NVDA / HORIZON-07 / 2026-10-06
H07：价格趋势混合与盈利误差研究H07: price-trend blending and earnings-error researchH07：價格趨勢混合與盈利誤差研究
研究候选 · 未晋升Research candidate · unpromoted研究候選 · 未晉升新增校正没有提升No improvement from the new correction新增校正沒有提升
H07-EPS-MOMENTUM-BLEND／协议v0.1，未晋升研究候选。使用2026-10-06收盘$239.24；EPS基础资料沿用纽约22:36的H06来源，收盘价于23:36再核验。16:30是收盘后可得性重建约定。0M是价格参考，6M／12M是市场锚定条件价格预测。H07-EPS-MOMENTUM-BLEND / protocol v0.1, an unpromoted research candidate. Use the2026-10-06 close of$239.24; EPS inputs retain H06 sources captured at22:36 New York and the close was rechecked at23:36.16:30 is an after-close availability reconstruction.0M is a price reference;6M/12M are market-anchored conditional price forecasts.H07-EPS-MOMENTUM-BLEND／協議v0.1，未晉升研究候選。使用2026-10-06收盤$239.24；EPS基礎資料沿用紐約22:36的H06來源，收盤價於23:36再核驗。16:30是收盤後可得性重建約定。0M是價格參考，6M／12M是市場錨定條件價格預測。
| 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 | 计算输出 USD/股Calculated USD/share計算輸出 USD/股 |
|---|
| 0M · 现价锚Price anchor現價錨 | 2026-10-06 → 2027-10-06 | 13.8827 | 17.2330× | $239.24 |
| 6M | 2027-04-06 → 2028-04-06 | 16.6606 | 17.2330× | $287.11 |
| 12M | 2027-10-06 → 2028-10-06 | 18.9528 | 17.2330× | $326.61 |
本轮没有取得准确率提升：当前训练权重为0，所有可训练历史起点也为0，H07逐值等于市场P/E不变基准。价格趋势混合没有贡献增量信息，主门槛失败；0M不是独立公允价值。This round does not improve accuracy: the current trained weight is zero, as at every trained historical origin. H07 equals held-market-P/E exactly. The price-trend blend contributes no incremental information and fails primary gates;0M is not an independent fair-value estimate.本輪沒有取得準確率提升：當前訓練權重為0，所有可訓練歷史起點也為0，H07逐值等於市場P/E不變基準。價格趨勢混合沒有貢獻增量信息，主門檻失敗；0M不是獨立公允價值。
固定一个候选，检验已有价格信息Freeze one candidate and test observable price information固定一個候選，檢驗已有價格信息
EPS与前10季经营路径继续沿用H03／H05。新增过去12个月对数价格涨幅m，单一权重w在盈利增长基准和价格趋势外推之间混合。它是预测路径混合，不是独立的内在价值或纯P/E均值回复参数。Keep H03/H05 EPS and the first10 operating quarters. Add past12-month log price returnm, using one weightw to blend the earnings-growth baseline and price-trend extrapolation. This mixes prediction paths rather than estimating independent intrinsic value or pure P/E mean reversion.EPS與前10季經營路徑繼續沿用H03／H05。新增過去12個月對數價格漲幅m，單一權重w在盈利增長基準和價格趨勢外推之間混合。它是預測路徑混合，不是獨立的內在價值或純P/E均值回覆參數。
| 变量Variable變量 | 定义Definition定義 |
|---|
| Baseline | Bₕ = Eₕ × (P₀/E₀) |
| Past return | m = ln(P₀/P₋₁₂ₘ) |
| EPS growth | eₕ = ln(Eₕ/E₀) |
| Disagreement | dₕ = (h/12)m − eₕ |
| Forecast | Pₕ = Bₕ exp(wdₕ) |
| Implied P/E | PEₕ = Pₕ/Eₕ |
