目录
核验图例:本篇每条承重引用后标注核验状态——✓ 主笔本人联网逐字复核;◐ 调研一手全文逐字提取、多源交叉印证、未经主笔逐字复核;⚠ 证据层级偏弱(二手转述/摘要级/未同行评审/付费墙未取正文);○ 仅题名。凡未核处一律显式标注,不以强度冒充。
课题完成态:先把”市场比谁都准”拆开
本篇不把”预测市场准不准”当成一个可一次验收的命题,而追问一个更窄、更可判决的问题:
一个”流动的预测市场能把分散信息聚合成常常校准良好的价格”这个有条件的经验事实,被讲成”市场价格=事件真概率”和”预测市场是永远打败专家、民调与模型的终极预报神谕”,各自越了多远?截至 2026-07-20:市场价格在本体上等于事件的真实概率吗?预测市场在预报准确性上真的碾压民调、模型与超预报者吗?薄市场与大户能不能扭曲价格?把”给信念定价”的工具外推成”把治理交给下注”(futarchy),成立吗?
结构胎记(一句压多物):把”流动的预测市场能把分散信息聚合成一个常常校准良好的价格”这个有条件的经验事实——① 冒充”市场价格=事件的真实概率“(校准→本体等同)× ② 冒充”预测市场永远打败专家、民调与模型,是群体智慧的终极化身“(局部平手/略胜→普适碾压)× ③ 用一次命中或一次翻车(IEM 2008/Brexit-Trump 2016/Polymarket 2024)冒充对整套方法的裁决 × ④ 把”给信念定价“的工具外推成”把治理交给下注”(futarchy)的制度方案。
全文把六层命题分账:
- 真锚层——信息聚合是真现象、理论根扎实、活跃市场常常打平或略胜民调(反祛魅不等于否认这些,更不等于把预测市场打成”赌博骗局”);
- 概念/术语层——”预测市场”不是单一物(real-money vs play-money、连续双向拍卖 CDA vs 做市商 LMSR、二元 vs 标量),”群体智慧””价格””概率””校准””准确”都必须消歧;
- 论证/机制层——市场为何能预测(边际交易者、真话激励、套利),及其边界(Grossman-Stiglitz 悖论);
- 命门一·价格≠真概率——市场价格是被风险偏好折射的、信念分布的一个分位数,而非事件真概率;系统性 favorite-longshot 偏差;
- 命门二·流动性与操纵——薄市场可被大户扭曲、wash trading、制度脆弱;但操纵往往被迅速抹平(对称,防虚无化);
- 叙事/对称层——升格侧(”市场知道一切””碾压专家””交给它治理”)与虚无侧(”就是赌博””被鲸鱼操纵””2016 就错了”)对称落刀,含极端事件失灵与聚合算法/超预报者对决。
研究任务清单
- [x] 重读项目协作规则(
AGENTS.md)、调研流程、当前状态、报告模板;确认署名 Claude Opus 4.8、采用格式 v5。 - [x] 完善课题:把”预测市场准不准”改写为带前提、带可反驳条件的六层承重问题,写定结构胎记与对称三向红线。
- [x] 查重独占边界:与市场有效性 EMH 篇、黑天鹅/肥尾篇、测量代理性/Goodhart 篇分工——本篇独占”预测市场作为聚合器与神谕之间的承重审计”,不重打 EMH 的半强式检验。
- [x] 六捆并行一手取证:真锚+理论根 / 校准实证 vs 民调模型 / 价格≠真概率 / 操纵流动性薄市场 / 聚合对决+极端事件 / 监管治理+叙事对称。
- [x] 核真锚与理论根:Hayek 1945(分散知识/价格即电信系统)、Wolfers-Zitzewitz 2004 JEP(定义/三功能/IEM 1.5 vs Gallup 2.1)、Arrow et al. 2008 Science(22 人联署)、IEM Berg 等、Hanson LMSR 逐字。
- [x] 核校准与概念纪律:Page-Clemen 2013(S 形 longshot)、Gneiting 2007(calibration vs sharpness)、Brier 的 Murphy 分解、Manifold/Metaculus 平台校准。
- [x] 核命门一:Manski 2006(价格非”市场概率”)、Wolfers-Zitzewitz 2006(何时≈/何时偏)、Thaler-Ziemba 1988、Snowberg-Wolfers 2010(favorite-longshot 量级)逐字。
- [x] 核命门二:Hanson-Oprea-Porter(操纵不损精度)、Rhode-Strumpf(44 点冲击被抹平)、Tetlock 2008(流动性不改善校准)、Théo whale 双向、wash trading、Intrade/PredictIt/Kalshi 判决。
- [x] 核命门三与极端事件:Dana 2019(聚合显著胜市价)、Atanasov 2017(approximately tied)、Mellers 2014(超预报者 Brier 0.07)、Brexit/Trump 2016 隐含概率与概率辩护、futarchy/PAM。
- [x] 主笔亲核最承重引用:Manski”refutes market probabilities”、Dana 2019 Brier 0.227/0.210 与 p=0.004、IEM”about as well as the average poll”、Hanson-Oprea-Porter”unable to distort price accuracy”、Kalshi v. CFTC 判决书——全部本人 curl+pdftotext/HTML 逐字确认无漂移。
- [x] 对称双向红队:攻升格(校准当真概率、局部平手当碾压、机制当永远有效、价格当预言)与攻虚无化(一次翻车当方法死、一个鲸鱼当市场必假、赌博性质当无认识价值、监管争议当道德败坏)。
简短结论
母裁决:信息聚合是真的(Hayek)、活跃厚市场校准常常良好且选前误差约为民调一半(IEM 市场 1.49% vs 民调 1.91%、Snowberg-Wolfers-Zitzewitz “half the forecast error”)、real-money 逼真话是真机制(Wolfers-Zitzewitz 三功能)、预测市场是 22 位顶尖经济学家联署背书的正规工具(Arrow et al., Science)——真的不是”市场价格=事件真概率”和”预测市场是永远打败一切的终极预报神谕”这两层被声称的胜利。截至 2026-07-20:(a)价格在本体上不是真概率,而是被风险偏好折射的、信念分布的一个分位数,Manski 一手判定”refutes the notion that prices in prediction markets are ‘market probabilities'”,系统性 favorite-longshot 偏差把价格压向 0.5、远期与政治市场更甚(Page-Clemen);(b)薄市场里单笔大单可在三分钟内把价格打飞 44 点(Rhode-Strumpf),wash trading 占成交量的四分之一到三分之一(哥大研究/Chaos Labs),但操纵往往被迅速抹平且不获利(Hanson-Oprea-Porter”unable to distort price accuracy”);(c)当自报信念被恰当聚合,它在统计上显著比市价更准(Dana 2019,p=0.004),精英超预报者的 Brier 打到 0.07、甚至胜过有机密权限的情报分析员——市场机制本身不是决定性优势。
术语层就已经埋着混淆:“预测市场”不是单一物——real-money(IEM、Kalshi、Polymarket)vs play-money、连续双向拍卖 vs Hanson 的对数市场评分规则做市商(LMSR,为薄市场补贴流动性)、二元合约 vs 标量合约,各自的准确性与可操纵性都不同。而”群体智慧“(Surowiecki 2004 的四条件:多样性、独立性、去中心化、聚合 ○)也不等于预测市场——市场价格恰恰由少数边际交易者驱动(IEM Berg:”Marginal traders, not average traders, drive market prices” ✓),交易者相互观察挂单,常常违反 Surowiecki 要求的”独立性”。把预测市场直接等同于”群体智慧”,本身就是一次术语升格。
命门一:价格不是真概率(本体最硬一击)。Manski 2006(主笔 pdftotext 亲核)证明:在风险中性、异质信念的交易者下,”The above analysis refutes the notion that prices in prediction markets are ‘market probabilities.’ … the price of an all-or-nothing futures contract does not equal the mean, median, or any other measure of the central tendency of traders’ beliefs“(✓)——价格只界定均值信念落在一个区间内。Wolfers-Zitzewitz 2006 补上对称边界:在对数效用等假设下价格≈均值信念,但风险中性投资者对应的是信念分布的第 100−π 百分位、而非均值(◐)。系统性 favorite-longshot 偏差是这一层的实证签名(Page-Clemen 2013 ◐)。
命门二:薄市场可被扭曲,但操纵常被抹平(对称)。升格侧要防”市场不可操纵”;虚无侧要防”市场必被鲸鱼操纵”。两边都有一手证据:Rhode-Strumpf 记录 2004-10-15 单一投资者三分钟把 Bush 合约打飞 44 点,但”these changes were quickly undone“、且攻击者亏钱(◐);Hanson-Oprea-Porter(主笔 pdftotext 亲核)实验里”manipulators are unable to distort price accuracy“、非操纵者反向抵消(✓);而 2024 Polymarket 的”Théo“四账户押 2860 万美元、净利约 7900 万美元,平台调查后判定其为”taking directional positions”而非操纵(◐),wash trading 则占相当比例。
