NVDA / H12 / ERROR ATTRIBUTION & GROWTH-STATE MULTIPLE
H12：H09误差归因与增长状态倍数实验H12: H09 error attribution and growth-state multiple experimentH12：H09誤差歸因與增長狀態倍數實驗
输入及市场收盘截面为2026-10-06。2026-10-07纽约10:27再次核验该日收盘$239.24；财报与指引沿用冻结资料，不能称全部输入刷新至10月7日。Input and market-close cutoff is 2026-10-06. The $239.24 close was rechecked on 2026-10-07 at 10:27 New York. Financials and guidance remain frozen; this is not a complete October 7 input refresh.輸入及市場收盤截面為2026-10-06。2026-10-07紐約10:27再次核驗該日收盤$239.24；財報與指引沿用凍結資料，不能稱全部輸入刷新至10月7日。
H12在12个历史起点和当前截面均拟合出β=0，输出精确回到H09，没有新增预测优势。共识比较相等而通过不恶化筛选，股价未改善，联合筛选未通过；继续保留H09研究参考。H12 fits beta=0 at all twelve historical origins and the current snapshot, returning exactly to H09 with no incremental predictive benefit. Equal consensus results pass nonworsening, but unchanged price errors fail the improvement requirement and joint screen. Retain H09 as the research reference.H12在12個歷史起點和當前截面均擬合出β=0，輸出精確回到H09，沒有新增預測優勢。共識比較相等而通過不惡化篩選，股價未改善，聯合篩選未通過；繼續保留H09研究參考。
| 模型Model模型 | 6M平均股价误差 · 10点6M mean price error · 10 points6M平均股價誤差 · 10點 | 12M平均股价误差 · 8点12M mean price error · 8 points12M平均股價誤差 · 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 |
|---|---|---|---|
| H09 | 14.33% | 14.91% | 16.86% |
| H12 | 14.33% | 14.91% | 16.86% |
当前H09参考与H12实验值Current H09 reference and H12 experimental values當前H09參考與H12實驗值
输入及市场收盘截面为2026-10-06。2026-10-07纽约10:27再次核验该日收盘$239.24；财报与指引沿用冻结资料，不能称全部输入刷新至10月7日。Input and market-close cutoff is 2026-10-06. The $239.24 close was rechecked on 2026-10-07 at 10:27 New York. Financials and guidance remain frozen; this is not a complete October 7 input refresh.輸入及市場收盤截面為2026-10-06。2026-10-07紐約10:27再次核驗該日收盤$239.24；財報與指引沿用凍結資料，不能稱全部輸入刷新至10月7日。
最终研究参考：H09 v0.1Final research reference: H09 v0.1最終研究參考：H09 v0.1
| 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 | 计算值 USD/股Calculated USD/share計算值 USD/股 |
|---|---|---|---|---|
| 0M | 2026-10-06 → 2027-10-06 | 13.86619 | 17.25348× | $239.24 |
| 6M | 2027-04-06 → 2028-04-06 | 16.62886 | 17.25348× | $286.91 |
| 12M | 2027-10-06 → 2028-10-06 | 18.90702 | 17.25348× | $326.21 |
固定4季度增长半衰期，市场PE不变，价格＝Core EPS×17.25348189。每个盈利窗口从目标日起覆盖未来12个月。0M=$239.24为市场价格锚，不是独立基本面公允价值。Use a fixed four-quarter growth half-life and held market P/E: price = Core EPS × 17.25348189. Each earnings window covers twelve months from the target date. The $239.24 0M value is a market-price anchor, not independent fundamental fair value.固定4季度增長半衰期，市場PE不變，價格＝Core EPS×17.25348189。每個盈利窗口從目標日起覆蓋未來12個月。0M=$239.24為市場價格錨，不是獨立基本面公允價值。
H12 v0.1：单独列示，未采用H12 v0.1: separate, not adoptedH12 v0.1：單獨列示，未採用
| 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 | 计算值 USD/股Calculated USD/share計算值 USD/股 |
