 2027-10-06 | 13.8827 | 25.9835× | $360.72 |
| 6M | 2027-04-06 → 2028-04-06 | 16.6606 | 25.9835× | $432.90 |
| 12M | 2027-10-06 → 2028-10-06 | 18.9528 | 25.9835× | $492.46 |
本轮统一预测EPS与历史倍数标签的口径，H05价格误差小于H04，但6M和12M均未通过主开发门槛，仍不及市场倍数不变基准。更高估值不等于更准确，未建立正式价格Champion。This round aligns forecast EPS with historical multiple labels. H05 price errors are smaller than H04, but both 6M and 12M fail the primary development gates and trail the held-market-P/E baseline. Higher values do not imply higher accuracy; no formal price champion exists.本輪統一預測EPS與歷史倍數標籤的口徑，H05價格誤差小於H04，但6M和12M均未通過主開發門檻，仍不及市場倍數不變基準。更高估值不等於更準確，未建立正式價格Champion。
从H04偏低到统一预测盈利口径From H04 underpricing to one forecast-earnings basis從H04偏低到統一預測盈利口徑
H04固定80%盈利现金代理、10Y加6%风险补偿和3%尾部，未完成现金留存、净现金和FCFE桥接；其低价格不能作为市场高估的证据。H05改用历史条件市场定价关系，不手动把P/E调到想要的价格。H04 fixed an 80% earnings cash proxy, 10Y plus 6% risk allowance and 3% tail without bridging retained cash, net cash or FCFE. Its low prices cannot establish market overvaluation. H05 uses historical conditional market pricing instead of manually targeting a desired price.H04固定80%盈利現金代理、10Y加6%風險補償和3%尾部，未完成現金留存、淨現金和FCFE橋接；其低價格不能作為市場高估的證據。H05改用歷史條件市場定價關係，不手動把P/E調到想要的價格。
H03-B历史标签用随后实现的NTM Core EPS，输出却乘预测EPS；H05历史标签改为当时股价／同一H03预测EPS。每个此前研究起点仅提供一条h=0标签，报价保守按次日可得，必须早于或等于新起点。当前股价不进入主模型。H03-B labels used subsequently realized NTM Core EPS while outputs multiplied forecast EPS. H05 labels instead divide historical origin prices by the same H03 forecast EPS. Each prior origin contributes one h=0 label, with quote availability conservatively dated the next day and no later than the new origin. Current P0 is excluded from the headline model.H03-B歷史標籤用隨後實現的NTM Core EPS，輸出卻乘預測EPS；H05歷史標籤改為當時股價／同一H03預測EPS。每個此前研究起點僅提供一條h=0標籤，報價保守按次日可得，必須早於或等於新起點。當前股價不進入主模型。
| 环节Step環節 | 固定定义Fixed definition固定定義 |
|---|---|
| EPS | 首10季及0M/6M/12M算法沿用H03；Core EPS=GAAP经营利润×85%／稀释股数。74%是下一季指引GM；15%是正常化研究税率。Retain H03 first10 quarters and 0M/6M/12M algorithms. Core EPS=GAAP operating profit×85%/diluted shares. 74% is next-quarter guided GM; 15% is normalized research tax.首10季及0M/6M/12M算法沿用H03；Core EPS=GAAP經營利潤×85%／稀釋股數。74%是下一季指引GM；15%是正常化研究税率。 |
| Label | yᵢ = ln(Pᵢ / forecast EPSᵢ,0) |
| Growth | gₕ = ln(forecast EPSₕ₊₁₂ / forecast EPSₕ) |
| Fit | sum残差平方＋4×标准化斜率平方；截距不罚，利率斜率≤0、增长斜率≥0；最少12个过去起点。Sum squared residuals +4×squared standardized slopes; no intercept penalty, rate slope≤0 and growth slope≥0; minimum12 prior origins.sum殘差平方＋4×標準化斜率平方；截距不罰，利率斜率≤0、增長斜率≥0；最少12個過去起點。 |