命门三:市场没有碾压聚合算法与超预报者。Dana et al. 2019(主笔 HTML 亲核,同 GJP 数据、开放获取)给出最干净的数字:市价 Brier 0.227,而恰当聚合的自报信念(加权+去偏 extremizing)0.210,二者组合”significantly more accurate than prediction-market prices alone“(配对 t(113)=2.92,p=0.004 ✓),因为”self-reports contained information that the market did not efficiently aggregate“(✓);Mellers 2014 里超预报者临近结算的 Brier 打到 0.07(未训练散户 0.26 ◐)、精英团队甚至”more accurate than professional intelligence analysts with access to classified information”(◐)。但这一切都不能反过来否定市场——Dana 同文亦写”Prediction markets appear to be a victory for the economic approach“、两法”complementarity … using both“(✓)。
对称金句(同时抵消神圣化与虚无化):用两句主笔亲核的话对冲——防虚无化的 Dana et al. 2019:”Prediction markets appear to be a victory for the economic approach, having yielded more accurate probability estimates than opinion polls or experts for a wide variety of events“(✓)∧ 防升格的 Manski 2006:”The above analysis refutes the notion that prices in prediction markets are ‘market probabilities.’“(✓)——它是真赢了经济学一局,但赢的不是”价格就是真概率”。
高风险/边界:本篇是金融与预报方法论的承重测试,不构成任何投注、交易、投资或监管政策建议;对市场机制的体检不构成对任何平台或参与者的品格指控。预测市场、民调、统计模型、超预报者都是严肃的预报工具;本篇祛魅的对象是”某一种已经永远赢了”的叙事,不是任何一方。
一、证据纪律:本篇在判什么,用什么尺
1.1 四种话语必须分开
预测市场的争论里,四种话语常被混为一谈,本篇全程分开记账:
- 理论机制(”价格能聚合分散信息”)——Hayek 的洞见、边际交易者假说、LMSR 的数学;这是为什么可能有效,不等于实际有多准。
- 经验校准/准确性(”市场价格与实际频率吻合到什么程度”)——IEM 的误差数字、Page-Clemen 的校准曲线、Brier 分数;这是实测,且校准≠准确≠有信息量(见 §2)。
- 本体解读(”价格是不是事件的真概率”)——Manski/Wolfers-Zitzewitz 的定理;这是价格数字该被读成什么,与前两者独立。
- 制度/治理外推(”该不该扩大它、该不该交给它做决策”)——Arrow et al. 的监管呼吁、futarchy、PAM;这是规范主张(ought),不能从前三者的经验事实(is)直接推出。
祛魅的核心,就是不让”机制上可能有效”滑成”实测很准”、不让”实测很准”滑成”价格就是真概率”、不让”是个好工具”滑成”该把治理交给它”。
1.2 六层 × 证据强度速览
| 层 | 审的是什么 | 证据强度 | 核验密度 | 一句裁决 |
|---|---|---|---|---|
| ① 真锚 | 信息聚合/理论根/是否常打平民调 | 硬(记录扎实) | ✓/◐一手多 | 真——祛魅 ≠ 打成赌博骗局 |
| ② 概念/术语 | 预测市场是不是单一物、群体智慧/校准是否被混用 | 硬(消歧清楚) | ✓/◐ | 一族工具;价格≠概率、校准≠准确、市场≠群体智慧 |
| ③ 论证/机制 | 为何能预测、边界在哪 | 中(真但有条件) | ◐ | 边际交易者/真话激励真;G-S 悖论与低信息场是边界 |
| ④ 命门一·价格≠真概率 | 价格能否读成事件真概率 | 强(否证升格) | ✓亲核 | 不能——价格是含风险偏好的分位数;favorite-longshot 系统偏 |
| ⑤ 命门二·流动性/操纵 | 薄市场能否被扭曲 | 强(但双面) | ✓亲核 | 能被扭曲∧常被抹平;wash trading 真∧操纵不获利 |
| ⑥ 叙事/对称 | 是否碾压专家、能否交给治理 | 软(两极皆软) | ✓/◐+部分⚠ | 升格只活在标题层;聚合/超预报者可平可胜;futarchy 反身性未立 |
二、真锚层:信息聚合是真的,市场常常打平或略胜民调
祛魅的第一件事,是承认预测市场不是骗局,它建立在一个严肃且被反复验证的洞见上。
分散知识与价格的信息功能(Hayek 1945, “The Use of Knowledge in Society” ✓,Econlib 经 AER 授权全文)逐字:”The peculiar character of the problem of a rational economic order is determined precisely by the fact that the knowledge of the circumstances of which we must make use never exists in concentrated or integrated form but solely as the dispersed bits of incomplete and frequently contradictory knowledge which all the separate individuals possess.” 价格系统的角色:”It is more than a metaphor to describe the price system as a kind of machinery for registering change, or a system of telecommunications which enables individual producers to watch merely the movement of a few pointers… in order to adjust their activities to changes of which they may never know more than is reflected in the price movement.” 他刻意用”marvel“一词——”I have deliberately used the word ‘marvel’ to shock the reader out of the complacency with which we often take the working of this mechanism for granted.” 预测市场正是把这一机制专门用于预报。
定义与三功能(Wolfers & Zitzewitz 2004, “Prediction Markets”, JEP 18(2) ◐)逐字定义:”markets where participants trade in contracts whose payoff depends on unknown future events”;三功能收束句:”prediction markets provide three important roles: 1) incentives to seek information; 2) incentives for truthful information revelation; and 3) an algorithm for aggregating diverse opinions.” 效率来源(含边际交易者):”In a truly efficient prediction market, the market price will be the best predictor of the event… This statement does not require that all individuals in a market be rational, as long as the marginal trade in the market is motivated by rational traders.“
经济学建制的联署背书(Arrow, Forsythe, …, Schelling, Shiller, V. Smith, Sunstein, Tetlock, Wolfers, et al. 2008, “The Promise of Prediction Markets”, Science 320 ◐,22 人联署)逐字:”There is mounting evidence that such markets can help to produce forecasts of event outcomes with a lower prediction error than conventional forecasting methods.” 53% 解读:”If the market price of an X contract is currently 53 cents, an interpretation is that the market ‘believes’ X has a 53% chance of winning.”
