NVDA / H07–H10 / THREE-YEAR COMPARISON
H07–H10重新比较：过去三年的股价预测误差H07–H10 revisited: three years of price forecast errorsH07–H10重新比較：過去三年的股價預測誤差
本轮按用户要求，只评价2023-10-06至2026-10-06期间发起的预测。使用相同12个季度起点，实际股价标签截止2026-10-06收盘；6M到期10个，12M到期8个。At the user’s request, this study evaluates only forecasts initiated from 2023-10-06 through 2026-10-06. Use the same twelve quarterly origins and actual price labels through the 2026-10-06 close: ten mature 6M observations and eight mature 12M observations.本輪按用户要求，只評價2023-10-06至2026-10-06期間發起的預測。使用相同12個季度起點，實際股價標籤截止2026-10-06收盤；6M到期10個，12M到期8個。
| 版本Version版本 | 6M平均误差 · 10点6M mean error · 10 points6M平均誤差 · 10點 | 12M平均误差 · 8点12M mean error · 8 points12M平均誤差 · 8點 | 分析师平均差异 · 12点Mean analyst gap · 12 points分析師平均差異 · 12點 |
|---|---|---|---|
| H07 | 20.83% | 18.95% | 22.64% |
| H08 | 20.73% | 18.67% | 23.14% |
| H09 | 14.33% | 14.91% | 16.86% |
| H10 | 21.61% | 22.07% | 18.64% |
在这个三年窗口内，H09的6M、12M平均绝对百分比误差和平均绝对对数误差均最低，三年分析师平均目标价差异也最小。H09应作为下一轮优先研究候选；12M中位误差则由H08最低。Within this three-year window, H09 has the lowest mean absolute percentage error and mean absolute log error at both horizons, plus the smallest mean gap to analyst targets. H09 is the priority candidate for further research; H08 has the lowest 12M median error.在這個三年窗口內，H09的6M、12M平均絕對百分比誤差和平均絕對對數誤差均最低，三年分析師平均目標價差異也最小。H09應作為下一輪優先研究候選；12M中位誤差則由H08最低。
三年回测如何定义How the three-year backtest is defined三年回測如何定義
按预测发起日筛选，实际12个起点为2023-11-22至2026-08-27；不是三年每日预测，也不是按目标日反向选取更早预测。当前2026-10-06截面单列，不加入历史评分。尚未到期的6M有2个、12M有4个，不使用未来价格。Filter by forecast-origin date. The twelve available origins run from 2023-11-22 to 2026-08-27. These are quarterly forecasts, not daily forecasts or older forecasts selected by their target dates. The current 2026-10-06 snapshot is separate. Two 6M and four 12M outcomes remain pending; no future prices are used.按預測發起日篩選，實際12個起點為2023-11-22至2026-08-27；不是三年每日預測，也不是按目標日反向選取更早預測。當前2026-10-06截面單列，不加入歷史評分。尚未到期的6M有2個、12M有4個，不使用未來價格。
| 期限Horizon期限 | 历史起点Historical origins歷史起點 | 已到期／待到期Mature / pending已到期／待到期 | 到期样本起点范围Mature-origin range到期樣本起點範圍 | 实际目标交易日范围Actual target-trading-date range實際目標交易日範圍 |
|---|---|---|---|---|
| 6M | 12 | 10 / 2 | 2023-11-22 → 2026-02-26 | 2024-05-22 → 2026-08-26 |
| 12M | 12 | 8 / 4 | 2023-11-22 → 2025-08-28 | 2024-11-22 → 2026-08-28 |
仅改变评价窗口。H07–H10各自的原公式、参数与当时可用训练历史保持不变；独立复算四个版本与冻结预测一致。若把训练也限制为三年，会产生新模型，本轮没有这样做。旧十年档案保留，未重算，也不作为本页排名依据。Only the evaluation window changes. Each H07–H10 formula, parameter rule and contemporaneously available training history remains unchanged. Independent reruns match the frozen forecasts. Restricting training itself to three years would create new models and is not done here. Earlier ten-year records remain archived, are not rescored and do not determine this page’s ranking.僅改變評價窗口。H07–H10各自的原公式、參數與當時可用訓練歷史保持不變；獨立復算四個版本與凍結預測一致。若把訓練也限制為三年，會產生新模型，本輪沒有這樣做。舊十年檔案保留，未重算，也不作為本頁排名依據。