|---|---|---|---|---|
| 0M | 2026-10-06 → 2027-10-06 | 13.86619 | 17.25348× | $239.24 |
| 6M | 2027-04-06 → 2028-04-06 | 16.62886 | 17.25348× | $286.91 |
| 12M | 2027-10-06 → 2028-10-06 | 18.90702 | 17.25348× | $326.21 |
| 期限Horizon期限 | 前12个月估计EPSEstimated preceding EPS前12個月估計EPS | G_h | D_h |
|---|---|---|---|
| 0M | 7.963151 | 0.554628 | 0.000000 |
| 6M | 10.939463 | 0.418763 | -0.135865 |
| 12M | 13.866187 | 0.310080 | -0.244548 |
H12使用相同EPS，价格差异只来自本分支的PE调整。当前β=0.000000；若有效β为0，输出精确等于H09，这是没有加入调整，不应称为新的准确率提升。H12 uses identical EPS; any price difference comes only from its branch-specific P/E adjustment. Current beta=0.000000. If effective beta is zero, outputs exactly equal H09: no adjustment is applied and this is not a new accuracy improvement.H12使用相同EPS，價格差異只來自本分支的PE調整。當前β=0.000000；若有效β為0，輸出精確等於H09，這是沒有加入調整，不應稱為新的準確率提升。
误差并非全部来自EPS或PEErrors do not come entirely from EPS or P/E誤差並非全部來自EPS或PE
H09以Pₕ=P₀×EPSₕ/EPS₀预测未来价格。统一提高所有期限的EPS会在比率中抵消，不改变预测。对整个盈利窗口已经公布的样本，可以将价格对数误差精确拆成两项：H09 predicts Pₕ=P₀×EPSₕ/EPS₀. Raising every EPS horizon by the same factor cancels in this ratio and leaves prices unchanged. Where complete earnings windows have been published, log-price error splits exactly into two terms:H09以Pₕ=P₀×EPSₕ/EPS₀預測未來價格。統一提高所有期限的EPS會在比率中抵消，不改變預測。對整個盈利窗口已經公佈的樣本，可以將價格對數誤差精確拆成兩項：
| 分量Component分量 | 定义Definition定義 |
|---|---|
| L | ln(actual price / H09 price) = A + B |
| A | ln[(actual EPS_h / actual EPS_0) / (forecast EPS_h / forecast EPS_0)] |
| B | ln[(actual price / actual EPS_h) / (P0 / actual EPS_0)] |
A是盈利增长比例预测误差，B是事后实际Core盈利隐含倍数的变化。B的分母包含后来才公布的盈利，绝不是当时已知的PIT远期PE，也不能解释为利率导致的变化。A is the forecast error in the earnings growth ratio. B is the ex-post change in the multiple on subsequently realized Core earnings. Its denominators contain earnings published later: B is not contemporaneously observable PIT forward P/E and is not an identified interest-rate effect.A是盈利增長比例預測誤差，B是事後實際Core盈利隱含倍數的變化。B的分母包含後來才公佈的盈利，絕不是當時已知的PIT遠期PE，也不能解釋為利率導致的變化。
| 期限Horizon期限 | 完整归因样本Complete attribution N完整歸因樣本 | 平均|A|Mean |A|平均|A| | 平均|B|Mean |B|平均|B| | 平均|L|Mean |L|平均|L| | A/B反向样本Opposite-sign A/BA/B反向樣本 |
|---|---|---|---|---|---|
| 6M | 5 | 0.139184 | 0.317399 | 0.178215 | 5/5 |
| 12M | 3 | 0.150990 | 0.372002 | 0.260520 | 2/3 |
表中是对数幅度，不是误差百分比或因果贡献占比。6M的5个样本全部反向抵消，12M为2/3；A与B共享实际EPS，不能将其当作独立原因。逐个剔除起点的诊断还显示，两期限平均B的符号都可能改变。这里的5/3点误差不可冒充完整10/8点价格回测。These are log magnitudes, not percentage errors or causal contribution shares. All five 6M observations have offsetting A/B signs, as do two of three at 12M. A and B share actual EPS and are not independent causes. Leave-one-origin-out diagnostics can reverse the mean B sign at both horizons. These 5/3-point errors must not be presented as the complete 10/8-point price backtest.表中是對數幅度，不是誤差百分比或因果貢獻佔比。6M的5個樣本全部反向抵消，12M為2/3；A與B共享實際EPS，不能將其當作獨立原因。逐個剔除起點的診斷還顯示，兩期限平均B的符號都可能改變。這裏的5/3點誤差不可冒充完整10/8點價格回測。