过去价格点按日历减12个月后，取该日或之前最近的交易收盘，最多回退7日；数据不覆盖则不补填。每个起点使用自己当时的EPS路径。训练只用已经到期的目标交易日价格，标签可得日设为交易日+1；事后实际EPS不参与拟合。Subtract12 calendar months, then take the latest trading close on or before that date, at most7 days earlier; missing coverage is not filled. Each origin uses its own EPS path. Fit only matured target closes, with availability dated target trading day+1; subsequently realized EPS is excluded from training.過去價格點按日曆減12個月後，取該日或之前最近的交易收盤，最多回退7日；數據不覆蓋則不補填。每個起點使用自己當時的EPS路徑。訓練只用已經到期的目標交易日價格，標籤可得日設為交易日+1；事後實際EPS不參與擬合。
每个过去起点的成熟期限合计权重1；至少12个不同起点，λ=4Sxx/N，先岭收缩再截到[0,1]，无截距。12、4、12个月窗口和边界在首次输出前固定。本轮没有根据成绩翻转符号或筛选其它窗口。Each past origin has total weight1 across mature horizons. Require12 distinct origins, set lambda=4Sxx/N, ridge-shrink then clip to[0,1], without an intercept.12,4, the12-month lookback and bounds were fixed before first output. This round did not flip signs or select alternative windows from scores.每個過去起點的成熟期限合計權重1；至少12個不同起點，λ=4Sxx/N，先嶺收縮再截到[0,1]，無截距。12、4、12個月窗口和邊界在首次輸出前固定。本輪沒有根據成績翻轉符號或篩選其它窗口。
零权重是拟合结果，不是数据不足Zero weight is a fitted result, not missing data零權重是擬合結果，不是數據不足
| 拟合字段Fit field擬合字段 | 结果Value結果 |
|---|
| Distinct origins | 34 |
| Mature rows | 66 |
| Sxx | 6.254126700 |
| Sxy | -1.211877389 |
| lambda | 0.735779612 |
| Unconstrained weight | -0.173375341 |
| Constrained weight | 0.000000 |
| Current log momentum | 0.254196740 |
| Lag target / trading date | 2025-10-06 / 2025-10-06 |
当前34个过去起点、66条成熟记录已经够资格拟合；未约束权重约−0.173，被预声明[0,1]约束归零。负数不在看到结果后改造成逆向模型。H07可训练历史样本21／19个全部权重0；另有17个warmup分别单列。The current34 prior origins and66 mature rows qualify for fitting. An unconstrained weight around−0.173 becomes zero under prespecified[0,1]bounds. It is not converted into a contrarian model after seeing results. All21/19 trained historical samples have zero weight;17 warmup origins are separately reported at each horizon.當前34個過去起點、66條成熟記錄已經夠資格擬合；未約束權重約−0.173，被預聲明[0,1]約束歸零。負數不在看到結果後改造成逆向模型。H07可訓練歷史樣本21／19個全部權重0；另有17個warmup分別單列。
主检验：与市场倍数基准完全相同Primary test: identical to held-market-P/E主檢驗：與市場倍數基準完全相同
| 期限Horizon期限 | 总到期Total mature總到期 | 可训练Trained可訓練 | WarmupWarmupWarmup | 可训练权重0Trained weight0可訓練權重0 |
|---|
| 6M | 38 | 21 | 17 | 21 |
| 12M | 36 | 19 | 17 | 19 |
| 期限Horizon期限 | 模型Model模型 | N | MALE | 中位APEMedian APE中位APE | P90 APE | 方向命中Direction方向命中 | 上涨／下跌数Up/down count上漲／下跌數 | 未来收益相关Forward-return correlation未來收益相關 |
|---|
| 6M | H07 | 21 | 0.3662 | 26.5% | 55.0% | 61.9% | 17/4 | -0.056 |
| 6M | MarketPEHeld | 21 | 0.3662 | 26.5% | 55.0% | 61.9% | 17/4 | -0.056 |
| 6M | PriceUnchanged | 21 | 0.3696 | 33.8% | 53.6% | 0.0% | 17/4 | — |