模型0M等于当期市场价格，不能当作独立预测来计算准确率。6M与12M均与目标日当日或之后首个交易日收盘比较；不要求未来EPS已公布才给价格评分。Model 0M equals the observed market price and is not scored as an independent prediction. Both future horizons use the first trading close on or after the target date. Price scoring does not require the future EPS window to have been published.模型0M等於當期市場價格，不能當作獨立預測來計算準確率。6M與12M均與目標日當日或之後首個交易日收盤比較；不要求未來EPS已公佈才給價格評分。
实际股价预测误差Realized-price forecast errors實際股價預測誤差
平均误差为MAPE＝平均值｜预测价÷实际价−1｜。MALE为平均绝对对数误差。两者越低越好，不能把1−误差写成准确率。所有版本在同一批样本上评分。MAPE is the mean of |forecast / actual − 1|. MALE is mean absolute log error. Lower is better for both; 1 minus error is not an accuracy rate. Every version uses identical matched observations.平均誤差為MAPE＝平均值｜預測價÷實際價−1｜。MALE為平均絕對對數誤差。兩者越低越好，不能把1−誤差寫成準確率。所有版本在同一批樣本上評分。
| 期限Horizon期限 | 模型Model模型 | N | MAPE | MALE | 中位误差Median APE中位誤差 | P90误差P90 APEP90誤差 | 平均带符号误差Mean signed error平均帶符號誤差 |
|---|---|---|---|---|---|---|---|
| 6M | H07 | 10 | 20.83% | 0.226953 | 20.03% | 34.86% | -4.87% |
| 6M | H08 | 10 | 20.73% | 0.224815 | 19.86% | 34.57% | -4.37% |
| 6M | H09 | 10 | 14.33% | 0.147441 | 12.21% | 30.89% | -1.72% |
| 6M | H10 | 10 | 21.61% | 0.239805 | 21.94% | 34.83% | -6.83% |
| 6M | 价格不变基准Unchanged-price baseline價格不變基準 | 10 | 21.14% | 0.253708 | 19.73% | 37.75% | -18.88% |
| 12M | H07 | 8 | 18.95% | 0.240083 | 7.67% | 41.13% | -16.85% |
| 12M | H08 | 8 | 18.67% | 0.236089 | 7.08% | 40.36% | -16.32% |
| 12M | H09 | 8 | 14.91% | 0.162365 | 10.46% | 26.31% | -3.65% |
| 12M | H10 | 8 | 22.07% | 0.282045 | 15.23% | 47.97% | -20.85% |
| 12M | 价格不变基准Unchanged-price baseline價格不變基準 | 8 | 32.20% | 0.420237 | 32.34% | 47.51% | -32.20% |
H09较H07的MAPE降低31.18%（6M）和21.29%（12M）。H09的P90误差也最低，分别30.89%与26.31%。但是12M中位误差为H08 7.08%、H07 7.67%、H09 10.46%，H09并非每项指标都胜出。H09 reduces MAPE versus H07 by 31.18% at 6M and 21.29% at 12M. H09 also has the lowest P90 errors, 30.89% and 26.31%. However, 12M median errors are 7.08% for H08, 7.67% for H07 and 10.46% for H09: H09 does not win every metric.H09較H07的MAPE降低31.18%（6M）和21.29%（12M）。H09的P90誤差也最低，分別30.89%與26.31%。但是12M中位誤差為H08 7.08%、H07 7.67%、H09 10.46%，H09並非每項指標都勝出。
H09的配对误差差区间H09 paired loss-difference intervalsH09的配對誤差差區間
| 期限Horizon期限 | 基准Baseline基準 | MALE改善MALE improvementMALE改善 | 95% CI |
|---|---|---|---|
| 6M | H07 | 35.03% | [-0.15946, -0.01988] |
| 6M | PriceUnchanged | 41.89% | [-0.17092, -0.00541] |
| 12M | H07 | 32.37% | [-0.21691, 0.05706] |
| 12M | PriceUnchanged | 61.36% | [-0.32878, -0.17802] |
