一、这个数是什么
三档赔率（主 / 平 / 客）的隐含概率之和，就是 overround：
overround = 1/主 + 1/平 + 1/客 返奖率 = 1 / overround
赔率如果是公平的，overround 正好等于 1.0。真实盘口永远大于 1，多出来的部分就是抽水 —— 在你对比赛的看法正确与否之前就已经收走的费用。
二、数据与方法
| 项 | 说明 |
|---|---|
| 样本 | 897 场，2026-05-17 → 2026-09-25（一个赛季片段） |
| 被测 | 中国竞彩胜平负固定赔率，以及同一批比赛的让球盘 |
| 对照 | 欧赔「99 家平均终指」—— 同一批比赛的市场共识价 |
| 方法 | 逐场分别算 overround，再逐场配对做差 |
原始赔率随报告附带（data/overround_897.csv，每场一行，只有赔率数字），
本页所有统计量都由它现算。
三、结果
3.1 它是公式定价，不是市场定价
看 σ。897 场里，竞彩的 overround 只在 1.1251 – 1.1299 之间摆动， 幅度不到半个百分点；欧赔的摆动是它的 19 倍，随联赛、流动性、开赛时间上下浮动。
| 市场 | 场次 | overround | 抽水 | 返奖率 | σ | 区间 |
|---|---|---|---|---|---|---|
| 竞彩 胜平负 | 897 | 1.1292 | 12.92% | 88.6% | 0.0006 | 1.1251 – 1.1299 |
| 竞彩 让球盘 | 897 | 1.1291 | 12.91% | 88.6% | 0.0006 | 1.1251 – 1.1299 |
| 欧赔 99 家平均 | 897 | 1.0685 | 6.85% | 93.6% | 0.0109 | 1.0403 – 1.1085 |
- 市场定价（欧赔）：由无数参与者博弈出来，所以会动。会动的价格才可能出错。
- 公式定价（竞彩）：由固定返奖参数算出来，不动，因此不存在错价可找。
同一个彩票的两种玩法（胜平负 / 让球）落在 12.92% 和 12.91% —— 同一个参数套所有玩法，这正是公式定价的样子。
3.2 它每一场都更贵
897 / 897 场，平均贵 6.07 个百分点（中位 5.97）。 下面每一个点都是一场比赛，没有例外：
推论：一个能在欧洲市场盈利的模型，搬到这里必然亏损。 对阵欧赔平均线要 ~6.85% 的优势才能打平，在这里要 12.92%。 职业博彩团队花几百万美元找的是 2~4% 的优势 —— 这里的门槛是他们的 3~6 倍。
3.3 串关把抽水连乘
| 串数 | 每 100 元长期回收 | 抽掉 |
|---|---|---|
| 单关 | 88.6 | 11.4 |
| 2 串 1 | 78.4 | 21.6 |
| 4 串 1 | 61.5 | 38.5 |
| 6 串 1 | 48.2 | 51.8 |
| 8 串 1 | 37.8 | 62.2 |
6 串 1 抽掉一半以上。作为尺度参照：美式轮盘（双零）抽 5.26%，6 串 1 是它的近十倍。
四、我也拿自己检验了一遍
在扩到 1617 注的样本上跑了 15 个策略。精确的结论（不是"全部亏损"）：
- 9 / 15 个显著为负（95%CI 上界 < 0），另有 3 个点估计为负但不显著。
- 3 / 15 个为正，但三个的 95%CI 全部跨过 0（最大 ±116%）， 其中最大的那个头两个月 +143% / +43%，最后三个月每月 −100%。
- 0 个通过我在看数据之前就写死的门槛：注数 ≥300 且 CI 下界 > 0。
所以诚实说法不是"所有策略都亏"，而是"没有一个能过样本量这关，正收益那几个和噪声分不开"。
五、局限
- 单一市场、单一时段：一个赛季片段，不是全年。
- 对照是 99 家平均终指，不是某一家庄家的实时盘口。
- overround 量的是入场成本，不等于某个人的实际损失率 —— 后者取决于你怎么押、押多少。
- σ 的比较证据充分（overround 是逐场常数）；更细的分层（按联赛、按月）证据不足，本文不作结论。
- 上面那 3 个正收益策略按噪声处理，不作为发现。
六、声明
本文是数据分析，不是投注建议。我不推荐比赛、不卖料、不承诺任何收益。
我量出这个数字，恰恰是为了说明反面：在这个市场里任何策略都要先跨过 12.92% 这道墙。 跨不过去，就是给别人付过路费。
1. The measure
Given three-way odds (home / draw / away), the overround is the sum of the implied probabilities:
overround = 1/home + 1/draw + 1/away payout = 1 / overround
If the odds were fair, overround would be exactly 1.0. Real books are always above 1.0, and the excess is the margin — the fee you pay before your opinion about the match is even relevant.