| 12M | H07 | 19 | 0.4508 | 19.4% | 71.2% | 63.2% | 16/3 | 0.124 |
| 12M | MarketPEHeld | 19 | 0.4508 | 19.4% | 71.2% | 63.2% | 16/3 | 0.124 |
| 12M | PriceUnchanged | 19 | 0.5916 | 39.7% | 70.4% | 0.0% | 16/3 | — |
主样本从2021-03-01开始，不能与H06的2022年起点样本直接比较。与市场P/E不变的MALE改善为0%，未达到预声明10%门槛；两期均不晋升。方向命中61.9%／63.2%，也不能据此宣称转折能力。MALE是平均绝对对数误差，越低越好。Primary origins begin2021-03-01 and cannot be directly compared with H06’s later sample. MALE improvement versus held-market-P/E is0%, failing the prespecified10%threshold; neither horizon is promoted. Direction scores of61.9%/63.2%also fail to establish turning-point skill. MALE is mean absolute log error, lower is better.主樣本從2021-03-01開始，不能與H06的2022年起點樣本直接比較。與市場P/E不變的MALE改善為0%，未達到預聲明10%門檻；兩期均不晉升。方向命中61.9%／63.2%，也不能據此宣稱轉折能力。MALE是平均絕對對數誤差，越低越好。
| 期限Horizon期限 | 基准Baseline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 | 开发门槛Development gate開發門檻 |
|---|
| 6M | MarketPEHeld | 0.00% | [0.00000, 0.00000] | false |
| 6M | PriceUnchanged | 0.93% | [-0.10759, 0.08434] | false |
| 12M | MarketPEHeld | 0.00% | [0.00000, 0.00000] | false |
| 12M | PriceUnchanged | 23.80% | [-0.42819, 0.08042] | true |
与H06只比较双方共同样本Compare H06 only on the shared trained intersection與H06只比較雙方共同樣本
| 期限Horizon期限 | 模型Model模型 | N | MALE | 中位APEMedian APE中位APE | P90 APE | 方向命中Direction方向命中 | 上涨／下跌数Up/down count上漲／下跌數 | 未来收益相关Forward-return correlation未來收益相關 |
|---|
| 6M | H07 | 17 | 0.3337 | 25.7% | 49.8% | 70.6% | 14/3 | -0.038 |
| 6M | MarketPEHeld | 17 | 0.3337 | 25.7% | 49.8% | 70.6% | 14/3 | -0.038 |
| 6M | PriceUnchanged | 17 | 0.3448 | 22.6% | 50.2% | 0.0% | 14/3 | — |
| 6M | H06 | 17 | 0.3361 | 25.7% | 50.2% | 64.7% | 14/3 | -0.053 |
| 12M | H07 | 15 | 0.4748 | 30.5% | 68.6% | 66.7% | 14/1 | 0.060 |
| 12M | MarketPEHeld | 15 | 0.4748 | 30.5% | 68.6% | 66.7% | 14/1 | 0.060 |
| 12M | PriceUnchanged | 15 | 0.6413 | 39.7% | 69.8% | 0.0% | 14/1 | — |
| 12M | H06 | 15 | 0.4803 | 30.5% | 70.0% | 60.0% | 14/1 | 0.040 |
双方都可训练的共同样本为17／15个。H07方向70.6%／66.7%，H06为64.7%／60.0%，但H07与简单基准逐值相同。这里的差异来自去掉H06一次非零校正失误，不能归功于新的动量信息。The intersection eligible for both models contains17/15 origins. H07 direction scores70.6%/66.7%versus H06’s64.7%/60.0%, while H07 equals the simple baseline exactly. The difference comes from removing one nonzero H06 correction error, not from new momentum information.雙方都可訓練的共同樣本為17／15個。H07方向70.6%／66.7%，H06為64.7%／60.0%，但H07與簡單基準逐值相同。這裏的差異來自去掉H06一次非零校正失誤，不能歸功於新的動量信息。
完整近10年流程，含warmupComplete nearly ten-year procedure, including warmup完整近10年流程，含warmup