最硬的准确性数字(IEM Berg, Forsythe, Nelson, Rietz, “Results from a Dozen Years…”,主笔 pdftotext 亲核):样本”49 markets covering 41 elections in 13 countries”;核心对比句”The market outperformed polls in 9 of 15 cases… Across all elections, the average poll error was 1.91% while the average market error was 1.49% and 1.58% by the two measures.”(✓)Wolfers-Zitzewitz 2004 与 Arrow et al. Science 独立给出同源数字:选前一周市场误差 1.5 个百分点 vs 最终 Gallup 民调 2.1 个百分点(◐)。Snowberg-Wolfers-Zitzewitz 2012(Brookings/NBER) 概括:”markets generally exhibit lower statistical errors than professional forecasters and polls“、”markets have half the forecast error of polls“(◐)。
真锚裁决:信息聚合是真现象、理论根扎实、real-money 激励真、活跃选举市场常常打平或略胜民调、被经济学建制正式背书。祛魅”市场是神谕”绝不等于否认这些,更不等于把预测市场打成”赌博骗局”。 但同一批亲市场的作者已自设护栏——Wolfers-Zitzewitz 2004 提醒,把市场与”民调的机械外推”对比”may not provide a particularly compelling comparison“(◐),真正该比的是市场 vs 独立分析师/模型(见 §7)。守真锚里最诚实的一句仍是 IEM Berg(✓):”the market does about as well as the average poll, sometimes worse but often better, even if by a small margin.“
三、概念/术语层:预测市场是一族工具,价格、概率、校准、群体智慧都要消歧
“预测市场”不是单一物:real-money(IEM、PredictIt、Kalshi、Polymarket)与 play-money(如已倒闭的 Hollywood Stock Exchange)激励结构不同;连续双向拍卖(CDA) 需要买卖双方撮合,而 Hanson 的对数市场评分规则做市商(LMSR) 用一个自动做市商为薄市场补贴流动性——Hanson(◐):”a market scoring rule… acts like a continuous automatic market maker… can also act like a subsidized betting market… rational agents should expect a positive profit from participating, and do not need to find another person willing to make a matching bet.” 二元合约(是/否)与标量/指数合约(连续取值)解读也不同。
价格 ≠ 概率(命门一的预告):一个 0.60 的合约价,只有在一串假设下才≈”事件 60% 会发生”(见 §5)。
校准 ≠ 准确 ≠ 锐度/区分度——这是最常被偷换的一组。Gneiting, Balabdaoui & Raftery 2007(JRSS-B)(◐)黄金定义:”Calibration refers to the statistical consistency between the distributional forecasts and the observations… Sharpness refers to the concentration of the predictive distributions and is a property of the forecasts only.” Page-Clemen 2013 的操作定义(◐):”by ‘calibrated’ we mean that on average, when the trader’s belief is p, the expected frequency of the corresponding outcome occurring is p.” Brier 分数的 Murphy 分解(Brier score ◐)把误差拆成 可靠度(校准)− 分辨度(区分/锐度)+ 不可约不确定性——两个预报可以校准同样好、区分度却天差地别。“校准好”不等于”有信息量”:永远报基率的预报可以完美校准却毫无用处。
群体智慧 ≠ 预测市场:Surowiecki 2004 的四条件是多样性、独立性、去中心化、聚合(○ 名称多源一致,原书受版权锁未取逐字括注)。而预测市场的价格由少数边际交易者驱动(IEM Berg:”Marginal traders, not average traders, drive market prices and, therefore, predictions.” ✓),交易者盯着挂单相互调整——这与 Surowiecki 的”意见不被周围人决定”的独立性明显张力。预测市场只是”聚合”机制的一种实现,且未必满足群体智慧的全部前提。
概念裁决:预测市场是一族工具而非单一物;价格≠概率、校准≠准确、市场≠群体智慧。祛魅要点:把预测市场直接叫作”群体智慧的化身”、把价格直接读成”真概率”、把”校准好”读成”最准”,是三处术语升格,任一都能让一场辩论变成打稻草人。
四、论证/机制层:为什么能预测,以及它的边界
边际交易者与真话激励:市场之所以能聚合信息,机制在于边际交易者是理性且有信息的(Wolfers-Zitzewitz 2004:只要”the marginal trade… is motivated by rational traders” ◐)——不需要人人理性;而 real-money 让人有动机去收集信息并说真话(三功能之一、二)。套利压力把明显错价拉回。LMSR 则从机制上解决薄市场”找不到对手盘”的死结(§3)。
边界一·Grossman-Stiglitz 悖论:如果价格已经完全聚合了所有信息,那么谁也没有动力再去付费收集信息——于是价格又无从聚合。预测市场与 EMH 共享这一结构性张力(本库市场有效性 EMH 篇已就 G-S 悖论详裁,此处不重打)。现实里市场停在”部分有效”,留出让信息交易者获利的空间——这既是它能工作的原因,也是它不可能是完美神谕的原因。
边界二·低信息与选择性信息场:Wolfers-Zitzewitz 2004(◐)明说市场”unlikely to perform well when there is little useful intelligence to aggregate or when public information is selective, inaccurate or misleading.” 没有可聚合的分散私有信息时,市场不会凭空变准。
边界三·反直觉:更高流动性未必更准。Tetlock 2008, “Liquidity and Prediction Market Efficiency”(◐,TradeSports 数据):”I find that liquidity does not reduce—and sometimes increases—deviations of securities prices from financial and sporting event outcomes… the prices of liquid securities are not better calibrated, and actually exhibit poorer resolution than the prices of illiquid securities.” 这提醒:不能简单地把”薄市场=更差、厚市场=更准”当规律。
机制裁决:市场能预测的机制(边际交易者、真话激励、套利、LMSR 补贴流动性)是真的,但有边界——G-S 悖论意味着它只能部分有效、低信息场里不会变准、流动性与准确的关系甚至非单调。机制”可能有效”到实测”很准”之间,隔着这些边界;把机制读成”永远有效”是第一处外推。
五、命门一:市场价格不是事件的真概率
这是全篇最硬的本体命门。日常里”合约价 0.60=事件 60% 会发生”被当成定义,但一手文献否证了这个等同。
