NVDA / V1 / RETROSPECTIVE EVALUATION
V1 已有历史检验记录V1 available historical evaluationV1 已有歷史檢驗記錄
以下是源对话报告的回顾性重建摘要，不是本站独立复跑结果、十年完整回测或样本外胜率。完整8期逐行底稿、价格序列与观察截止日尚未取得，因此本页不能逐行复算100%和207.06%。These are source-reported retrospective reconstruction results, not an independent rerun, complete ten-year backtest or out-of-sample win rate. The full eight-row working file, price series and observation cutoff are unavailable here; 100% and 207.06% cannot be recomputed row by row from this page.以下是源對話報告的回顧性重建摘要，不是本站獨立復跑結果、十年完整回測或樣本外勝率。完整8期逐行底稿、價格序列與觀察截止日尚未取得，因此本頁不能逐行復算100%和207.06%。
已有汇总数据Available summary data已有彙總數據
| 指标Metric指標 | 源记录数值Source-reported value源記錄數值 | 解读边界Interpretation解讀邊界 |
|---|---|---|
| 成熟样本数Mature samples成熟樣本數 | 8 | 窗口重叠，不是8个独立试验。Overlapping windows, not eight independent trials.窗口重疊，不是8個獨立試驗。 |
| 12M最高收盘达到或超过估值12M maximum close at or above target12M最高收盤達到或超過估值 | 100.00% | 8/8；是历史频率，不是未来概率。8/8; a historical frequency, not a future probability.8/8；是歷史頻率，不是未來概率。 |
| 平均最高收盘价／估值Mean maximum close / target平均最高收盤價／估值 | 207.06% | 不是投资收益率、准确率或目标价放大系数。Not an investment return, accuracy rate or target multiplier.不是投資收益率、準確率或目標價放大係數。 |
| 期末价格MAPEEndpoint price MAPE期末價格MAPE | 42.55% | 分母是实际期末价；越低越好。Denominator is actual endpoint price; lower is better.分母是實際期末價；越低越好。 |
| 不变价格基准MAPEUnchanged-price baseline MAPE不變價格基準MAPE | 31.44% | 模型误差比该基准高11.11个百分点。Model error is 11.11 percentage points higher than this baseline.模型誤差比該基準高11.11個百分點。 |
| 起点股价已高于估值Initial price already above target起點股價已高於估值 | 6 / 8 | 较低门槛会提高达到比例及触及率。Lower thresholds can increase attainment ratios and hit frequencies.較低門檻會提高達到比例及觸及率。 |
统计口径不能悄悄改变Keep the metric definition fixed統計口徑不能悄悄改變
Hitᵢ = 1(max closing price in the next 12 months ≥ original valuationᵢ) Attainmentᵢ = max closing price in the next 12 months / original valuationᵢ Mean attainment = mean(Attainmentᵢ) MAPE = mean(|original valuationᵢ − endpoint priceᵢ| / endpoint priceᵢ)
本摘要沿用后续对话确认的规则：上行、下行估值都比较未来最高收盘价，不对下行目标改用最低价。它与最初研究稿中“下行目标用最低价、接近起点目标排除”的方案不同，不能共用分母或合并成绩。比较对象为原情景加权估值，不是单独Base价格。This summary follows the later conversation’s rule: compare future maximum closing price with both upward and downward valuations, without switching downward targets to minimum prices. This differs from the original report’s minimum-price rule for downward targets and near-origin exclusion; denominators and results cannot be combined. The comparison target is the original scenario-weighted valuation, not the Base price alone.本摘要沿用後續對話確認的規則：上行、下行估值都比較未來最高收盤價，不對下行目標改用最低價。它與最初研究稿中“下行目標用最低價、接近起點目標排除”的方案不同，不能共用分母或合併成績。比較對象為原情景加權估值，不是單獨Base價格。
六个下行估值案例：原输入摘录Six downward-valuation examples: original input excerpts六個下行估值案例：原輸入摘錄
这六行只是对话列出的下行案例，不是完整8期回测表。没有最高收盘价、期末价或逐期触及日期的原始数据，不补造对应列，也不从这六行重新计算总体统计。EPS和P/E为Base输入，加权价格是三情景组合，不能直接由这两列相乘复现。These six rows are the downward examples listed in the conversation, not the complete eight-sample backtest table. Raw maximum closes, endpoint prices and individual attainment dates are unavailable; those columns are not invented and aggregate statistics are not recalculated from this subset. EPS and P/E are Base inputs; the weighted price combines three scenarios and cannot be reproduced by multiplying these two columns alone.這六行只是對話列出的下行案例，不是完整8期回測表。沒有最高收盤價、期末價或逐期觸及日期的原始數據，不補造對應列，也不從這六行重新計算總體統計。EPS和P/E為Base輸入，加權價格是三情景組合，不能直接由這兩列相乘復現。