| 期限Horizon期限 | 模型Model模型 | N | MALE | 中位APEMedian APE中位APE | P90 APE | 方向命中Direction方向命中 | 上涨／下跌数Up/down count上漲／下跌數 | 未来收益相关Forward-return correlation未來收益相關 |
|---|
| 6M | H07 | 38 | 0.3225 | 26.3% | 51.4% | 73.7% | 31/7 | 0.027 |
| 6M | MarketPEHeld | 38 | 0.3225 | 26.3% | 51.4% | 73.7% | 31/7 | 0.027 |
| 6M | PriceUnchanged | 38 | 0.3546 | 32.2% | 54.4% | 0.0% | 31/7 | — |
| 6M | H06 | 38 | 0.3236 | 26.3% | 51.9% | 71.1% | 31/7 | 0.018 |
| 12M | H07 | 36 | 0.4901 | 36.9% | 74.7% | 66.7% | 29/7 | 0.166 |
| 12M | MarketPEHeld | 36 | 0.4901 | 36.9% | 74.7% | 66.7% | 29/7 | 0.166 |
| 12M | PriceUnchanged | 36 | 0.6126 | 46.1% | 68.7% | 0.0% | 29/7 | — |
| 12M | H06 | 36 | 0.4923 | 36.9% | 75.7% | 63.9% | 29/7 | 0.154 |
全流程38／36个到期样本，H07方向73.7%／66.7%，一律看涨81.6%／80.6%。未来收益相关仅0.027／0.166，仍不能证明可靠预测。Across38/36 mature full-procedure samples, H07 direction scores73.7%/66.7%, versus always-up81.6%/80.6%. Forward-return correlations are only0.027/0.166, providing no evidence of reliable prediction.全流程38／36個到期樣本，H07方向73.7%／66.7%，一律看漲81.6%／80.6%。未來收益相關僅0.027／0.166，仍不能證明可靠預測。
EPS和P/E误差会相互抵消EPS and P/E errors can offset each otherEPS和P/E誤差會相互抵消
H06事后分解只用盈利窗口实际EPS已经公布的子样本：6M完整／有效训练33／12，12M31／10，各排除最新5个起点。EPS与事后隐含P/E共享分母，不能把组件当作独立因果贡献，也不能用实际EPS回填预测。H06 ex-post decomposition uses only already-published actual EPS windows: full/trained counts33/12 at6M and31/10 at12M, excluding the latest five origins at each horizon. EPS and ex-post implied P/E share a denominator, so components are not independent causal contributions and actual EPS must not fill predictions.H06事後分解只用盈利窗口實際EPS已經公佈的子樣本：6M完整／有效訓練33／12，12M31／10，各排除最新5個起點。EPS與事後隱含P/E共享分母，不能把組件當作獨立因果貢獻，也不能用實際EPS回填預測。
| 期限Horizon期限 | N | EPS平均有符号对数误差Mean signed EPS log errorEPS平均有符號對數誤差 | P/E平均有符号对数误差Mean signed P/E log errorP/E平均有符號對數誤差 | 合计价格误差Combined price error合計價格誤差 |
|---|
| 6M | 12 | -0.6867 | 0.5860 | -0.1007 |
| 12M | 10 | -1.0685 | 0.5464 | -0.5221 |
部分下跌漏判主要来自倍数没有及时下降；AI起飞阶段也出现过明显EPS低估。因此“EPS始终没问题，只优化P/E”不符合这个历史子样本的证据。本轮EPS不改是为了控制变量，不能理解为EPS已经验证准确。Some missed downturns mainly reflect multiples failing to fall, while AI takeoff also contains substantial EPS underforecasts. This historical subset does not support treating EPS as always correct and only adjusting P/E. Holding EPS fixed controls the experiment; it does not validate EPS accuracy.部分下跌漏判主要來自倍數沒有及時下降；AI起飛階段也出現過明顯EPS低估。因此“EPS始終沒問題，只優化P/E”不符合這個歷史子樣本的證據。本輪EPS不改是為了控制變量，不能理解為EPS已經驗證準確。
未采纳端点，不能替换主输出Unselected endpoints cannot replace headline outputs未採納端點，不能替換主輸出
| Weight | 0M | 6M | 12M |