结论是值得检验一个小规模倍数假设，而不是已经证明“盈利准确、只需修PE”。这些实际EPS及归因分量全部留在事后诊断，不进入候选训练。The finding motivates a small multiple hypothesis; it does not establish that earnings are accurate and only P/E needs repair. Realized EPS and attribution components remain ex-post diagnostics and never enter candidate training.結論是值得檢驗一個小規模倍數假設，而不是已經證明“盈利準確、只需修PE”。這些實際EPS及歸因分量全部留在事後診斷，不進入候選訓練。
唯一改动：增长状态变化与未来倍数One change: growth-state changes and future multiples唯一改動：增長狀態變化與未來倍數
H12保持H09整条经营路径与0M、6M、12M EPS精确不变。先在同一预测日构建目标日前后各12个月的盈利窗口，再观察前后盈利比随期限如何变化。H12 preserves the entire H09 operating path and exact 0M, 6M and 12M EPS values. At one forecast origin, construct the twelve-month earnings windows before and after each target date, then measure how their ratio changes across horizons.H12保持H09整條經營路徑與0M、6M、12M EPS精確不變。先在同一預測日構建目標日前後各12個月的盈利窗口，再觀察前後盈利比隨期限如何變化。
| 变量Variable變量 | 定义Definition定義 |
|---|---|
| E_h | H09 Core EPS over [target, target+12 months) |
| T_h | As-of-origin estimated Core EPS over [target−12 months, target) |
| G_h / D_h | G_h = ln(E_h/T_h); D_h = G_h−G_0 |
| P/E | PE_h = (P0/E0) × exp(βD_h) |
| Price | P_h = E_h × PE_h; β=0 returns exact H09 |
Tₕ由当时已公布实际季度与该起点H09预测拼接，包含尚未公布的季度部分，不能称为全部已报告TTM。不得用后来实际盈利替换。现有10季度路径已覆盖全部窗口，无需额外延长；T₁₂与E₀是同一个窗口，必须相等。Tₕ combines actual quarters published by the origin with that origin’s H09 forecasts, including unreported intervals. It is not entirely reported TTM. Later realized earnings cannot replace it. The existing ten-quarter path covers all windows without extension; T₁₂ and E₀ are the same window and must be equal.Tₕ由當時已公佈實際季度與該起點H09預測拼接，包含尚未公佈的季度部分，不能稱為全部已報告TTM。不得用後來實際盈利替換。現有10季度路徑已覆蓋全部窗口，無需額外延長；T₁₂與E₀是同一個窗口，必須相等。
| 当前目标期限Current horizon當前目標期限 | 前12个月窗口：已知实际天数Preceding window: known-actual days前12個月窗口：已知實際天數 | 预测覆盖天数Forecast-covered days預測覆蓋天數 |
|---|---|---|
| 0M | 294 | 71 |
| 6M | 112 | 253 |
| 12M | 0 | 365 |
β≥0检验“增长状态改善时给予更高倍数、放缓时给予更低倍数”的假设。这是可被否定的预测关系，不是因果定律。PE₀仍来自当前市场锚，本轮没有额外加入国债或美联储利率折价。Nonnegative beta tests the hypothesis that improving growth states receive higher multiples and slowing states lower multiples. This is a falsifiable predictive relationship, not a causal law. PE₀ retains the current market anchor and no additional Treasury or Federal Reserve discount is added.β≥0檢驗“增長狀態改善時給予更高倍數、放緩時給予更低倍數”的假設。這是可被否定的預測關係，不是因果定律。PE₀仍來自當前市場錨，本輪沒有額外加入國債或美聯儲利率折價。