价格 refutes”市场概率”(Manski 2006, “Interpreting the Predictions of Prediction Markets”,主笔 pdftotext 亲核):在风险中性、异质信念的价格接受者假设下——摘要句”the price of a contract in a prediction market reveals nothing about the dispersion of traders’ beliefs and partially identifies the central tendency… The mean belief of traders lies in an interval whose midpoint is the equilibrium price.”;均值信念界”E(qm) ∈ (Bm², 2Bm − Bm²)… price does not generally equal the mean belief of traders“;结论句(✓):”The above analysis refutes the notion that prices in prediction markets are ‘market probabilities.’ … the price of an all-or-nothing futures contract does not equal the mean, median, or any other measure of the central tendency of traders’ beliefs. Instead, price solves equation (1).” 一个直觉版:价格 0.75 只告诉你”75% 的交易者认为概率高于 0.75″,均值主观概率落在一个区间里(例中 0.5625–0.935)。
何时≈、何时偏(Wolfers & Zitzewitz 2006, “Interpreting Prediction Market Prices as Probabilities” ◐):给出对称边界——在对数效用下价格恰等于均值信念”market prices are equal to the mean belief among traders”;但风险中性投资者”always invests her entire wealth whenever market prices diverge from her beliefs”,对应的是信念分布的”100−πth percentile“而非均值——”Interestingly, this is the only case considered by Manski (2004)”。风险偏好决定偏向哪端:”traders with low risk aversion… prices will be biased toward ½… a ‘favorite-longshot bias'”,而高风险厌恶则可能”biased toward zero or one”。结论:价格是被风险偏好折射的量,只在特定假设下≈均值信念,且”divergence between prices and average beliefs is greatest for prices closest to \$0 and \$1″。
风险中性概率的语言:无套利下二元合约价”are interpretable as risk-neutral probabilities“(预印本框架 ⚠ 未同行评审,仅作术语佐证);其严格经济学根据已由 Manski(”price reflects their risk preferences as well as their expectations”)与 Wolfers-Zitzewitz 2006 一手锁定。风险中性概率与物理/客观概率之间隔着一个风险溢价。
实证签名·favorite-longshot 偏差:
- Thaler & Ziemba 1988(JEP)(◐)定义句:”Favorites win more often than the subjective probabilities imply, and longshots less often.” 极端量级:”the typical 100-1 shot has real odds of about 730 to 1!”
- Snowberg & Wolfers 2010(JPE)(◐):赛马数据里”the rate of return to betting on horses with odds of 100/1 or greater is about −61%… while betting the favorite in every race yields losses of only 5.5%“,并判两类解释(风险偏好 vs 概率错觉)”observationally equivalent in win bet data”。
- Page & Clemen 2013(Economic Journal)(◐,Intrade 1,787 市场/50 万+笔):”high-likelihood events are underpriced, and low-likelihood events are over-priced.” 硬数字(S 形,压向 0.5):”a price of 0.20 is on average associated with a relative frequency of 15.3%; conversely, a price of 0.80 is on average associated with a relative frequency of 87.4%.” 政治市场更甚:”for a price of 0.20, the relative frequency… is 10.9% and, for a price of 0.80… 92.8%“,且”the longshot bias is stronger for the longer time horizon.“
一处必须写准的概念纠偏:文献记载的系统偏差方向是把价格向 0.5 压缩(低概率事件被高估、高概率事件被低估,即 favorite-longshot),而远期合约与政治市场更严重——不是通俗说的”极端处过度自信”。事实上 Page-Clemen 反而发现最极端的 0/1 附近校准尚可(其模型无法解释这一点,归因于极端处流动性枯竭)。
命门一裁决:截至可核证据,市场价格不是事件的真概率——它是被风险偏好折射的、信念分布的一个分位数(Manski/Wolfers-Zitzewitz),带有系统性的 favorite-longshot 偏差(Thaler-Ziemba/Snowberg-Wolfers/Page-Clemen),远期与政治市场偏得更多。“价格 0.60=事件 60% 会发生”在短周期厚市场里是好用的近似,但把它当成本体等同(”市场读出了真概率”)是升格。 这与本库模型成功不等于世界如此篇同构:工具的经验成功不发本体身份证。
六、命门二:薄市场可被扭曲,但操纵常被迅速抹平(对称)
这一层必须两面同审:升格侧要防”市场坚不可摧、不可操纵”,虚无侧要防”市场必被鲸鱼操纵、就是被操纵的赌局”。一手证据把两边都收进来。
防虚无化·操纵不易得逞:
- Hanson, Oprea & Porter 2006(JEBO)(主笔 pdftotext 亲核)实验:”manipulators are unable to distort price accuracy. Subjects without manipulation incentives compensate for the bias… by setting a different threshold”;”the presence of manipulators does not harm the information aggregation properties of trade”;非操纵者”effectively cancelled out the distortionary effects“(✓)。作者诚实自设限:”when agents suspect the presence of manipulators and know in what directions… manipulation is ineffective… Although these results are suggestive, they are by no means conclusive.“(✓)
- Rhode & Strumpf(◐)百年样本+田野实验总论:”the speculative attack initially moved prices, but these changes were quickly undone and prices returned close to their previous levels… little evidence that political stock markets can be systematically manipulated beyond short time periods.”