|---|
| 0 | $239.24 | $287.11 | $326.61 |
| 1 | $239.24 | $271.66 | $308.48 |
权重1完全外推过去价格涨幅，未被当前训练采纳；端点未评分，不是预测区间。P/E是预测价格除以EPS的隐含值，通常不满足旧H06的PE6²=PE0×PE12。Weight1 fully extrapolates past price returns and is not adopted by the current fit. Endpoints are unscored, not prediction intervals. P/E is forecast price divided by EPS and generally does not obey H06’s old PE6²=PE0×PE12 identity.權重1完全外推過去價格漲幅，未被當前訓練採納；端點未評分，不是預測區間。P/E是預測價格除以EPS的隱含值，通常不滿足舊H06的PE6²=PE0×PE12。
实现检查与预测证据分开Separate implementation checks from predictive evidence實現檢查與預測證據分開
| 独立检查组Independent check group獨立檢查組 | 检查数Count檢查數 | 通过Passed通過 |
|---|
| 独立代数与时间重建Independent algebra and timing reconstruction獨立代數與時間重建 | 814 | true |
| 未来信息扰动与边界Future-information perturbations and boundaries未來信息擾動與邊界 | 75 | true |
旧H02至H06的8,861个文件在本轮计算前后hash一致。信息隔离与代数检查通过只证明遵循协议，不能证明准确率提升。此前历史已反复使用，期限重叠，未完成独立样本外验证。All8,861 protected H02–H06 files retain their hashes across the experiment. Timing and algebra checks establish protocol compliance, not improved accuracy. History has been repeatedly used, horizons overlap and independent OOS validation is incomplete.舊H02至H06的8,861個文件在本輪計算前後hash一致。信息隔離與代數檢查通過只證明遵循協議，不能證明準確率提升。此前歷史已反覆使用，期限重疊，未完成獨立樣本外驗證。
下一步应先取得预测发布日的盈利修订、订单／交付和资本开支证据，改进EPS误差识别，再用新增起点验证。停止在相同40个起点上反复改价格校正追逐成绩。Next obtain publication-dated earnings revisions, orders/deliveries and capital-expenditure evidence, improve EPS-error diagnosis, then validate on new origins. Repeatedly changing price corrections on the same40 origins cannot establish new predictive evidence.下一步應先取得預測發佈日的盈利修訂、訂單／交付和資本開支證據，改進EPS誤差識別，再用新增起點驗證。停止在相同40個起點上反覆改價格校正追逐成績。
公开仅派生预测和聚合结果，逐期价格、训练行和原始证据保留内部。研究网页已发布不代表正式预测模型部署。Publish derived forecasts and aggregates only; individual price records, training rows and raw evidence remain internal. Publishing a research page does not deploy a validated prediction model.公開僅派生預測和聚合結果，逐期價格、訓練行和原始證據保留內部。研究網頁已發佈不代表正式預測模型部署。
12个月趋势研究依据Twelve-month trend research source12個月趨勢研究依據
原研究来自分散期货市场；H07的NVDA单股EPS混合不是原论文复现，也不是已知有效策略。The cited work studies diversified futures. H07’s single-NVDA EPS blend is neither a replication nor an established effective strategy.原研究來自分散期貨市場；H07的NVDA單股EPS混合不是原論文復現，也不是已知有效策略。
Protocol SHA256 cc6fb57ea6a9b9bb8aaab91e9180230d41fdedbbb85a78e7ab6f0b78f8eb1b8d
Record SHA256 a59ac74b9740da2706b5bef6e05ec60d93604f54dbdd1ada6e448a3ab0712123
查看H07研究结果Read H07 research查看H07研究結果 · H06近10年趋势回测H06 nearly ten-year trend auditH06近10年趨勢回測 · 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