它与H10的区别在于：log(Pₕ/P₀)=(1+β)log(Eₕ/E₀)−βlog(Tₕ/T₀)，增加了前12个月估计盈利的变化项；仍需检查信号相关性，不能假设增加一个表达式就增加了独立信息。Unlike H10, log(Pₕ/P₀)=(1+β)log(Eₕ/E₀)−βlog(Tₕ/T₀), adding a change in estimated preceding-window earnings. Signal correlation must still be checked; an additional expression is not automatically independent information.它與H10的區別在於：log(Pₕ/P₀)=(1+β)log(Eₕ/E₀)−βlog(Tₕ/T₀)，增加了前12個月估計盈利的變化項；仍需檢查信號相關性，不能假設增加一個表達式就增加了獨立信息。
系数怎样学习，怎样回退How the coefficient is learned and falls back係數怎樣學習，怎樣回退
只使用严格早于预测日、6M与12M实际价格都已到期的原季度起点；每个起点总权重1，两期限各0.5。价格在实际交易日后一个日历日才视为可用。旧于三年的起点仅供当时训练，不进行十年评分。Use strictly earlier original quarterly origins whose 6M and 12M price outcomes have both matured. Each origin has total weight one, split equally across horizons. A close becomes available one calendar day after its trading date. Origins older than three years support as-of training only; there is no ten-year evaluation.只使用嚴格早於預測日、6M與12M實際價格都已到期的原季度起點；每個起點總權重1，兩期限各0.5。價格在實際交易日後一個日曆日才視為可用。舊於三年的起點僅供當時訓練，不進行十年評分。
| 项目Item項目 | 固定规则Fixed rule固定規則 |
|---|---|
| x / y | x=D_h; y=ln(actual target close / frozen H09 price_h) |
| Fit | β_raw=Sxy/(Sxx+λ); λ=4Sxx/N; β=clip(β_raw,0,1) |
| 最少训练起点Minimum training origins最少訓練起點 | 12 complete origin pairs |
| 先验与回退Prior and fallback先驗與回退 | β=0; exact H09 |
| 当前训练N / β原值 / 有效值Current N / raw beta / effective beta當前訓練N / β原值 / 有效值 | 36 / -0.481904 / 0.000000 |
| 最新训练起点／标签可用日Latest training origin / label availability最新訓練起點／標籤可用日 | 2025-08-28 / 2026-08-29 |
β向0收缩，0至1的范围和4个先验起点在评分前固定，不是最优参数证明。已拟合得到β=0是有效边界解，与样本不足回退不同。当前必要增长特征无效时整条分支回退H09，不能只挑某个期限替换。Beta shrinks toward zero. The [0,1] bounds and four prior-equivalent origins are fixed before scoring and are not proven optimal. A fitted beta of zero is an active boundary solution, distinct from insufficient-data fallback. Invalid required current growth features cause the entire branch to fall back to H09, without selecting individual horizons.β向0收縮，0至1的範圍和4個先驗起點在評分前固定，不是最優參數證明。已擬合得到β=0是有效邊界解，與樣本不足回退不同。當前必要增長特徵無效時整條分支回退H09，不能只挑某個期限替換。
当前36个完整起点、72行配对的未约束β为−0.481904，在预设非负范围内的最优值为0。这不证明增长越慢PE必然越高，也没有在看到结果后允许负系数另跑一个版本。市场提前计入增长变化、模型误差相关等都是可能解释，尚未被因果验证。The current 36 complete origins and 72 paired rows yield unconstrained beta=−0.481904; the optimum under the prespecified nonnegative range is zero. This does not establish that slower growth necessarily produces higher P/E, and no negative-coefficient alternative is run after seeing results. Prior market anticipation and correlated model errors are possible, not causally verified, explanations.當前36個完整起點、72行配對的未約束β為−0.481904，在預設非負範圍內的最優值為0。這不證明增長越慢PE必然越高，也沒有在看到結果後允許負係數另跑一個版本。市場提前計入增長變化、模型誤差相關等都是可能解釋，尚未被因果驗證。
拟合的价格残差还包含盈利预测误差及其他市场变化，β不能解释为纯PE敏感度。分析师目标价、未来实际EPS以及A/B归因分量均未用于拟合。The fitted price residual also contains earnings forecast errors and other market changes; beta is not pure P/E sensitivity. Analyst targets, future realized EPS and the A/B attribution terms are excluded from fitting.擬合的價格殘差還包含盈利預測誤差及其他市場變化，β不能解釋為純PE敏感度。分析師目標價、未來實際EPS以及A/B歸因分量均未用於擬合。