| Three horizons | 每个起点只有一个fit，三个期限共用系数和起点10Y。P/E=exp(a+bᵣzᵣ+b_gz_g)，V=Core EPS×P/E。One fit per origin; all horizons share coefficients and origin10Y. P/E=exp(a+bᵣzᵣ+b_gz_g); V=Core EPS×P/E.每個起點只有一個fit，三個期限共用係數和起點10Y。P/E=exp(a+bᵣzᵣ+b_gz_g)，V=Core EPS×P/E。 |
实际当前5.31%用于主表，没有截低利率。训练仅覆盖0.68%至4.66%，当前明确属于外推。增长特征超支持时按此前训练边界裁剪；当前三期均未触发。保持当前利率是条件情景，不是预测未来美联储政策。The headline uses the actual current 5.31%, without cutting the rate. Training covers only 0.68% to 4.66%, so the current output is explicitly extrapolated. Growth outside prior training support is clipped; none of the current horizons triggers it. Holding the origin rate is a conditional scenario, not a forecast of future Fed policy.實際當前5.31%用於主表，沒有截低利率。訓練僅覆蓋0.68%至4.66%，當前明確屬於外推。增長特徵超支持時按此前訓練邊界裁剪；當前三期均未觸發。保持當前利率是條件情景，不是預測未來美聯儲政策。
P/E的增长、回报与风险理论P/E theory: growth, payout and riskP/E的增長、回報與風險理論
25.9835倍如何算出How the 25.9835× multiple is calculated25.9835倍如何算出
ln(P/E) = 3.651542436 -0.202201334 × (5.31 − 2.934000) / 1.219122225 + 0 × z_growth
当前40个过去起点拟合出增长斜率0，故同一当前利率下三个期限P/E相同。预测后续年度增长分别约36.5%、24.4%、16.7%，但本次受约束估计没有保留正增长溢价。不能把这解释为增长无价值，也不能称已经识别NVDA的成熟阶段。The current fit uses40 prior origins and returns a zero growth slope, so all horizons have the same P/E under the shared origin rate. Subsequent annual forecast growth is about36.5%,24.4%,16.7%, but this constrained estimator retains no positive growth premium. This does not mean growth has no value or that NVIDIA’s maturity stage has been identified.當前40個過去起點擬合出增長斜率0，故同一當前利率下三個期限P/E相同。預測後續年度增長分別約36.5%、24.4%、16.7%，但本次受約束估計沒有保留正增長溢價。不能把這解釋為增長無價值，也不能稱已經識別NVDA的成熟階段。
增长特征与历史P/E标签共享预测EPS分母，可能吸收盈利误差；利率与AI周期又存在相关性。斜率方向是经济先验，不是因果估计。当前25.9835倍不能直接当成公允价值真值或分析师共识倍数。The growth feature and P/E labels share a forecast-EPS denominator and may absorb earnings errors; rates also correlate with the AI cycle. Slope signs are economic priors, not causal estimates. The current25.9835× is neither verified fair-value truth nor an analyst-consensus multiple.增長特徵與歷史P/E標籤共享預測EPS分母，可能吸收盈利誤差；利率與AI週期又存在相關性。斜率方向是經濟先驗，不是因果估計。當前25.9835倍不能直接當成公允價值真值或分析師共識倍數。
| 期限Horizon期限 | 本期EPS窗口Current EPS window本期EPS窗口 | 后续EPS窗口末日Next EPS-window end後續EPS窗口末日 | 后续预测增长Subsequent forecast growth後續預測增長 | 增长裁剪Growth clipped增長裁剪 |
|---|---|---|---|---|
| 0M | 2026-10-06 → 2027-10-06 | 2028-10-06 | 36.5% | false |
| 6M | 2027-04-06 → 2028-04-06 | 2029-04-06 | 24.4% | false |