命门实证·薄市场确能被单笔大单打飞:
- 同一 Rhode-Strumpf(◐)记录 TradeSports 2004-10-15:”a series of thirty trades in less than a second dropped the price… prices fell by 44 points in just three minutes, suggesting that Bush went from a slight favorite to serious underdog”,且证实系”the large trades of a single investor“——但攻击者”losing over ten percent of his investment”,冲击不可持续、不获利。薄市场量能之稀(同文):”average volume of 9.7 shares (or \$56.51) per minute“。
- 2024 Polymarket 的”Théo”:Forbes(◐)——四账户合计押 Trump “\$28.6 million“;The Block 引 Chainalysis(◐)估其”net profit of nearly \$79 million“,法国博彩监管 ANJ 介入调查。对称关键:Polymarket 官方回应(Forbes 直引 ✓)——该交易员”is taking a directional position based on personal views of the election”,平台”has not identified any information to suggest that this user manipulated, or attempted to manipulate, the market.“(”大户具体如何推高赔率”的因果逐字因 FT 付费墙未取到,本篇不以一般原理冒充该具体因果——见不确定点。)
wash trading(虚假成交量):Fortune(◐)——Chaos Labs 估”wash trading constituted around one-third of trading volume on Polymarket’s presidential market”;哥伦比亚大学研究(CoinDesk 转述)(◐)”nearly 25% of Polymarket’s trading volume may be fake”;Chainalysis(◐)点出加密预测市场里 wash trading 的独特危害——”in prediction markets, it can be used to fabricate consensus.”(哥大研究作者自陈为估算、未指控平台本身,Polymarket 未置评。)
制度脆弱:预测市场的历史是一部被监管反复冲击、平台反复倒闭/受限的历史——
- Intrade:CFTC 2018 新闻稿(◐)——2012 起诉、2013 关停,2018 法院判 Intrade/TEN 缴”\$3 million civil monetary penalty“。
- PredictIt:Clarke v. CFTC, 5th Cir. 2023(◐)——2014 拿到 no-action letter、2022 被 CFTC 撤销,第五巡回判”the CFTC’s rescission… was likely arbitrary and capricious“。
- Kalshi:KalshiEX LLC v. CFTC, D.C. Cir. 2024-10-02(主笔 pdftotext 亲核)——CFTC 曾以选举合约”amount to gaming or election gambling”为由禁止(援引 CEA 特别规则把 terrorism/assassination/war/gaming 并列 ✓),地区法院判”elections are not games”、上诉法院拒绝 stay 放行交易(✓);同案确认 Polymarket 2022 与 CFTC 和解 缴 140 万美元罚款并限制为非美用户(◐)。
命门二裁决:薄市场确实可被单笔大单在数分钟内打飞(Rhode-Strumpf 44 点、Théo 2860 万美元、wash trading 占两成到三成),制度上也脆弱(Intrade/PredictIt/Kalshi 反复被监管冲击)——这一层否证”市场坚不可摧、价格永远干净”的升格。但同样一手的证据也否证虚无化:操纵往往被迅速抹平且不获利(Rhode-Strumpf/Hanson-Oprea-Porter),Théo 被平台判定为”directional position”而非操纵。两面都不可偏废:能被扭曲 ∧ 常被抹平。
七、叙事/对称层:升格活在标题层,聚合算法与超预报者可平可胜
升格侧一·”市场碾压专家/民调”被聚合算法与超预报者打平甚至反超:
- Dana, Atanasov, Tetlock & Mellers 2019, “Are markets more accurate than polls? … ‘just asking'”(JDM)(主笔 HTML 亲核,同 GJP 数据、37,000 预测/113 事件):简单平均的自报信念 Brier 0.283 输给市价 0.227;但”the hybrid method registering a Brier score of 0.218… adding belief extremization reduced mean Brier scores to 0.210“,二者组合”was significantly more accurate than prediction-market prices alone“(t(113)=2.92,p=0.004 ✓),directionally 胜市价”on 85% of the 113 forecasting questions”(✓);总结”self-reported beliefs were at least as informative as prediction market prices when beliefs were properly aggregated… Many economists may now be surprised by how informative it is to ‘just ask.'”(✓)一处关键不对称:”extremizing helped the aggregate, but did not help market prices“(✓)。
- Atanasov et al. 2017, “Distilling the Wisdom of Crowds: Prediction Markets vs. Prediction Polls”(Management Science)(⚠ 付费墙,摘要级)+ 后续 Crowd Prediction Systems(◐ 全文)逐字复述:”CDA markets underperform team-based prediction polls when question resolutions are months away but are approximately tied in accuracy in the last few weeks”;主结论”prediction markets and prediction polls are approximately tied in terms of accuracy”,而”the advantages of elite over sub-elite crowds are substantially larger than the differences between prediction markets and prediction polls”——谁在预测 > 用什么机制预测。
- Mellers et al. 2014, “Psychological Strategies for Winning a Geopolitical Forecasting Tournament”(◐):训练/组队/精英化的 Brier(越低越好)——未训练散户临结算 0.26,超预报者 0.07;精英团队”were more accurate than professional intelligence analysts with access to classified information“(引 Goldstein et al. 2016)。Baron et al. 2014(◐)证 extremizing 对聚合的 Brier 增益(低专业组 0.196→0.139)。
升格侧二·极端事件失灵(对称:既认错,又给概率辩护):
- Brexit 2016:Gelman & Rothschild, Slate(◐)——市场”holding steady with a predicted probability 25 percent during the week leading up to the vote”(Leave),实际 Leave 以 51.9% 胜。更深的一刀是反身性:”pollsters looked to the markets, which were firmly supporting ‘Remain,’ and doubted their own polls“——价格被当信息后,反过来污染了本该独立的民调。
- Trump 2016:Josh Hannah, Medium(◐)——选前 Betfair”Clinton has a 78% chance… Trump has a 22% chance”。概率辩护一侧(防虚无化):同文”the market is NOT saying ‘Clinton will win’ and the market is not ‘wrong’ if Trump wins”——25% 的事件本就该有 25% 的时候发生。两侧都成立:市场没”预言”错,但也远非神谕。
升格侧三·治理外推(futarchy)——把定价工具当制度:
- Robin Hanson, “Shall We Vote on Values, But Bet on Beliefs?” (JPP 2013)(◐)提案逐字:”we could ‘vote on values, but bet on beliefs‘… when speculative markets clearly estimate that a proposed policy would increase national welfare, that policy becomes law”,并自陈是”taking the idea to an extreme”。
- 反身性/Goodhart 的致命批评:一旦决策市场的输出被拿去行动,它就无法再估计因果效应——LessWrong “Futarchy is Parasitic…”(◐):”conditional decision markets are structurally incapable of estimating causal policy effects once their outputs are acted upon… There is no payout structure that simultaneously incentivizes… causal knowledge and allows that knowledge to be acted upon.” Rethink Priorities(◐)补 Goodhart 与财富加权:”when a measure becomes a target, it ceases to be a good measure”;信念”weighted according to the average wealth of traders”。这与本库测量代理性/Goodhart 篇同构。
- DARPA PAM “恐怖期货” 2003(对称样本):虚无侧——CBS/AP(◐)参议员 Wyden”ridiculous… grotesque”、Dorgan”unbelievably stupid”;国会记录(◐)Daschle”the most irresponsible… poorly thought out”。升格侧——Wolfers-Zitzewitz 2004(◐)反讽辩护:”the aftermath of the DARPA controversy provided a vivid illustration of the power of markets to provide information… rather than spend political capital defending a tiny program, the proposal was dropped.”