校准汇总记录Calibration summary校準彙總記錄
| 字段Field字段 | 值Value值 |
|---|---|
| activeOriginN | 12 |
| fallbackOriginN | 0 |
| betaZeroN | 12 |
| betaOneN | 0 |
| betaInteriorN | 0 |
| betaMin | 0 |
| betaMax | 0 |
| 当前拟合字段Current fit field當前擬合字段 | 值Value值 |
|---|---|
| N | 36.0000000 |
| rowN | 72.0000000 |
| Sxx | 2.7734792 |
| Sxy | -1.4850551 |
| penalty | 0.3081644 |
| betaRaw | -0.4819036 |
| beta | 0.0000000 |
| trained | true |
| fallbackReason | — |
| gradient | 2.9701102 |
| objective | 10.2302554 |
| priorBeta | 0.0000000 |
| priorEquivalentOrigins | 4.0000000 |
| boundary | zero |
| oldestOrigin | 2016-11-23 |
| latestOrigin | 2025-08-28 |
| latestLabelAvailability | 2026-08-29 |
| xMinimum | -0.9192270 |
| xMaximum | 0.4026543 |
| signalMean | -0.1503464 |
| EPSReturnMean | 0.1459176 |
| signalVariance | 0.0544370 |
| EPSReturnVariance | 0.0181239 |
| signalEPSReturnCovariance | -0.0141155 |
| signalEPSReturnCorrelation | -0.4493876 |
| excludedOriginN | 4.0000000 |
过去三年的股价预测检验Price forecast evaluation over the last three years過去三年的股價預測檢驗
| 模型Model模型 | 6M平均股价误差 · 10点6M mean price error · 10 points6M平均股價誤差 · 10點 | 12M平均股价误差 · 8点12M mean price error · 8 points12M平均股價誤差 · 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 |
|---|---|---|---|
| H09 | 14.33% | 14.91% | 16.86% |
| H12 | 14.33% | 14.91% | 16.86% |
仍按2023-10-06至2026-10-06的预测发起日筛选，保留原12个季度起点。6M到期10个、待到期2个；12M到期8个、待到期4个。当前截面单列。目标标签为目标日当日或之后首个交易日收盘，未公布的未来EPS不妨碍已到期价格评分。Filter by forecast origins from 2023-10-06 through 2026-10-06, retaining twelve original quarterly dates. There are ten mature and two pending 6M prices, and eight mature and four pending 12M prices. The current snapshot is separate. Targets use the first trading close on or after the target date; unpublished future EPS does not prevent scoring mature prices.仍按2023-10-06至2026-10-06的預測發起日篩選，保留原12個季度起點。6M到期10個、待到期2個；12M到期8個、待到期4個。當前截面單列。目標標籤為目標日當日或之後首個交易日收盤，未公佈的未來EPS不妨礙已到期價格評分。
| 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 | H12 | 10 | 14.33% | 0.147441 | 12.21% | 30.89% | -1.72% |
| 6M | H09 | 10 | 14.33% | 0.147441 | 12.21% | 30.89% | -1.72% |
| 6M | H07Reference | 10 | 20.83% | 0.226953 | 20.03% | 34.86% | -4.87% |
| 6M | PriceUnchanged | 10 | 21.14% | 0.253708 | 19.73% | 37.75% | -18.88% |
| 12M | H12 | 8 | 14.91% | 0.162365 | 10.46% | 26.31% | -3.65% |
| 12M | H09 | 8 | 14.91% | 0.162365 | 10.46% | 26.31% | -3.65% |
| 12M | H07Reference | 8 | 18.95% | 0.240083 | 7.67% | 41.13% | -16.85% |
| 12M | PriceUnchanged | 8 | 32.20% | 0.420237 | 32.34% | 47.51% | -32.20% |
