香港賽馬量化研究報告Hong Kong Racing Quantitative Research香港赛马量化研究报告
獨立研究報告，延伸短文見 @sigma_quant（Instagram／Threads）。Standalone research research PDFs in Traditional Chinese and English, shorter notes continue at @sigma_quant on Instagram and Threads.独立研究报告，延伸短文见 @sigma_quant（Instagram／Threads）。
PDF REPORTS
研究報告下載Download research PDFs研究报告下载
冷門偏誤分析：獨贏與位置賠率、命中率及回報Longshot Bias Analyses: Closing Win and Place Odds versus Realised Outcomes冷门偏误分析：独赢与位置赔率、命中率及回报
冷門馬的較高派彩能否補償較低的勝出及入位機會？3,485場、42,463馬次中，高賠率獨贏／位置組估算回報約 −44%／−41%，低賠率組約 −14%／−16%。冷門平均更差，但20個獨贏組及20個位置組全部為負：關鍵是價格相對機率，而非派彩看似多大。Do larger payouts compensate for how rarely longshots win or place? In 42,463 Hong Kong horse-runs, high-odds Win/Place groups returned about −44% / −41%, versus about −14% / −16% for low-odds groups. Longshots were worse value on average, but every bin still lost: relative price, not payout size.冷门马的较高派彩能否补偿较低的胜出及入位机会？3,485场、42,463马次中，高赔率独赢／位置组估算回报约 −44%／−41%，低赔率组约 −14%／−16%。冷门平均更差，但20个独赢组及20个位置组全部为负：关键是价格相对机率，而非派彩看似多大。
資金流與「跟大戶」：為何高命中率不等於正期望值Money Flow and "Following Smart Money": Why a High Hit Rate Is Not Positive Expected Value资金流与「跟大户」：为何高命中率不等于正期望值
檢驗「跟大戶」能否長期獲利：資金最重的一匹勝率約 25.5%，但平均回報約 −22.6%，低於全場平均。數學證明跟注期望不超過零；資金流應與自有機率模型按精度融合，不宜單靠資金落注。Tests whether following heavy money can profit long term: the top money-share horse wins ~25.5% of bets yet returns ~−22.6% on average, worse than the field. Math shows followers cannot beat zero EV; blend money flow with your own probability model—do not bet on flow alone.检验「跟大户」能否长期获利：资金最重的一匹胜率约 25.5%，但平均回报约 −22.6%，低于全场平均。数学证明跟注期望不超过零；资金流应与自有概率模型按精度融合，不宜单靠资金下注。
連贏／位置Q研究：確認層、殘差公式與組合建構QIN / QPL Research: Confirmation Layer, Residual Formula and Ticket Design連贏／位置Q研究：確認層、殘差公式與組合建構
Q／QP 多數重述 Win／Place 共識。組合排名有助縮窄票面；原始落飛看似很強，但殘差與封鎖驗證後獨立增量減弱。定位為確認與組合建構，而非第二套勝率模型。Q/QP mostly restates the Win/Place consensus. Pair rank helps narrow tickets; raw steam looks strong but fades after residual and blocked tests. Treat it as confirmation and ticket design—not a second horse model.Q／QP 多數重述 Win／Place 共識。組合排名有助縮窄票面；原始落飛看似很強，但殘差與封鎖驗證後獨立增量減弱。定位為確認與組合建構，而非第二套勝率模型。
市場價格水平仍是主訊號；賠率路徑僅帶來小幅增量。溫和且獲尾段確認的支持較可信，持續大幅飄冷是最清楚的風險否決。適合作信心疊加與風險濾鏡，尚未證明投注優勢。Price level remains the main signal; odds paths add only a modest lift. Mild, late-confirmed support is the most credible upgrade, while sustained strong drift is the clearest risk veto. Use as a confidence overlay—not a proven betting edge.市場價格水平仍是主訊號；賠率路徑僅帶來小幅增量。溫和且獲尾段確認的支持較可信，持續大幅飄冷是最清楚的風險否決。適合作信心疊加與風險濾鏡，尚未證明投注優勢。
騎練組合三年分析：穩定偏差與樣本風險Jockey–Trainer Combinations Across Three Seasons: Stable Deviations and Sample Risk騎練組合三年分析：穩定偏差與樣本風險
比較三季高量騎練組合相對馬房基準的勝負／位置偏差。跨季重複的正偏差較耐看；小樣本會放大絕對偏差。歷史配合表現不等于賠率價值。Compares high-volume pairs across three seasons on win/place deviations versus stable baselines. Cross-season positive repeats travel better; small samples inflate absolute gaps. Historical pairing strength is not odds value.比較三季高量騎練組合相對馬房基準的勝負／位置偏差。跨季重複的正偏差較耐看；小樣本會放大絕對偏差。歷史配合表現不等于賠率價值。
以賽前平均賠率變動描述騎師、練馬師與組合的市場注意力。部分名字持續獲提前支持，極端組合波幅常伴小樣本。適合讀盤上下文，不宜自動跟注。Average pre-race odds moves map market attention across jockeys, trainers and pairs. Some names keep drawing early support; extreme pair swings often sit on thin samples. Use as reading context, not an auto-bet list.以賽前平均賠率變動描述騎師、練馬師與組合的市場注意力。部分名字持續獲提前支持，極端組合波幅常伴小樣本。適合讀盤上下文，不宜自動跟注。
研究內容只作數據分析與教育用途，不構成投注建議，亦不保證任何回報。賽馬活動涉及風險。Research is for analysis and education only. It is not betting advice and does not guarantee returns. Racing involves risk.研究内容只作数据分析与教育用途，不构成投注建议，亦不保证任何回报。赛马活动涉及风险。
Member login
Membership signup
Membership unlock
Upgrade toPro+to unlock
Stay on this page, or open membership plans to continue.