| 12M | 2027-10-06 → 2028-10-06 | 2029-10-06 | 16.7% | false |
估值更高，但仍不及简单基准Higher values, still behind simple baselines估值更高，但仍不及簡單基準
| 期限Horizon期限 | H05 | 同日H03-BSame-date H03-B同日H03-B | H04结构诊断H04 structural diagnosticH04結構診斷 | 市场P/E不变基准Held-market-P/E baseline市場P/E不變基準 |
|---|---|---|---|---|
| 0M | $360.72 | $300.39 | $203.57 | $239.24 |
| 6M | $432.90 | $304.19 | $214.74 | $287.11 |
| 12M | $492.46 | $346.07 | $215.49 | $326.61 |
本轮保留40个已观察历史起点；主价格样本28／26／24，所有模型使用相同起点。MALE是平均绝对对数误差，越低越好，不是准确率。H05对同样本H04的6M／12M MALE降低约45.9%／53.0%，但两期都未胜过市场倍数不变，主开发门槛失败。Retain40 previously observed origins; primary price samples are28/26/24, matched across models. MALE is mean absolute log error, lower is better, not an accuracy rate. H05 reduces same-sample H04 MALE by about45.9%/53.0%at6M/12M, but both horizons trail held-market-P/E and fail primary development gates.本輪保留40個已觀察歷史起點；主價格樣本28／26／24，所有模型使用相同起點。MALE是平均絕對對數誤差，越低越好，不是準確率。H05對同樣本H04的6M／12M MALE降低約45.9%／53.0%，但兩期都未勝過市場倍數不變，主開發門檻失敗。
| 期限Horizon期限 | 模型Model模型 | N | MALE | 中位APEMedian APE中位APE | P90 APE |
|---|---|---|---|---|---|
| 0M | H05 | 28 | 0.408 | 31.4% | 88.7% |
| 0M | MarketPEHeld | 28 | 0.000 | 0.0% | 0.0% |
| 0M | PriceUnchanged | 28 | 0.000 | 0.0% | 0.0% |
| 0M | H04 | 28 | 0.955 | 62.5% | 89.4% |
| 6M | H05 | 26 | 0.606 | 48.7% | 91.9% |
| 6M | MarketPEHeld | 26 | 0.340 | 26.1% | 52.4% |
| 6M | PriceUnchanged | 26 | 0.362 | 32.6% | 52.2% |
| 6M | H04 | 26 | 1.120 | 62.2% | 93.4% |
| 12M | H05 | 24 | 0.674 | 38.2% | 81.1% |
| 12M | MarketPEHeld | 24 | 0.445 | 25.0% | 68.7% |
| 12M | PriceUnchanged | 24 | 0.620 | 41.9% | 69.9% |
| 12M | H04 | 24 | 1.434 | 70.6% | 94.8% |
| 期限Horizon期限 | 对照基准Baseline對照基準 | MALE改善MALE improvementMALE改善 | 配对95%误差差区间Paired95% loss-difference interval配對95%誤差差區間 | 相对该基准门槛Gate versus this baseline相對該基準門檻 |
|---|---|---|---|---|
| 6M | MarketPEHeld | -78.4% | [0.091, 0.436] | false |
| 6M | PriceUnchanged | -67.5% | [0.042, 0.443] | false |
| 6M | H04 | 45.9% | [-0.834, -0.318] | true |
| 12M | MarketPEHeld | -51.6% | [0.094, 0.435] | false |
| 12M | PriceUnchanged | -8.9% | [-0.197, 0.345] | false |
| 12M | H04 | 53.0% | [-0.984, -0.508] | true |
0M市场倍数基准必等于$239.24，零误差是恒等式，不是公允价值验证。H04表列值扣假想现金，未桥接实际现金留存；新旧输出口径不同，不能把值变高全归因于一个参数或把H04比较当完整股权估值验证。The0M market-P/E baseline necessarily equals$239.24; its zero error is an identity, not fair-value validation. H04 is after hypothetical proxy cash and lacks retained-cash bridging. The model bases differ, so higher outputs cannot be attributed to one parameter or the H04 comparison treated as full equity-valuation validation.0M市場倍數基準必等於$239.24，零誤差是恆等式，不是公允價值驗證。H04表列值扣假想現金，未橋接實際現金留存；新舊輸出口徑不同，不能把值變高全歸因於一個參數或把H04比較當完整股權估值驗證。