MAPE为平均绝对百分比误差，MALE为平均绝对对数误差，均越低越好。不能将1−MAPE写成准确率。预设两期限均须MALE改善至少5%、MAPE下降、中位与P90恶化不超过10%；不在看到结果后更换门槛。MAPE is mean absolute percentage error and MALE mean absolute log error; lower is better. One minus MAPE is not an accuracy rate. At both horizons, the prespecified screen requires at least 5% MALE improvement, lower MAPE and no more than 10% median/P90 worsening. Thresholds do not change after seeing results.MAPE為平均絕對百分比誤差，MALE為平均絕對對數誤差，均越低越好。不能將1−MAPE寫成準確率。預設兩期限均須MALE改善至少5%、MAPE下降、中位與P90惡化不超過10%；不在看到結果後更換門檻。
| 期限Horizon期限 | 相对H09的MALE改善MALE improvement vs H09相對H09的MALE改善 | 95% paired CI | 方向命中Direction accuracy方向命中 | 一直预测上涨Always-up baseline一直預測上漲 |
|---|---|---|---|---|
| 6M | 0.00% | [0.00000, 0.00000] | 90.00% | 90.00% |
| 12M | 0.00% | [0.00000, 0.00000] | 100.00% | 100.00% |
配对区间[0,0]来自两模型预测完全相同，不表示未来没有预测风险。区间采用四起点移动区块、5000次抽样，只作探索性诊断；小样本、重叠期限和反复使用历史均限制其解释。EPS与H09逐值相同，成熟7／5／3个盈利窗口的误差也相同，本轮不声称盈利预测改善。The paired interval [0,0] follows from identical model forecasts, not an absence of future forecast risk. Intervals use four-origin moving blocks and 5,000 draws, for exploratory diagnosis only. Small samples, overlapping horizons and repeatedly reused history limit interpretation. EPS is identical to H09, so errors in the 7 / 5 / 3 mature earnings windows are also identical; this study claims no EPS improvement.配對區間[0,0]來自兩模型預測完全相同，不表示未來沒有預測風險。區間採用四起點移動區塊、5000次抽樣，只作探索性診斷；小樣本、重疊期限和反覆使用歷史均限制其解釋。EPS與H09逐值相同，成熟7／5／3個盈利窗口的誤差也相同，本輪不聲稱盈利預測改善。
分析师比较单独评价Evaluate analyst comparisons separately分析師比較單獨評價
12个同日配对：模型12M(t)对StockAnalysis All Analysts平均一年目标价(t)。它是预期差异，不是未来股价误差或真实公允价值；未用于参数训练。Twelve same-date pairs compare model 12M(t) with the StockAnalysis All Analysts average one-year target(t). This is a gap from expectations, not realized-price error or true fair value, and targets are not used for training.12個同日配對：模型12M(t)對StockAnalysis All Analysts平均一年目標價(t)。它是預期差異，不是未來股價誤差或真實公允價值；未用於參數訓練。
| 模型Model模型 | N | 平均绝对差异Mean absolute gap平均絕對差異 | MALE | 中位差异Median gap中位差異 | 平均带符号差异Mean signed gap平均帶符號差異 | ±10% | ±20% |
|---|---|---|---|---|---|---|---|
| H12 | 12 | 16.86% | 0.146112 | 8.81% | 13.24% | 7/12 | 9/12 |
| H09 | 12 | 16.86% | 0.146112 | 8.81% | 13.24% | 7/12 | 9/12 |
| H07Reference | 12 | 22.64% | 0.222992 | 20.79% | 9.76% | 3/12 | 5/12 |
共识筛选要求MALE、平均绝对百分比差异、绝对平均带符号差异都不恶化。沿用10月7日图表悬停记录，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，Pro原始数据保留内部。The consensus screen requires nonworsening MALE, mean absolute percentage gap and absolute mean signed gap. October 7 chart-tooltip records are reused; original historical release times, revisions and split lineage remain unverified. Publish only aggregates and keep raw Pro data internal.共識篩選要求MALE、平均絕對百分比差異、絕對平均帶符號差異都不惡化。沿用10月7日圖表懸停記錄，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，Pro原始數據保留內部。
