预测市场对中期选举获胜者的预测有多准确?进入选举日之际,共和党人似乎很可能保住南卡罗来纳州的参议院席位。参议员达琳·格雷厄姆正在竞选连任,以接替她的哥哥、已故参议员林赛·格雷厄姆,完成一个完整任期,并且她已获得特朗普总统的背书。
How Accurate Were Prediction Markets at Picking Midterm Winners? Heading into Election Day, Republicans appeared likely to hold on to South Carolina’s Senate seat. Senator Darline Graham was running to succeed her older brother, the late Senator Lindsey Graham, for a full term, and she had received President Trump’s endorsement.
Polymarket和Kalshi等预测市场往往也持相同看法。11月3日上午,在线投注者认为格雷厄姆女士赢得这场竞选的概率约为70%,这背后有超过2700万美元的投注活动推动。
Prediction markets, like Polymarket and Kalshi, tended to agree. On the morning of Nov. 3, online bettors had given Ms. Graham a roughly 70 percent chance of winning the race, fueled by more than $27 million in betting activity.
随着周二晚间计票结果陆续出炉,她在Polymarket及其他平台上的胜算迅速反转。民主党挑战者安妮·安德鲁斯乘着全国性民主党支持浪潮上台,截至周四,她以约9万票、即4.4个百分点的优势获胜。
As results began to come in on Tuesday evening, the odds of her winning on Polymarket and elsewhere rapidly inverted. The Democratic challenger, Annie Andrews, was swept into power as part of a national wave of Democratic support, winning by about 90,000 votes, or 4.4 percentage points, as of Thursday.
中期选举之后,预测市场能否准确预见选举结果已成为一个关键问题。《纽约时报》运用统计方法进行的一项分析发现,在中期选举的大部分时间里,预测市场的表现并不比选举预测更好——后者利用民调和其它指标而非投注活动来生成概率。但随着选举日临近,由于预测和民调变得过时,预测市场的表现反而更好。
Whether prediction markets can accurately anticipate election results has become a key question after the midterms. An analysis by The New York Times, using statistical methods, found that for most of the midterm race, prediction markets performed no better than election forecasts, which produce odds using polling and other indicators instead of betting activity. But as Election Day neared, prediction markets performed better as forecasts and polling became stale.
这些平台已从一种鲜为人知的在线博彩形式——允许人们对从体育到流行文化几乎任何事情下注——转变为价值数十亿美元、吸引全球关注的现象。其运作方式是允许任何人对几乎任何事件的结果下注,用户投入的资金会转化为某件事发生的概率。
The platforms have gone from an obscure form of online betting — allowing people to place wagers on nearly anything, from sports to pop culture — to an international fixation worth billions of dollars. They work by allowing anyone to bet on the outcome of pretty much any event, with the money that users spend translating into odds that something will happen.
Polymarket 和 Kalshi 在中期选举季大肆吹嘘其优于民调和选举预测,声称当参与人数足够多时,押注者能够正确预测未来 —— 即群众的智慧。它们表示,自己比民调更具反应速度,因为民调需要在数天内询问选民,并利用这些信息估算每位候选人将获得的票数。
Polymarket and Kalshi spent the midterm season boasting about their superiority to polls and election forecasts, claiming that bettors can correctly predict the future when working in large enough numbers — the wisdom of the crowd. They have said they are more reactive than polls, which query voters over several days and use that information to estimate how many votes each candidate will receive.
它们还声称胜过选举预测者,后者通过将民调数据与经济指标、历史趋势、专家意见及其他指标进行复杂融合,生成显示候选人获胜概率的赔率。
They have also claimed to beat election forecasters, which produce odds showing whether a candidate will win by using a complex blend of polling data with economic metrics, historical trends, expert opinion and other indicators.
Kalshi 和 Polymarket 现在试图利用选举日的结果,证明它们作为可靠预测工具具有公共价值。
Kalshi and Polymarket are now seeking to use the results from Election Day to prove that they offer public value as a reliable predictor.
“Polymarket 是对的,”该公司周三在社交媒体上得意洋洋地宣称。
“Polymarket was right,” the company crowed on social media on Wednesday.
“过去几个月,人们来到 Kalshi 向世界表明他们将做什么,”Kalshi 首席执行官 Tarek Mansour 周三发帖称。“昨晚,扩音器交到了他们手中。” 然而,《纽约时报》发现,在中期选举前最后三个月里,预测市场生成的赔率并不比 Nate Silver 等数据专家制作的知名选举预测更准确,Silver 创办了 Silver Bulletin,这是一份选举预测和通讯。
“Over the last few months, the people showed up on Kalshi to tell the world what they were going to do,” Tarek Mansour, the chief executive of Kalshi, posted on Wednesday. “Last night, the megaphone was passed to them.”The Times found, however, that the odds produced by prediction markets over the last three months before the midterms were no more accurate than those from prominent election forecasts produced by data wonks like Nate Silver, who created Silver Bulletin, an election forecast and newsletter.