与H03比较：另列完全共同样本H03 comparison: separate fully matched sample與H03比較：另列完全共同樣本
六模型都可计算的共同样本仍为24／20／16。不得拿扩大的主样本分数直接与旧研究分数比较。H05的同样本MALE小于H03-B，仍未胜过简单基准；EPS方法没有改善，预测EPS口径倍数也可能补偿EPS误差。The six-model common samples remain24/20/16. Do not compare expanded primary-sample scores directly with older-study scores. H05 has lower same-sample MALE than H03-B but still trails simple baselines; EPS methods are unchanged, and forecast-basis multiples may compensate for EPS errors.六模型都可計算的共同樣本仍為24／20／16。不得拿擴大的主樣本分數直接與舊研究分數比較。H05的同樣本MALE小於H03-B，仍未勝過簡單基準；EPS方法沒有改善，預測EPS口徑倍數也可能補償EPS誤差。
| 期限Horizon期限 | 模型Model模型 | N | MALE | 中位APEMedian APE中位APE | P90 APE |
|---|---|---|---|---|---|
| 0M | H05 | 24 | 0.442 | 35.6% | 89.8% |
| 0M | MarketPEHeld | 24 | 0.000 | 0.0% | 0.0% |
| 0M | PriceUnchanged | 24 | 0.000 | 0.0% | 0.0% |
| 0M | H04 | 24 | 1.026 | 65.1% | 89.6% |
| 0M | H03B | 24 | 0.559 | 46.6% | 77.9% |
| 0M | H03A | 24 | 0.517 | 45.8% | 134.9% |
| 6M | H05 | 20 | 0.680 | 59.2% | 101.4% |
| 6M | MarketPEHeld | 20 | 0.369 | 26.8% | 61.3% |
| 6M | PriceUnchanged | 20 | 0.364 | 28.2% | 57.0% |
| 6M | H04 | 20 | 1.207 | 69.8% | 94.4% |
| 6M | H03B | 20 | 0.700 | 53.5% | 82.8% |
| 6M | H03A | 20 | 0.765 | 72.2% | 130.5% |
| 12M | H05 | 16 | 0.740 | 41.8% | 86.7% |
| 12M | MarketPEHeld | 16 | 0.503 | 31.8% | 74.7% |
| 12M | PriceUnchanged | 16 | 0.642 | 45.2% | 70.7% |
| 12M | H04 | 16 | 1.527 | 65.0% | 96.0% |
| 12M | H03B | 16 | 0.944 | 51.8% | 91.7% |
| 12M | H03A | 16 | 0.911 | 71.2% | 130.7% |
这些是已观察历史上的序贯开发诊断。按时间只用过去标签控制了信息泄漏，却不能称新的独立样本外验证。未来标签尚未成熟，不能晋升正式模型。These are sequential development diagnostics on already observed history. Prior-only labels control information timing but do not create a new independent out-of-sample test. Future labels have not matured, so the model is not promoted.這些是已觀察歷史上的序貫開發診斷。按時間只用過去標籤控制了信息泄漏，卻不能稱新的獨立樣本外驗證。未來標籤尚未成熟，不能晉升正式模型。
未评分的条件情景Unscored conditional scenarios未評分的條件情景
| 情景Scenario情景 | 0M | 6M | 12M |
|---|---|---|---|
| 10Y减1个百分点10Y minus1pp10Y減1個百分點 | $425.80 | $511.00 | $581.30 |
| 10Y加1个百分点10Y plus1pp10Y加1個百分點 | $305.59 | $366.74 | $417.20 |
| 增长特征减0.10Growth feature minus0.10增長特徵減0.10 | $360.72 | $432.90 | $492.46 |