StockAnalysis平均目标价图表StockAnalysis average-target chartStockAnalysis平均目標價圖表
研究结论与下一步Research conclusion and next step研究結論與下一步
H12在12个历史起点和当前截面均拟合出β=0，输出精确回到H09，没有新增预测优势。共识比较相等而通过不恶化筛选，股价未改善，联合筛选未通过；继续保留H09研究参考。H12 fits beta=0 at all twelve historical origins and the current snapshot, returning exactly to H09 with no incremental predictive benefit. Equal consensus results pass nonworsening, but unchanged price errors fail the improvement requirement and joint screen. Retain H09 as the research reference.H12在12個歷史起點和當前截面均擬合出β=0，輸出精確回到H09，沒有新增預測優勢。共識比較相等而通過不惡化篩選，股價未改善，聯合篩選未通過；繼續保留H09研究參考。
归因样本提示盈利误差与倍数变化可能抵消；更改其中一项不保证总误差下降。当前0M是否低估，需要另立现金流或合理PE方案，不能用现价恒等式来验证。本轮没有生成独立0M估值。Attribution suggests earnings errors and multiple changes can offset; changing one component need not reduce total error. Whether today’s 0M price is undervalued requires a separate cash-flow or justified-P/E protocol, not validation of a market-price identity. No independent 0M value is generated here.歸因樣本提示盈利誤差與倍數變化可能抵消；更改其中一項不保證總誤差下降。當前0M是否低估，需要另立現金流或合理PE方案，不能用現價恆等式來驗證。本輪沒有生成獨立0M估值。
独立计算检查1478项、信息隔离检查502项通过，审计未读取候选评分表。9014个旧研究文件保持原样；实现正确不代表预测优势成立。Independent calculation checks (1478) and information-isolation checks (502) pass without reading candidate score tables. All 9014 older research files remain unchanged. Correct implementation does not establish predictive advantage.獨立計算檢查1478項、信息隔離檢查502項通過，審計未讀取候選評分表。9014箇舊研究文件保持原樣；實現正確不代表預測優勢成立。
按本次捕获时间保存完整预测和版本，不将重复发布的H09旧截面计作新的独立预测。下一步积累真正向前的冻结记录；历史三年已被多次用于开发，原始不可变PIT资料仍不完整，独立样本外验证尚未完成。本页发布研究，不部署H12为正式模型。Save complete predictions and versions at this capture time; reissuing H09’s old cutoff does not count as a new independent forecast. Accumulate genuinely prospective freezes next. The historical three-year window has been repeatedly used for development, immutable original PIT data is incomplete and independent OOS validation remains incomplete. This page publishes research and does not deploy H12 as a formal model.按本次捕獲時間保存完整預測和版本，不將重複發佈的H09舊截面計作新的獨立預測。下一步積累真正向前的凍結記錄；歷史三年已被多次用於開發，原始不可變PIT資料仍不完整，獨立樣本外驗證尚未完成。本頁發佈研究，不部署H12為正式模型。
按预测期限进行滚动验证Rolling validation by forecast horizon按預測期限進行滾動驗證
Frozen UTC 2026-10-07T14:33:05.277559+00:00 · Protocol SHA256 78b10d3159b98d6ec3671988019c841dd35bae39b5a8bdb5734a6c683e75aeb7
Record SHA256 31a6d89de4791997ea2a810af58260d26fcf00f971ee71da51d98f57e48c9266
H11持续率实验H11 persistence experimentH11持續率實驗 · H07–H10三年比较H07–H10 three-year comparisonH07–H10三年比較 · H10档案H10 archiveH10檔案 · H09档案H09 archiveH09檔案 · 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