虽然预测市场的赔率在选举日前几天变得更准确,但早期的押注活动生成的赔率对胜者的把握程度不如一些预测,且经常准确度更低。
While the odds on prediction markets became more accurate in the days before Election Day, earlier betting activity produced odds that were less confident about the winner than some forecasts and were frequently less accurate.
在一份声明中,Kalshi 表示,即使在选举日前几个月,其生成的赔率也能与复杂的选举预测相抗衡,这证明了它们的预测能力。它们还指出,随着选举临近,赔率有所改善。预测者表示,这部分原因在于选举预测高度依赖民调,而民调到了选举日已变得陈旧。
In a statement, Kalshi said producing odds that were competitive with sophisticated election forecasts — even months before Election Day — was proof of their predictive power. They also pointed to odds that improved as the election neared. That is partly because election forecasts are highly reliant on polling, which becomes stale by Election Day, forecasters said.
“简单来说,我们的主张就是:我们平台上的选举预测市场校准良好,即便参与度较低,在大选前三个月内依然如此,”Kalshi Research(Kalshi研究部,该预测市场的分析部门)负责人妮可·卡根(Nicole Kagan)表示。Polymarket未回应置评请求。
“The claim that we’re making, very simply put, is just that the election markets we have on our platform are well calibrated, up to say three months out, even with low participation,” said Nicole Kagan, the head of Kalshi Research, the prediction market’s analysis division. Polymarket did not respond to a request for comment.
《纽约时报》采用多种方法评估预测市场的表现。主要方法称为“布里尔评分”(Brier score),它在衡量预测准确性的同时,也会考量预测的置信度。设想一个预测市场给某候选人90%的获胜几率。若该候选人落败,预测犯下重大错误,最终评分应反映失误的程度。但若候选人获胜几率仅51%,错误则更可原谅。布里尔评分会综合考虑这些因素。
The Times used several approaches to assess how the prediction markets performed. The main method, called the Brier score, measures a forecast’s accuracy while also considering its confidence. Imagine that a prediction market gave one candidate a 90 percent chance of winning. If that candidate lost, the forecast made a major error and its final score should reflect the size of that miss. But if the candidate had only a 51 percent chance of winning, the error is more forgivable. Brier scores consider those factors.
这张图表展示了预测市场与选举预测在数月内的准确性排名,从优到劣。总体而言,Polymarket和Kalshi在某些时刻领先于选举预测,但大选大部分时间里落后于预测者。
This chart shows the ranking of prediction markets and election forecasts, from best to worst, over several months. Overall, Polymarket and Kalshi led election forecasts at certain moments but spent most of the election trailing forecasters.
预测准确性排名 在8月至11月的多个时间点,预测市场的准确性低于《纽约时报》评分的两份选举预测:“Silver Bulletin”(西尔弗通讯)和“FiftyPlusOne”。这些选举预测倾向于给出更可靠的赔率,更能反映竞选实况,即使在结果揭晓数月前也是如此。当它们看错候选人时,给出的赔率通常低于预测市场,为冷门留出可能;当它们选对候选人时,通常更有信心。
Predictive accuracy rank At multiple points between August and November, prediction markets were less accurate than two election forecasts scored by The Times: Silver Bulletin and FiftyPlusOne. Those election forecasts tended to produce more reliable odds that better reflected the realities in the race, even months before results were known. When they favored the wrong candidate, they tended to assign lower odds than prediction markets, leaving open the possibility for an upset.
在缅因州参议院竞选中,预测者给民主党人特洛伊·杰克逊(Troy Jackson)击败现任共和党参议员苏珊·柯林斯(Susan Collins)的机会评价更高。这一点在大选前30天和11月3日均成立:当时预测市场对杰克逊获胜的信心增强,但预测者的信心增强得更多。中位数预测市场赔率与预测模型赔率与最终结果的对比。
When they chose the correct candidate, they were generally more confident. In Maine’s Senate race, forecasters gave Troy Jackson, the Democrat, a better shot of defeating Susan Collins, the incumbent Republican, than prediction markets. That was true 30 days before the election and on Nov. 3, when prediction markets grew more confident about Mr. Jackson’s odds but forecasters grew even more confident. The median prediction market and forecast odds compared with the final result.
缅因州参议员 赢面 预测市场 预测模型 结果 Jackson Troy D. Jackson 以 6.6 个百分点获胜,得票率 53.3% 对 46.7% 尽管如此,与预测模型相比,预测市场在准确性排名中总体处于中游位置。这或许是不可避免的:大量投注者将所有可用信息来源(包括选举预测)综合考量后做出决策,形成了一种平均信念。
Maine Senate Chance of winning Prediction markets Forecasts Result Jackson Troy D. Jackson won by 6.6 points, 53.3% to 46.7% Still, prediction markets generally fell in the middle of the accuracy rankings when compared with forecasts. That may be inevitable: Large groups of bettors combine information from all available sources, including election forecasts, into their decisions, creating what could amount to an average belief.