四起点移动区块、5000次抽样，与既有研究方法一致。6M相对H07的区间在零以下；12M相对H07仍跨零。8至10个重叠样本下，区间仅作探索性诊断，也没有校正多版本及多窗口选择。Use four-origin moving blocks and 5,000 draws, consistent with earlier studies. The 6M interval versus H07 is below zero; the 12M interval still includes zero. With only eight to ten overlapping observations, these intervals are exploratory and are not corrected for selection across versions and windows.四起點移動區塊、5000次抽樣，與既有研究方法一致。6M相對H07的區間在零以下；12M相對H07仍跨零。8至10個重疊樣本下，區間僅作探索性診斷，也沒有校正多版本及多窗口選擇。
方向命中不能代替误差比较Direction does not replace error comparison方向命中不能代替誤差比較
| 期限Horizon期限 | 版本Version版本 | 方向命中Direction accuracy方向命中 | 实际上涨／下跌数Actual up / down實際上漲／下跌數 | 一直预测上涨Always-up baseline一直預測上漲 |
|---|---|---|---|---|
| 6M | H07 | 70.00% | 9 / 1 | 90.00% |
| 6M | H08 | 70.00% | 9 / 1 | 90.00% |
| 6M | H09 | 90.00% | 9 / 1 | 90.00% |
| 6M | H10 | 70.00% | 9 / 1 | 90.00% |
| 12M | H07 | 75.00% | 8 / 0 | 100.00% |
| 12M | H08 | 75.00% | 8 / 0 | 100.00% |
| 12M | H09 | 100.00% | 8 / 0 | 100.00% |
| 12M | H10 | 75.00% | 8 / 0 | 100.00% |
H09的方向命中为90%／100%，恰好等于一直预测上涨。6M只有一个实际下跌，H09没有预测到；12M的8个实际结果全部上涨。因此不能把这组结果解释为已验证拐点预测能力。H09 direction accuracy is 90% / 100%, exactly equal to always predicting up. The sole 6M decline was missed by H09, and all eight 12M outcomes rose. These results do not establish turning-point prediction ability.H09的方向命中為90%／100%，恰好等於一直預測上漲。6M只有一個實際下跌，H09沒有預測到；12M的8個實際結果全部上漲。因此不能把這組結果解釋為已驗證拐點預測能力。
同一三年内的分析师目标价比较Analyst-target comparison within the same three years同一三年內的分析師目標價比較
此处为12个同日配对：模型12M(t)对StockAnalysis All Analysts平均一年目标价(t)。它衡量与分析师预期的距离；不是未来实际股价误差，也没有与上表合并为总分。These are twelve same-date pairs: model 12M(t) versus the StockAnalysis All Analysts average one-year target(t). This measures distance from analyst expectations, not future realized-price error; it is not combined with the price table into one total score.此處為12個同日配對：模型12M(t)對StockAnalysis All Analysts平均一年目標價(t)。它衡量與分析師預期的距離；不是未來實際股價誤差，也沒有與上表合併為總分。
| 版本Version版本 | N | 平均绝对差异Mean absolute gap平均絕對差異 | 中位差异Median gap中位差異 | MALE | 平均带符号差异Mean signed gap平均帶符號差異 | ±10% | ±20% |
|---|---|---|---|---|---|---|---|
| H07 | 12 | 22.64% | 20.79% | 0.222992 | 9.76% | 3/12 | 5/12 |
| H08 | 12 | 23.14% | 21.10% | 0.226602 | 10.37% | 3/12 | 5/12 |
| H09 | 12 | 16.86% | 8.81% | 0.146112 | 13.24% | 7/12 | 9/12 |
| H10 | 12 | 18.64% | 19.39% | 0.191669 | 5.09% | 3/12 | 7/12 |
H09最接近共识，但平均带符号差异为+13.24%，高于H07的+9.76%；绝对差异更小不代表整体更保守。H10的平均带符号差异为+5.09%。分析师目标价没有用于任何版本的本次拟合。H09 is closest in absolute terms, but its mean signed gap is +13.24%, versus H07’s +9.76%; smaller absolute gaps do not mean less optimism overall. H10’s mean signed gap is +5.09%. Analyst targets are not used to fit any version in this comparison.H09最接近共識，但平均帶符號差異為+13.24%，高於H07的+9.76%；絕對差異更小不代表整體更保守。H10的平均帶符號差異為+5.09%。分析師目標價沒有用於任何版本的本次擬合。
