✦   ✦   ✦

breaking news

News, read through The Primal Race
← 全部评论 · all commentary

三趴人开盘,全民当先知Three Percent Sets the Odds

哲学 文化层 · 结构层 · 元暴力 The Guardian ↗ 2026-11-02 § 链接
新闻摘要《卫报》评论作者 Judith Levine 指出,预测市场平台(如 Polymarket、Kalshi)以「群体智慧」为名吸引散户下注,但研究显示准确率主要来自约 3% 的专业账户,四分之三的利润流向 0.1% 的账户。她认为这类平台影响选民对选举结果的预期甚至投票行为,并援引特朗普家族对 Polymarket、Kalshi 的投资与任职关系,主张禁止对政治选举下注。
The newsGuardian columnist Judith Levine argues that prediction markets such as Polymarket and Kalshi sell themselves as "wisdom of the crowd" while accuracy comes from roughly 3% of skilled accounts and three-quarters of profits go to 0.1% of accounts. She says the platforms shape voters' expectations and behavior, notes Trump family ties to both firms, and calls for banning election betting.
群体智慧是招牌,定价权在三趴人手里。
"Wisdom of the crowd" is the sign; three percent holds the pricing power.

预测市场卖的是「群体智慧」,赚的是「三趴人定价」的差价。Levine 引的伦敦商学院与耶鲁的新工作论文说得很清楚:几乎每个下注者的准确率都不比抛硬币强,准确度来自约 3% 的持续盈利账户。人群贡献了大部分交易量,几乎不贡献信息。

这就是第三章那套「制造可能性的艺术」换了个壳。入口从宗教、浪漫爱、色情产业换成了赔率,机制一模一样:夺取话语权,夺取解释权,最后夺取「什么是事实」的制造权。只是这次连「事实」都不用编,直接给出一个数字,再让媒体把它当成民调来引。

最妙的是它自称反精英。Kalshi 的发言人说这是「自下而上的数据,不是精英驱动的」。可《华尔街日报》拆了 160 万个钱包:四分之三的利润进了 0.1% 的账户,Polymarket 上 2000 个账户净赚近五亿美元。这些人买每年二十万美元的数据流,用算法比任何人反应都快,一天做上万笔交易。这叫哪门子的「人群」?这叫华尔街换了个界面,还顺手把「精英」这个词的定价权也拿走了。

散户的钱养着庄家的利润,庄家的赔率再回头塑造选民的预期。Brennan Center 资助的研究说预测市场赔率「显著改变选民的预期」;Barnard 的经济学家 Rajiv Sethi 说,人相信什么将要发生,会真的影响什么发生。CoinMarketCal 那份给加密交易者的通讯说得更直白:一个资金充足的人可以在选民情绪毫无变化的情况下推动一份流动性稀薄的合约。

于是整条链闭合了:多数人出钱,少数人定价,媒体转述定价,选民照着定价调整预期,预期再变成投票行为。加尔通那条公式在这里有了新算法——差额不是被谁用枪逼出来的,是被一个数字说服出来的。文化层和结构层在这里咬合得严丝合缝,而元层那根轴照旧:解释权归谁,谁的赔率就算事实。

腐败的部分 Levine 写得很不客气。George Santos 买了自己会不会出席国情咨文的合约,发帖说会去,然后待在家里,赚了一万八,律师说「绝对没有欺骗任何人的意图」。今年春天有幕僚拿自己的候选人下注。Polymarket 查了几十起中期选举相关的潜在欺诈。而监管者 Michael Selig 上任前就是预测市场和加密行业的公司律师;小特朗普的 1789 Capital 往 Polymarket 投了五亿美元,Kalshi 在特朗普就职第二天宣布「无比兴奋」地聘他当付费顾问。

Levine 说该禁。禁当然该禁,但禁的理由不能停在「赌博有害」。她自己在这点上倒是诚实——手机上的赌场不会把每个人都变成赌鬼,赌板球、赌外星人,随你。问题在于民主不是板球。政治本来就是一场被钱严重污染的博弈,预测市场做的不是污染它,是给它装上老虎机的界面,再把赔率包装成民意。