叙事两极(供对称打靶):升格侧 CNN 2024(✓)标题”How prediction markets saw something the polls and pundits didn’t”;虚无侧 The Nerve(◐)”money masquerading as evidence”、Undark(◐)监管者视其为”gambling platforms that allow market manipulation”。两极皆软——同一篇 CNN 自带打脸:”In 2016… PredictIt gave Hillary Clinton an 82% chance… the smart money was wrong”,并紧接一句最锋利的居中话(✓):”just like polls, prediction markets are far from perfect.“
叙事裁决:升格只活在标题层——聚合算法与超预报者在硬基准上打平甚至反超市场(Dana 2019 p=0.004、Atanasov “approximately tied”、Mellers 超预报者 0.07),极端事件处市场给的 25%/22% 既非”错”也非神谕,治理外推(futarchy)撞上反身性/Goodhart 而未立。但两极皆软:一次翻车不证方法死、一个鲸鱼不证市场必假、”就是赌博”抹不掉 IEM 的记录与 real-money 的校准。
八、母裁决、灵魂句与对称三向红线
母裁决:信息聚合是真的、活跃厚市场校准常常良好且选前误差约为民调一半、real-money 逼真话是真机制、预测市场是被顶尖经济学家联署背书的正规工具——真的不是”市场价格=事件真概率”和”预测市场是永远打败一切的终极预报神谕”这两层被声称的胜利。价格是被风险偏好折射的、信念分布的一个分位数(Manski”refutes market probabilities”),系统性 favorite-longshot 偏差把价格压向 0.5;薄市场可被大户与 wash trading 扭曲,虽常被抹平;恰当聚合的自报信念与精英超预报者在硬基准上显著打平甚至反超市场。
灵魂句:一个真的信息聚合机制(Hayek 1945)、一份真的准确记录(IEM 选前误差约民调一半)、一次真的经济学胜利(Dana 2019“a victory for the economic approach”)——这些都是真的,也真有 22 位顶尖经济学家为它联署。但把”价格常常很准”(一个校准事实)讲成”价格就是真概率”(一个本体等同),中间隔着一个被 Manski 一手否证的等号和一条系统性的 favorite-longshot 偏差;而把”是个好工具”讲成”永远打败一切、该交给它做决定”,则被恰当聚合的自报信念、被超预报者、被反身性一一挡住。预测市场没有消除不确定性,它只是给分散的信念标了一个价——这个价在足够厚的市场里往往很准,但它始终是一枚被风险、流动性与下注结构折射过的概率,不是世界的真概率,更不是能替你做决定的神谕。
对称三向红线:
- 不升格:校准好 ≠ 价格是真概率(Manski“refutes market probabilities”)· 打平/略胜民调 ≠ 碾压专家(Dana 2019 聚合显著胜市价、Atanasov“approximately tied”)· 聚合机制 ≠ 永远有效(G-S 悖论 + Wolfers-Zitzewitz 低信息场局限)· 价格 ≠ 预言(Brexit 25%/Trump 22% 既不”错”也非神谕)。
- 不虚无化:一次翻车(2016)≠ 方法死(概率本就该有 25% 的时候发生)· 一个鲸鱼(Théo)≠ 市场必假(被平台判定为 directional position、大冲击常被抹平)· 操纵个案 ≠ 不可聚合信息(Hanson-Oprea-Porter“unable to distort price accuracy”、Rhode-Strumpf“quickly undone”、IEM 记录硬、Dana 2019“victory for the economic approach”)。
- 不污名泛化:赌博性质 ≠ 无认识价值(它同时是下注与信息聚合器)· 监管争议 ≠ 参与者道德败坏(Kalshi v. CFTC 是”gaming”定义之争,PAM 关停是政治而非因市场邪恶)· wash trading 个案 ≠ 领域腐败(研究者自陈估算、平台受监管纠错)。
对称金句:用两句主笔亲核的话把两侧同时钉住——防虚无化 Dana et al. 2019(✓):”Prediction markets appear to be a victory for the economic approach“;防升格 Manski 2006(✓):”The above analysis refutes the notion that prices in prediction markets are ‘market probabilities.’“
高风险/边界:金融题——本篇不构成任何投注、交易、投资或监管政策建议;对市场机制的体检不构成对任何平台或参与者的品格指控。
九、自指与关联
- 与 EMH 篇:”永远不可战胜”的恒真冒充,与本篇”永远打败民调”的普适碾压是同一种普适越界;两篇共享 Grossman-Stiglitz 悖论这一结构性张力(价格完全有效则无人有动力采集信息)。
- 与 测量代理性/Goodhart 篇:futarchy 的反身性(”once outputs are acted upon, decision markets can’t estimate causal effects”)与 Brexit”pollsters doubted their own polls”都是 Goodhart 的活体——价格一旦从测量变成目标/决策依据,其作为聚合器的效度就退化。
- 与 黑天鹅/肥尾篇:极端事件(Brexit/Trump 2016)处的失灵,以及 favorite-longshot 在小概率端的系统偏差,是尾部风险在预报市场里的映像。
- 与 模型成功不等于世界如此篇、有效理论 vs 本体论篇:市场价格”校准好”(工具成功)不等于”价格是真概率”(本体身份)——解释/预测的经验成功只发一张分级许可证,不发本体身份证。
不确定点(如实标注)
- “大户如何推高赔率”的具体因果逐字未取到:Théo 案已锁定持仓集中度(占市场逾 1%、Trump 侧前五占四)与平台否认操纵;但”Polymarket 的 Trump 赔率因该鲸鱼而高于其他市场”的 FT 报道因付费墙未取到 verbatim。本篇不以 Chainalysis 的一般原理冒充该具体因果。
- Atanasov 2017 内部百分比数字(如”13% lower””22%/30% more accurate”)为付费墙摘要级(⚠),本篇承重改用同数据、开放获取、已主笔亲核的 Dana 2019(0.227 vs 0.210,p=0.004)。
- Sethi et al. 2025″2024 总统赛抛硬币更准”、Erikson-Wlezien 2012″prices add nothing beyond polls”(⚠ 付费墙/摘要转述):作为防升格辅证提及,但未上承重;命门三的承重压在 Dana 2019 / Mellers 2014 / Atanasov 上。
- Surowiecki 四条件的逐字括注:原书受版权锁未取到(○),四条件名称多源一致可用,精确措辞待实体书 p.10。
- Superforecasting 原书逐字:Internet Archive 借阅限制、Google Books 无 snippet,取证失败(○);以 Mellers 2014 论文替代。
- 部分平台校准数字(Manifold 当日 Brier 0.174 为快照会变;Metaculus FAQ 403 未直连;哥大 wash-trade 分类别百分比为摘要级)标 ◐/⚠,不作承重。
- 两条 2026 加密预测市场校准预印本(arXiv 2602.19520 / 2510.15205)未同行评审(⚠),仅作当代佐证,命门一承重压在 Manski + Page-Clemen + Snowberg-Wolfers 三根同行评审硬梁上。
后续问题
- Kalshi/Polymarket 2024–2026 大样本校准的同行评审复核:把本篇 ◐/⚠ 的加密市场校准结论(”政治市场压向 50%””长周期偏差”)升到同行评审一手。
- “市场 vs 模型(非民调)”的判决性对比:Wolfers-Zitzewitz 自陈真正该比的是市场 vs 独立分析师/统计模型;2024 大选的 Polymarket vs 538/Silver Bulletin 的严肃学术裁决(Sethi 2025 等)待取正文。
- futarchy 的活体案例:是否存在真实决策市场被”反身性”污染的可核案例,可与 Goodhart 篇合写。
- 信息级联/社交媒体篇(backlog 第五梯队邻接项):预测市场的反身性(价格塑造它所预测的结果)与信息级联同源,可作社会侧姊妹篇。
关联笔记
- 独占边界:EMH审”不可战胜”、黑天鹅审尾部、Goodhart审”测量变目标”;本篇独占”预测市场作为聚合器与神谕之间的承重审计”。
- 方法论自指:本篇再次示范”工具的经验成功 ≠ 本体身份证”——与本库有效理论 vs 本体论、模型成功同一形状。
主要来源清单(分层 · 标准化 · 均带链接)
体例:每条给作者/机构、标题、年份/出处、可点击链接、访问方式与核验标记。付费墙/访问失败均如实标注。
① 真锚层 · 理论根与准确性记录
- Friedrich A. Hayek, “The Use of Knowledge in Society,” American Economic Review 35(4), 1945, pp.519–530 — https://www.econlib.org/library/Essays/hykKnw.html (Econlib 经 AER 授权全文;curl 全文,✓主笔可核)
- Justin Wolfers & Eric Zitzewitz, “Prediction Markets,” Journal of Economic Perspectives 18(2), 2004, pp.107–126,DOI 10.1257/0895330041371321 — https://jmvidal.cse.sc.edu/library/wolfers04a.pdf (pdftotext 全文,◐)
- K. J. Arrow, R. Forsythe, …, T. C. Schelling, R. J. Shiller, V. L. Smith, C. R. Sunstein, P. E. Tetlock, J. Wolfers, E. Zitzewitz(22 人联署), “The Promise of Prediction Markets,” Science 320(5878), 2008, pp.877–878,DOI 10.1126/science.1157679 — https://eriksnowberg.com/papers/science.pdf (pdftotext 全文,◐)