记录于2026-10-07从图表悬停读取，原始历史发布时间、修订与拆股沿革尚未独立核验。只公开聚合比较，原始Pro目标价留内部。Records were read from chart tooltips on 2026-10-07. Original publication times, revisions and split lineage remain unverified. Only aggregate comparisons are public; raw Pro targets remain internal.記錄於2026-10-07從圖表懸停讀取，原始歷史發佈時間、修訂與拆股沿革尚未獨立核驗。只公開聚合比較，原始Pro目標價留內部。
StockAnalysis平均目标价图表StockAnalysis average-target chartStockAnalysis平均目標價圖表
四个版本的当前0M、6M、12M输出Current 0M, 6M and 12M outputs for all four versions四個版本的當前0M、6M、12M輸出
输入与实际收盘截面为2026-10-06；本轮于2026-10-07纽约09:07再次核验$239.24收盘。财报与指引沿用冻结资料，不能称为全部输入刷新至10月7日。Input and actual-close cutoff is 2026-10-06. The $239.24 close was rechecked on 2026-10-07 at 09:07 New York. Financial and guidance inputs remain frozen; this is not a complete input refresh to October 7.輸入與實際收盤截面為2026-10-06；本輪於2026-10-07紐約09:07再次核驗$239.24收盤。財報與指引沿用凍結資料，不能稱為全部輸入刷新至10月7日。
| 版本／身份Version / role版本／身份 | 0M | 6M | 12M |
|---|---|---|---|
| H07 · 保留最终参考 · retained final reference · 保留最終參考 | $239.24 | $287.11 | $326.61 |
| H08 · 研究诊断 · research diagnostic · 研究診斷 | $239.24 | $287.35 | $326.89 |
| H09 · 三年优先研究候选 · priority three-year candidate · 三年優先研究候選 | $239.24 | $286.91 | $326.21 |
| H10 · 研究诊断 · research diagnostic · 研究診斷 | $239.24 | $282.35 | $317.42 |
| 期限Horizon期限 | 目标日Target目標日 | 盈利窗口末日Earnings-window end盈利窗口末日 |
|---|---|---|
| 0M | 2026-10-06 | 2027-10-06 |
| 6M | 2027-04-06 | 2028-04-06 |
| 12M | 2027-10-06 | 2028-10-06 |
各分支EPS、倍数与公式Branch-specific EPS, multiples and formulas各分支EPS、倍數與公式
| 版本Version版本 | 期限Horizon期限 | Core EPS | P/E | USD |
|---|---|---|---|---|
| RA-NVDA-HORIZON-07-v0.1 | 0M | 13.882658 | 17.233011 | 239.24 |
| RA-NVDA-HORIZON-07-v0.1 | 6M | 16.660556 | 17.233011 | 287.11 |
| RA-NVDA-HORIZON-07-v0.1 | 12M | 18.952766 | 17.233011 | 326.61 |
| RA-NVDA-HORIZON-08-v0.1 | 0M | 14.158083 | 16.897768 | 239.24 |
| RA-NVDA-HORIZON-08-v0.1 | 6M | 17.005343 | 16.897768 | 287.35 |
| RA-NVDA-HORIZON-08-v0.1 | 12M | 19.344990 | 16.897768 | 326.89 |
| RA-NVDA-HORIZON-09-v0.1 | 0M | 13.866187 | 17.253482 | 239.24 |
| RA-NVDA-HORIZON-09-v0.1 | 6M | 16.628864 | 17.253482 | 286.91 |
| RA-NVDA-HORIZON-09-v0.1 | 12M | 18.907019 | 17.253482 | 326.21 |
| RA-NVDA-HORIZON-10-v0.1 | 0M | 13.882658 | 17.233011 | 239.24 |
| RA-NVDA-HORIZON-10-v0.1 | 6M | 16.660556 | 16.947171 | 282.35 |
| RA-NVDA-HORIZON-10-v0.1 | 12M | 18.952766 | 16.748029 | 317.42 |