| 起点Origin起點 | 参考价格 $Reference price $參考價格 $ | Base目标NTM EPS $Base target NTM EPS $Base目標NTM EPS $ | Base P/E | 加权估值 $Weighted valuation $加權估值 $ |
|---|---|---|---|---|
| 2024-02-22 | 75.03 | 2.809 | 22.0× | 66.60 |
| 2024-05-23 | 102.03 | 3.080 | 21.5× | 70.95 |
| 2024-08-29 | 121.36 | 3.722 | 24.0× | 96.97 |
| 2024-11-21 | 149.35 | 4.040 | 21.6× | 93.07 |
| 2025-02-27 | 135.00 | 4.939 | 22.2× | 118.95 |
| 2025-05-29 | 142.25 | 3.951 | 21.4× | 87.64 |
为什么100%不代表有效预测Why 100% does not establish predictive skill為什麼100%不代表有效預測
8期中6期在起点已高于估值；较低估值容易被后续价格达到。207.06%也会随估值分母降低而上升。期末MAPE反而高于不变价格基准，提示中心预测可能偏低。重叠窗口、小样本、事后选择模型和缺少完整PIT共识进一步限制解释。Six of eight origins already trade above the valuation; low targets are easier for subsequent prices to exceed. The 207.06% ratio also rises as the valuation denominator falls. Endpoint MAPE exceeding the unchanged-price baseline suggests potential downward bias in central forecasts. Overlapping windows, small sample size, hindsight model selection and missing complete point-in-time consensus further limit interpretation.8期中6期在起點已高於估值；較低估值容易被後續價格達到。207.06%也會隨估值分母降低而上升。期末MAPE反而高於不變價格基準，提示中心預測可能偏低。重疊窗口、小樣本、事後選擇模型和缺少完整PIT共識進一步限制解釋。
原始版本严格PIT合格样本N=0，与这里回顾性重建N=8分开标注。历史总体起止范围、价格源版本与最终观察截止日未在可读取摘要中完整提供；本站不将这份摘要命名为“过去十年回测”，不输出置信区间或V2胜率。Keep zero qualifying strict point-in-time samples in the original version separate from eight retrospective reconstructions. The complete sample range, price-source vintage and final observation cutoff are not fully available in the readable summary. This is not labeled a ten-year backtest; no confidence interval or V2 win rate is reported.原始版本嚴格PIT合格樣本N=0，與這裏回顧性重建N=8分開標註。歷史總體起止範圍、價格源版本與最終觀察截止日未在可讀取摘要中完整提供；本站不將這份摘要命名為“過去十年回測”，不輸出置信區間或V2勝率。
来源与发布边界Sources and publication boundary來源與發佈邊界
依据“英伟达估值研究”对话及2026-09-25原始研究稿整理，2026-09-26迁入本站。V1/V2属于该对话的研究版本，不等同于本站P11.1、N2.2或模型族v0.1；不将不同分支的参数和成绩混合。Compiled from the NVIDIA Valuation Research conversation and the original 2026-09-25 report; migrated to this site on 2026-09-26. These V1/V2 labels belong to that research conversation, not the site’s P11.1, N2.2 or family v0.1. Parameters and results from different branches are not blended.依據“英偉達估值研究”對話及2026-09-25原始研究稿整理，2026-09-26遷入本站。V1/V2屬於該對話的研究版本，不等同於本站P11.1、N2.2或模型族v0.1；不將不同分支的參數和成績混合。
模型结构是已有研究的摘要，不代表新的估值运行。本次只刊登已有V1历史检验摘要，不刊登V2单点诊断价格、前瞻目标价或尚未取得的十年回测。未对完整回测底稿重新运行，不声称独立复现。The structures summarize existing research, not a new valuation run. This release includes only the available V1 historical-evaluation summary; it excludes V2 single-origin diagnostic prices, forward targets and unavailable ten-year backtests. Complete backtest working files were not rerun; independent replication is not claimed.模型結構是已有研究的摘要，不代表新的估值運行。本次只刊登已有V1歷史檢驗摘要，不刊登V2單點診斷價格、前瞻目標價或尚未取得的十年回測。未對完整回測底稿重新運行，不聲稱獨立復現。
本次整理数据 SHA-256Compiled data SHA-256本次整理數據 SHA-256: 999064b7916f493d522c086041c7e05fe6ffdcb5accb8cf21d9eeb4eb0a2fb06
内容校验值不是历史公开时间证明或第三方审计认证。研究源稿与对话记录保留在本地；不提供失效的附件链接。The checksum is not proof of historical publication or third-party audit certification. Source reports and conversation provenance are retained locally; no expired attachment links are presented.內容校驗值不是歷史公開時間證明或第三方審計認證。研究源稿與對話記錄保留在本地；不提供失效的附件鏈接。