2. Data and method
| item | detail |
|---|---|
| sample | 897 matches, 2026-05-17 → 2026-09-25 (one season fragment) |
| under test | Chinese Sports Lottery 1X2 fixed odds, and the handicap market on the same matches |
| benchmark | European “99 bookmakers average” closing odds — the market consensus for the same fixture |
| method | overround per match per market, then compared pairwise |
Raw odds ship with the report (data/overround_897.csv, one row per match,
odds only). Every statistic on this page is computed from that file.
3. Results
3.1 It is a formula, not a market
Look at σ. Across 897 matches the lottery's overround stays inside 1.1251 – 1.1299 — a swing of under half a percentage point. The European average swings 19× more, following league, liquidity and kick-off time.
| market | n | overround | margin | payout | σ | range |
|---|---|---|---|---|---|---|
| JC 1X2 (fixed odds) | 897 | 1.1292 | 12.92% | 88.6% | 0.0006 | 1.1251 – 1.1299 |
| JC handicap | 897 | 1.1291 | 12.91% | 88.6% | 0.0006 | 1.1251 – 1.1299 |
| EU avg (99 books) | 897 | 1.0685 | 6.85% | 93.6% | 0.0109 | 1.0403 – 1.1085 |
- Market pricing (EU): produced by the interaction of many participants, so it moves. A moving price can be wrong.
- Formula pricing (JC): produced by a fixed payout parameter. It does not move, and therefore there is no mispricing to find.
Two independent play types in the same lottery (1X2 and handicap) land on 12.92% and 12.91% — the same parameter applied to everything, which is what formula pricing looks like.
3.2 It is more expensive in every single match
897 of 897, mean gap 6.07 percentage points (median 5.97). Each dot below is one match. No exceptions:
Consequence: a model that is profitable at European margins is necessarily unprofitable here. You need ~6.85% edge to break even against the average book; you need 12.92% here. Professional syndicates spend millions building models to find 2–4%. The bar here is 3–6× that.
3.3 Multi-leg bets compound the margin
| legs | returned per 100 staked | taken |
|---|---|---|
| 1 | 88.6 | 11.4 |
| 2 | 78.4 | 21.6 |
| 4 | 61.5 | 38.5 |
| 6 | 48.2 | 51.8 |
| 8 | 37.8 | 62.2 |
A 6-leg accumulator gives up more than half. For scale: American roulette (double zero) takes 5.26%; a 6-leg ticket here takes about 10× that.
4. I also tested this on myself
I ran 15 betting rules over an extended 1,617-bet sample. Result, stated precisely (not “every strategy loses”):
- 9 of 15 are significantly negative (95% CI upper bound below 0); another 3 have a negative point estimate but are not significant.
- 3 of 15 are positive, and all three have 95% CIs that cross zero (widest ±116%). The largest of them makes +143% and +43% in its first two months and −100% in each of the last three.
- 0 pass the rule I fixed before looking at the data: n ≥ 300 and CI lower bound > 0.
So the honest summary is not “every strategy loses” — it is “nothing survives the sample-size bar, and the positive results are indistinguishable from noise”.
5. Limitations
- One market, one period: a season fragment, not a full year.
- The benchmark is the 99-bookmaker average close, not one book's live price.
- Overround measures the price of entry, not any individual's loss rate — that depends on what and how much you actually stake.
- The σ comparison is well supported (overround is a per-match constant); the finer breakdowns (by league, by month) are not, and are not claimed.
- The 3 positive rules are reported as noise, not findings.
6. Disclaimer
This is data analysis, not betting advice. I do not recommend matches, sell selections, or promise any return.
I measured this number precisely to make the opposite point: in this market every strategy must first clear a 12.92% wall. If it doesn't, you are paying someone else's toll.