- Joyce Berg, Robert Forsythe, Forrest Nelson, Thomas Rietz, “Results from a Dozen Years of Election Futures Markets Research”(Handbook of Experimental Economics Results 第 80 章,2008;核 2003 工作稿) — https://www.biz.uiowa.edu/faculty/trietz/papers/iemresults.pdf (pdftotext 全文,✓主笔亲核守真金句与 1.91/1.49 数字)
- Berg, Nelson & Rietz, “Prediction market accuracy in the long run,” International Journal of Forecasting 24(2), 2008, pp.285–300,DOI 10.1016/j.ijforecast.2008.03.007 — https://www.biz.uiowa.edu/faculty/trietz/papers/long%20run%20accuracy.pdf (pdftotext 全文,◐;MAE 1.82 vs 3.37、74% 频次)
- Snowberg, Wolfers & Zitzewitz, “Prediction Markets for Economic Forecasting,” NBER WP 18222 / Brookings, 2012 — https://www.brookings.edu/wp-content/uploads/2016/06/13-prediction-markets-wolfers.pdf (pdftotext 全文,◐;”half the forecast error””not a panacea”)
- Robin Hanson, “Logarithmic Market Scoring Rules for Modular Combinatorial Information Aggregation,” 2002 工作稿 / Journal of Prediction Markets 1(1), 2007 — http://hanson.gmu.edu/mktscore.pdf (pdftotext 全文,◐;LMSR 自动做市商)
- James Surowiecki, The Wisdom of Crowds, Doubleday, 2004(四条件) — https://en.wikipedia.org/wiki/The_Wisdom_of_Crowds (原书受版权锁,四条件名称多源一致 ○,逐字括注待实体书 p.10)
② 概念/术语层 · 校准与准确性的定义
- Tilmann Gneiting, Fadoua Balabdaoui, Adrian Raftery, “Probabilistic forecasts, calibration and sharpness,” JRSS-B 69(2), 2007, pp.243–268 — https://sites.stat.washington.edu/raftery/Research/PDF/Gneiting2007jrssb.pdf (pdftotext 全文,◐;calibration vs sharpness 黄金定义)
- “Brier score”(含 Murphy 1973 分解 REL−RES+UNC) — https://en.wikipedia.org/wiki/Brier_score (WebFetch,◐;一手为 Murphy, J. Appl. Meteorol. 12(4), 1973, pp.595–600)
- Manifold Markets, “Platform calibration”(当日 Brier 0.174 快照) — https://manifold.markets/calibration (WebFetch,◐)
③ 命门一 · 价格 ≠ 真概率
- Charles F. Manski, “Interpreting the Predictions of Prediction Markets,” Economics Letters 91(3), 2006, pp.425–429,DOI 10.1016/j.econlet.2006.01.004(核底本 NBER WP 10359) — https://www.nber.org/system/files/working_papers/w10359/w10359.pdf (✓主笔 pdftotext 亲核 “refutes… ‘market probabilities'”)
- Justin Wolfers & Eric Zitzewitz, “Interpreting Prediction Market Prices as Probabilities,” NBER WP 12200, 2006 — https://www.nber.org/system/files/working_papers/w12200/w12200.pdf (pdftotext 全文,◐;对数效用≈均值、风险中性→100−π 分位)
- Richard Thaler & William Ziemba, “Anomalies: Parimutuel Betting Markets,” JEP 2(2), 1988, pp.161–174,DOI 10.1257/jep.2.2.161 — https://pubs.aeaweb.org/doi/pdfplus/10.1257/jep.2.2.161 (pdftotext 全文,◐;favorite-longshot 定义)
- Erik Snowberg & Justin Wolfers, “Explaining the Favorite-Long Shot Bias,” JPE 118(4), 2010, pp.723–746,DOI 10.1086/655844(核底本 NBER WP 15923) — https://www.nber.org/system/files/working_papers/w15923/w15923.pdf (pdftotext 全文,◐;−61% vs −5.5%)
- Lionel Page & Robert Clemen, “Do Prediction Markets Produce Well-Calibrated Probability Forecasts?” The Economic Journal 123(568), 2013, pp.491–513,DOI 10.1111/j.1468-0297.2012.02561.x — https://people.duke.edu/~clemen/bio/Published%20Papers/45.PredictionMarkets-Page&Clemen-EJ-2013.pdf (pdftotext 全文,◐;0.20→15.3%、政治 0.20→10.9%)
④ 命门二 · 操纵、流动性与制度脆弱
- Robin Hanson, Ryan Oprea, David Porter, “Information aggregation and manipulation in an experimental market,” JEBO 60(4), 2006, pp.449–459 — http://mason.gmu.edu/~rhanson/biastest.pdf (✓主笔 pdftotext 亲核 “unable to distort price accuracy”)
- Paul Rhode & Koleman Strumpf, “Manipulating Political Stock Markets: A Field Experiment and a Century of Observational Data,” 2008 工作稿 — https://users.wfu.edu/strumpks/papers/ManipIHT_June2008%28KS%29.pdf (pdftotext 全文,◐;44 点/3 分钟、quickly undone);配套 “Historical Presidential Betting Markets,” JEP 18(2), 2004 — https://users.wfu.edu/strumpks/papers/JEP_2004.pdf
- Paul C. Tetlock, “Liquidity and Prediction Market Efficiency,” Columbia, 2008 — https://business.columbia.edu/sites/default/files-efs/pubfiles/3098/Tetlock_SSRN_Liquidity_and_Efficiency.pdf (pdftotext 全文,◐;流动性不改善校准)
- Derek Saul, “Who Is Polymarket’s Trump Whale…,” Forbes, 2024-10-24 — https://www.forbes.com/sites/dereksaul/2024/10/24/who-is-polymarkets-trump-whale-site-reveals-french-trader-bet-28-million-on-trump-win/ (WebFetch,◐;\$28.6M + 平台否认操纵 ✓)
- “French Polymarket whale… net profit,” The Block, 2024-11-07(引 Chainalysis) — https://www.theblock.co/post/324996/french-polymarket-whale-us-election-profit-france-ban (WebFetch,◐;净利约 \$79M、ANJ 调查)
- “Polymarket… crypto wash trading,” Fortune, 2024-10-30 — https://fortune.com/crypto/2024/10/30/polymarket-trump-election-crypto-wash-trading-researchers/ (WebFetch,◐;Chaos Labs 约 1/3)
- “Polymarket’s trading volume may be 25% fake, Columbia study finds,” CoinDesk, 2025-11-07 — https://www.coindesk.com/markets/2025/11/07/polymarket-s-trading-volume-may-be-25-fake-columbia-study-finds (WebFetch,◐)
- Chainalysis, “Crypto Prediction Markets Explained,” 2026-05-07 — https://www.chainalysis.com/blog/crypto-prediction-markets/ (WebFetch,◐;”fabricate consensus”)