“市场处于所有其他信号的下游,”在线博彩与预测市场公司 High Roller Technologies 首席商务官 Flip Pidot 表示,“所有民调、所有预测者、所有评论、所有早期投票数据、所有实时流出的数据——这些全都汇入了市场。” 这种动态在选举夜显现出来,当时预测市场实时对结果做出反应,在美联社等机构宣布胜选前很久就给出了高度自信的预测。投票刚一结束,随着早期迹象指向民主党支持率的海啸式上涨,预测市场的赔率便开始变动。
“The markets are downstream of all those other signals,” said Flip Pidot, chief business officer at High Roller Technologies, an online gaming and prediction markets company. “All the polls, all the forecasters, all the commentary, all the early-vote data, all the real-time data coming out — that all gets fed into the market.”That dynamic appeared to play out on election night, when prediction markets reacted to results in real time, producing highly confident predictions far before races were called by organizations like The Associated Press. Almost as soon as polls closed, odds on prediction markets began shifting as early indications pointed to a tsunami of Democratic support.
Silver Bulletin 背后的预测者 Silver 表示,预测市场的准确性直接源于民调机构和预测者的工作,是他们创造了投注者主要依赖的信息。
Mr. Silver, the forecaster behind Silver Bulletin, said the accuracy of prediction markets was a direct result of work by pollsters and forecasters, who create the information that bettors have mostly relied on.
他在一份电邮声明中说:“在一个这些模型不存在、或者模型保持预测私密的世界里,我认为预测市场不会那么准确。” 他补充道:“市场在周二早上形成的价格,我们往往提前几天到几周就有了。”
“In a world where these models didn’t exist, or if the models kept their forecast private, I don’t think the prediction markets would be as accurate,” he said in an emailed statement. “The prices markets wound up with on Tuesday morning, we often had days to weeks sooner,” he added.
在爱荷华州的参议员竞选中,民调在民主党人 Josh Turek 和共和党人 Ashley Hinson 之间基本势均力敌。选前三十天,预测模型看好 Turek 先生,而预测市场则看好 Hinson 女士。图雷克最终以超过4个百分点的优势获胜。
In Iowa’s Senate race, polls were fairly evenly split between Josh Turek, the Democrat, and Ashley Hinson, the Republican. Thirty days before the election, forecasters gave the edge to Mr. Turek, while prediction markets favored Ms. Hinson. Mr. Turek ended up winning by more than four percentage points.
爱荷华州参议院 图雷克 乔什·图雷克以4.1个百分点获胜,得票率50.5%对46.4% 预测市场总体上可能略微偏向共和党,原因是人们错误地认为民调总是低估共和党的支持率。
Iowa Senate Turek Josh Turek won by 4.1 points, 50.5% to 46.4% Prediction markets may have slightly favored Republicans overall because of a false belief that polls always underestimate support for Republicans.
选举日后的共识是,今年的民调总体上准确。《纽约时报》/锡耶纳学院10月28日发布的一项民调令评论人士意外,该民调显示南卡罗来纳州的参议院选情已变成势均力敌。
The consensus after Election Day is that polls this year were, in general, on the mark. A New York Times/Siena poll, published on Oct. 28, surprised pundits by showing that the Senate race in South Carolina had become a toss-up.
“在近几个选举周期中,预测市场总体上比民调表现更好,因为它们正确预判了民调误差的方向,”研究预测市场的达特茅斯学院教授埃里克·齐策维茨说。“看来,在这次的情况下,预测市场认为民调可能略微高估了民主党的支持率——而看起来我们遇到的情况多少恰恰相反。”有一场选举似乎是民调搞错了,而预测市场也犯了同样的错误。民调曾预测共和党将赢得密西西比州的参议院选举。预测市场对此高度认同,在选举日早晨给共和党现任议员辛迪·海德-史密斯86%的胜选概率。
“In general, prediction markets have done better than the polls in recent cycles because they’ve correctly anticipated the sign of a polling error,” said Eric Zitzewitz, a Dartmouth professor who has studied prediction markets. “It looks like, in this case, prediction markets were thinking there might be a slight overstatement of Democratic support in the polls — and it looks like we sort of had the opposite.”There was one race that polls seemed to get wrong, and prediction markets made the same mistake. Polls predicted that Republicans would win Mississippi’s Senate race. Prediction markets strongly agreed, giving Cindy Hyde-Smith, the Republican incumbent, an 86 percent chance of winning on the morning of Election Day.
密西西比州参议院 科洛姆 斯科特·科洛姆以0.1个百分点领先,得票率49%对48.9% 已统计96% 民主党人斯科特·科洛姆与海德-史密斯女士几乎打成平手。如果他最终赢得这场选举,那将是所有人的失误:民调、预测——以及预测市场。
Mississippi Senate Colom Scott Colom leads by 0.1 points, 49% to 48.9% 96% counted Scott Colom, the Democrat, is practically tied with Ms. Hyde-Smith. If he ends up winning the race, it will be a miss by everyone: polls, forecasts — and prediction markets.