每个盈利窗口从目标日开始，覆盖未来12个月。各分支均满足价格＝本分支Core EPS×本分支PE。H07／H08／H09用各自PE0＝P0／EPS0并保持倍数；H10用P_h＝P0×exp[β×log(EPS_h／EPS0)]，当前β＝0.908303，对应PE_h＝P_h／EPS_h。不得混用不同版本的EPS和PE。Each earnings window begins at its target date and covers the next twelve months. Every branch satisfies price = its own Core EPS × its own P/E. H07/H08/H09 hold their own PE0=P0/EPS0; H10 uses P_h=P0×exp[beta×log(EPS_h/EPS0)], current beta=0.908303, with PE_h=P_h/EPS_h. Do not mix EPS and P/E across versions.每個盈利窗口從目標日開始，覆蓋未來12個月。各分支均滿足價格＝本分支Core EPS×本分支PE。H07／H08／H09用各自PE0＝P0／EPS0並保持倍數；H10用P_h＝P0×exp[β×log(EPS_h／EPS0)]，當前β＝0.908303，對應PE_h＝P_h／EPS_h。不得混用不同版本的EPS和PE。
本轮结论与验证状态Conclusion and validation status本輪結論與驗證狀態
在这个三年窗口内，H09的6M、12M平均绝对百分比误差和平均绝对对数误差均最低，三年分析师平均目标价差异也最小。H09应作为下一轮优先研究候选；12M中位误差则由H08最低。Within this three-year window, H09 has the lowest mean absolute percentage error and mean absolute log error at both horizons, plus the smallest mean gap to analyst targets. H09 is the priority candidate for further research; H08 has the lowest 12M median error.在這個三年窗口內，H09的6M、12M平均絕對百分比誤差和平均絕對對數誤差均最低，三年分析師平均目標價差異也最小。H09應作為下一輪優先研究候選；12M中位誤差則由H08最低。
评价重点已切换到三年，本页不以十年结果决定排名。H09是这个窗口下更值得继续优化的起点；当前发布的是模型比较，未自动替换H07最终参考，也未晋升正式Champion。The evaluation focus has switched to three years; ten-year results do not determine this ranking. H09 is the stronger starting point for further research under this window. This publication compares models; it does not automatically replace the H07 final reference or promote a formal Champion.評價重點已切換到三年，本頁不以十年結果決定排名。H09是這個窗口下更值得繼續優化的起點；當前發佈的是模型比較，未自動替換H07最終參考，也未晉升正式Champion。
该窗口是在已有研究后按用户要求选定，属于回顾性窗口敏感性分析。季度样本少、预测期限重叠，且上涨阶段占主导；不是独立样本外验证。本轮只检查股价误差与共识差异，没有声称EPS准确率因此提高。This window was requested after earlier research, making the exercise a retrospective window-sensitivity analysis. Quarterly samples are small, horizons overlap and rising markets dominate. It is not independent out-of-sample validation. This round assesses price errors and consensus gaps and makes no claim of improved EPS accuracy.該窗口是在已有研究後按用户要求選定，屬於回顧性窗口敏感性分析。季度樣本少、預測期限重疊，且上漲階段佔主導；不是獨立樣本外驗證。本輪只檢查股價誤差與共識差異，沒有聲稱EPS準確率因此提高。
独立复算及一致性检查1364项通过；旧研究和分析师记录8965个文件保持原样。公开仅派生当前模型值、方法和聚合结果；逐期价格、目标价、预测与训练记录保留内部。All 1364 independent rerun and consistency checks pass; 8965 older research and benchmark files are unchanged. Publish only current derived model values, methods and aggregates; retain per-origin prices, targets, forecasts and training records internally.獨立復算及一致性檢查1364項通過；舊研究和分析師記錄8965個文件保持原樣。公開僅派生當前模型值、方法和聚合結果；逐期價格、目標價、預測與訓練記錄保留內部。
滚动预测时点验证方法Rolling-origin validation method滾動預測時點驗證方法
Protocol SHA256 8e1cc7be769d8f15dc024e400977a0269e73b9a08e21db46a7125b845515791e
Record SHA256 61c1ede03edc246e8676b11c3ce9d069a7bf4a0c9c4876b3e0e2c991b5e8848e
H10原研究H10 original studyH10原研究 · H09原研究H09 original studyH09原研究 · H08档案H08 archiveH08檔案 · H07档案H07 archiveH07檔案 · H06档案H06 archiveH06檔案 · H05档案H05 archiveH05檔案 · H04档案H04 archiveH04檔案 · H03档案H03 archiveH03檔案 · H02档案H02 archiveH02檔案