- CFTC Press Release 7758-18(Intrade \$3M penalty, 2018) — https://www.cftc.gov/PressRoom/PressReleases/7758-18 (WebFetch,◐)
- Clarke v. CFTC, No. 22-51124 (5th Cir. 2023-07-21)(PredictIt) — https://www.ca5.uscourts.gov/opinions/pub/22/22-51124-CV0.pdf (pdftotext 全文,◐;”arbitrary and capricious”)
- KalshiEX LLC v. CFTC, No. 24-5205 (D.C. Cir. 2024-10-02) — https://media.cadc.uscourts.gov/opinions/docs/2024/10/24-5205-2077790.pdf (✓主笔 pdftotext 亲核;”gaming”越界之争、拒绝 stay);CFTC 8478-22(Polymarket 2022 \$1.4M 和解) — https://www.cftc.gov/PressRoom/PressReleases/8478-22
⑤ 命门三 · 聚合对决、极端事件与治理
- Jason Dana, Pavel Atanasov, Philip Tetlock, Barbara Mellers, “Are markets more accurate than polls? The surprising informational value of ‘just asking’,” Judgment and Decision Making 14(2), 2019, pp.135–147 — https://www.sas.upenn.edu/~baron/journal/18/18919/jdm18919.html (✓主笔 HTML 亲核;Brier 0.227 vs 0.210,p=0.004,”victory for the economic approach”)
- Pavel Atanasov et al., “Distilling the Wisdom of Crowds: Prediction Markets vs. Prediction Polls,” Management Science 63(3), 2017, pp.691–706,DOI 10.1287/mnsc.2015.2374 — https://pubsonline.informs.org/doi/10.1287/mnsc.2015.2374 (⚠ 付费墙,摘要级);后续 “Crowd Prediction Systems” — https://faculty.wharton.upenn.edu/wp-content/uploads/2016/11/Crowd-Prediction-Systems.pdf (pdftotext 全文,◐;”approximately tied”)
- Barbara Mellers et al., “Psychological Strategies for Winning a Geopolitical Forecasting Tournament,” Psychological Science 25(5), 2014, pp.1106–1115,DOI 10.1177/0956797614524255 — https://escholarship.org/content/qt4rg4n9vr/qt4rg4n9vr.pdf (pdftotext 全文,◐;超预报者 Brier 0.07)
- Jonathan Baron, Barbara Mellers, Philip Tetlock, Eric Stone, Lyle Ungar, “Two Reasons to Make Aggregated Probability Forecasts More Extreme,” Decision Analysis 11(2), 2014, pp.133–145,DOI 10.1287/deca.2014.0293 — https://faculty.wharton.upenn.edu/wp-content/uploads/2015/07/2015—two-reasons-to-make-aggregated-probability-forecasts_1.pdf (pdftotext 全文,◐;extremizing 增益)
- Andrew Gelman & David Rothschild, “Something’s Odd About the Political Betting Markets,” Slate, 2016-07-12 — https://slate.com/news-and-politics/2016/07/why-political-betting-markets-are-failing.html (curl 全文,◐;Brexit 25%、反身性)
- Josh Hannah, “Betfair and the 2016 Presidential Election,” Medium — https://medium.com/@jdh/betfair-and-the-2016-presidential-election-6833849e4549 (WebFetch,◐;Clinton 78%/Trump 22% + 概率辩护)
- Robin Hanson, “Shall We Vote on Values, But Bet on Beliefs?” Journal of Political Philosophy 21(2), 2013, pp.151–178,DOI 10.1111/jopp.12008 — https://mason.gmu.edu/~rhanson/futarchy2013.pdf (pdftotext 全文,◐)
- “Futarchy is Parasitic on What It Tries to Govern,” LessWrong — https://www.lesswrong.com/posts/mW4ypzR6cTwKqncvp/futarchy-is-parasitic-on-what-it-tries-to-govern (reader 代理全文,◐;反身性批评);”Issues with futarchy,” Rethink Priorities — https://rethinkpriorities.org/research-area/issues-with-futarchy/ (WebFetch,◐;Goodhart)
- “Terror Betting Torpedoed,” CBS/AP, 2003-07-31(PAM) — https://www.cbsnews.com/news/terror-betting-torpedoed-31-07-2003/ (WebFetch,◐);国会记录 2003-07-29 — https://sgp.fas.org/congress/2003/s072903.html
- “How prediction markets saw something the polls and pundits didn’t,” CNN, 2024-11-08 — https://www.cnn.com/2024/11/08/business/polymarket-election-trump-nightcap (reader 代理全文,✓;”far from perfect”、2016 打脸)
- Ramin Skibba, “Prediction Markets Aren’t Likely to Replace Polling,” Undark, 2026-05-05 — https://undark.org/2026/05/05/prediction-markets-polling/ (◐);Ian Tucker, The Nerve, 2026-05-15 — https://www.thenerve.news/p/polymarket-guide-betting-trump-prediction-market-venezuela-farage-uk-regulation-synthetic-truth (◐)
辅证/未上承重(如实标注)
- Robert Erikson & Christopher Wlezien, “Markets vs. polls as election predictors,” Electoral Studies 31(3), 2012, pp.532–539 — https://www.sciencedirect.com/science/article/abs/pii/S0261379412000467 (⚠ 付费墙,”prices add nothing” 摘要转述,未上承重)
- Rajiv Sethi et al., “Political Prediction and the Wisdom of Crowds,” CI ’25, 2025,DOI 10.1145/3715928.3737483 — https://dl.acm.org/doi/10.1145/3715928.3737483 (⚠ 付费墙,”coin flip more accurate” 摘要转述,未上承重)
证据纪律备注
- 不凭记忆:六层命题全部来自六捆并行一手取证(多为 curl+pdftotext/HTML 全文级)+主笔亲核;凡二手、摘要、付费墙未取正文、未同行评审者一律标 ⚠/○,且不上承重。
- 命门承重的三根硬梁:命门一压在 Manski(✓主笔亲核)+ Wolfers-Zitzewitz 2006 + Page-Clemen;命门二压在 Hanson-Oprea-Porter(✓)+ Rhode-Strumpf + Kalshi/Intrade/PredictIt 判决(✓);命门三压在 Dana 2019(✓主笔亲核 p=0.004)+ Mellers 2014 + Atanasov。付费墙的 Atanasov 内部百分比、Sethi”抛硬币”、Erikson-Wlezien”prices add nothing”均降格为辅证。
- 对称纪律:每一层都同时取升格侧与虚无侧的一手证据(操纵能得逞 ∧ 被抹平;市场很准 ∧ 聚合可反超;价格像概率 ∧ 不是真概率),不做单边打靶。
- 纠错入账:Hanson LMSR 系 2002 工作稿/2007 刊出(非 2003 的姊妹作);文献里的系统偏差方向是”压向 0.5″(favorite-longshot、远期与政治更甚)而非”极端处过度自信”,且 Page-Clemen 发现 0/1 极端处校准反而尚可——正文已按一手写准。
- 高风险边界:金融题不作投注/交易/投资/监管政策建议;机制体检不构成对任何平台或个人的品格指控。