还有一个数字值得盯着:中期选举的交易量,反贪数据组织说到 2026 年已超过 2.2 亿美元,CNN 引两家平台发言人说合计 14 亿。同一件事差六倍,而这两家的估值分别是 220 亿和 90 亿。一门连成交量都说不清的生意,正在给民主定价。

真正被卖掉的不是选票,是「谁有资格说话」这件事本身。以前是民调机构、电视评论员、政党机器;现在是一群连名字都不出现在报道里的账户。他们不需要说服你,只需要让你看见一个数字,然后让你以为那是你自己想出来的。

Prediction markets sell "wisdom of the crowd" and pocket the spread from a crowd that isn't wise. The London Business School and Yale working paper Levine cites is blunt: nearly every bettor's record is no better than chance, and the accuracy comes from about 3% of persistently skilled accounts. The crowd supplies most of the volume and almost none of the information.

This is Chapter Three's "art of manufacturing possibilities" in a new casing. The entry point used to be religion, romance, porn; now it's odds. Same mechanism: seize the discourse, seize the interpretation, seize the power to manufacture what counts as fact. Only this time nobody has to invent the fact. You publish a number and let the press cite it as a poll.

And it bills itself as anti-elite. A Kalshi spokesperson calls the data "bottom-up, not elites-driven." The Wall Street Journal analyzed 1.6 million wallets: three-quarters of profits went to 0.1% of accounts, and 2,000 accounts on Polymarket netted nearly half a billion dollars. These people buy $200,000-a-year data streams and run algorithms faster than any human, tens of thousands of trades a day. That isn't a crowd. That's Wall Street with a new interface, having also seized the right to define the word "elite."

Retail money funds the house's profits, and the house's odds then reshape what voters expect. A Brennan Center-funded study finds prediction market odds shift expectations substantially; Barnard economist Rajiv Sethi notes that believing something will happen can make it happen. CoinMarketCal, a crypto newsletter, puts it plainly: a well-funded trader can move a thin contract with no change in voter sentiment at all.

The loop closes: the many pay, the few price, the press repeats the price, voters adjust to it, and the adjustment becomes a vote. Galtung's formula gets a new engine here. The gap between potential and actual isn't enforced by a gun. It's enforced by a number people are persuaded to trust. Cultural and structural violence interlock, and the meta layer holds as always: whoever owns the interpretation owns what counts as fact.

The corruption Levine documents is not subtle. George Santos bought Kalshi contracts on whether he'd attend the State of the Union, posted that he would, stayed home, and made $18,000; his lawyer insisted there was "absolutely no intent to deceive any person." Aides bet on their own candidates this spring. Polymarket has investigated dozens of potential fraud cases around the midterms. The regulator, Michael Selig, was a corporate lawyer for prediction markets and crypto before taking the chair. Donald Trump Jr.'s 1789 Capital put half a billion into Polymarket; Kalshi announced it was "beyond excited" to hire him as a paid adviser the day after the inauguration.

Levine says ban it. Fine. But not because gambling is inherently poisonous — she's honest enough to say a casino on your phone doesn't turn everyone into a fiend, and if you want to bet on cricket or aliens, go ahead. Democracy isn't cricket. Politics is already a game badly corrupted by money; prediction markets don't corrupt it further so much as bolt a slot machine onto it and rebrand the payout as public opinion.

One more number worth staring at: midterm trading volume is $220 million by the Anti-Corruption Data Collective's count and $1.4 billion according to the two platforms' own spokespeople. A sixfold gap, for firms valued at $22 billion and $9 billion. A business that can't agree on its own volume is currently pricing democracy.

What's really being sold isn't votes. It's the question of who gets to speak. It used to be pollsters, cable pundits, party machines. Now it's a set of accounts whose names never appear in the story. They don't need to persuade you. They just need you to see a number and believe you thought of it yourself.