| 增长特征加0.10Growth feature plus0.10增長特徵加0.10 | $360.72 | $432.90 | $492.46 |
| 仅诊断：利率裁剪至4.66%Diagnostic only: rate clipped to4.66%僅診斷：利率裁剪至4.66% | $401.78 | $482.18 | $548.52 |
情景均用同一fit，未评分、未选择，不是置信区间。利率裁剪会把0M抬到$401.78，所以只作诊断，不替换实际5.31%的主值。增长系数为0使本轮增长扰动不改变输出。All scenarios use the same fit, are unscored and unselected, and are not confidence intervals. Rate clipping raises0M to$401.78, so it is diagnostic only and never substitutes for the actual5.31% headline. The zero growth slope makes these growth perturbations leave outputs unchanged.情景均用同一fit，未評分、未選擇，不是置信區間。利率裁剪會把0M抬到$401.78，所以只作診斷，不替換實際5.31%的主值。增長係數為0使本輪增長擾動不改變輸出。
回购授权如何进入后续研究How repurchase authorization enters future research回購授權如何進入後續研究
9月28日新增$150B回购授权，剩余$235B。授权不是已执行回购，不是可以直接加到股权价值的现金，也不证明股数已减少。本轮不改EPS；后续需要核验实际现金支出、股数减少和经营变化。September28 added$150B in repurchase authorization, with$235B remaining. Authorization is not executed repurchase, cash to add directly to equity value or proof that shares have declined. EPS is unchanged this round; later research must verify actual cash spent, share reductions and operating changes.9月28日新增$150B回購授權，剩餘$235B。授權不是已執行回購，不是可以直接加到股權價值的現金，也不證明股數已減少。本輪不改EPS；後續需要核驗實際現金支出、股數減少和經營變化。
验证状态与下一研究Validation status and next research驗證狀態與下一研究
独立实现及未来数据隔离检查通过；历史价格主门槛失败，未晋升、未自动交易。旧H02/H03/H04冻结记录保持原样。公开仅保留原创方法、派生预测和聚合结果，原始证据与Pro／账户资料继续留在内部。Independent implementation and future-data-isolation checks pass. Primary historical price gates fail; no promotion or automatic trading. Frozen H02/H03/H04 records remain intact. Publish only original methods, derived forecasts and aggregates, retaining raw evidence and Pro/account data internally.獨立實現及未來數據隔離檢查通過；歷史價格主門檻失敗，未晉升、未自動交易。舊H02/H03/H04凍結記錄保持原樣。公開僅保留原創方法、派生預測和聚合結果，原始證據與Pro／賬户資料繼續留在內部。
下一步用不共享预测EPS分母的业务指标检验成长及市场制度变化，桥接实际资本回报，并保留当前市场倍数基准；真正新增证据需要冻结新起点并等待标签成熟。本轮不因估值变高而继续调参。Next, test growth and market-regime changes with operating indicators that do not share forecast-EPS denominators, bridge actual capital returns and retain the current market-multiple baseline. New evidence requires frozen new origins and mature future labels. Stop retuning this round despite higher values.下一步用不共享預測EPS分母的業務指標檢驗成長及市場制度變化，橋接實際資本回報，並保留當前市場倍數基準；真正新增證據需要凍結新起點並等待標籤成熟。本輪不因估值變高而繼續調參。
Fed H15利率与发布时间Fed H15 rates and release timeFed H15利率與發佈時間
Protocol SHA256 53a2315eb1cc929930839da2568b2d448c512fbe9772725358716721aee9886f
Record SHA256 a6825e6295840cdd281b93a4755bb67ee580dcdc9f82ea516df8d52e699a1821