预测市场中期选举交易量达25亿美元准确率98%

Bloomberg Audio Studios podcasts, radio news.

You're listening to Bloomberg Business Week with Carol Masser and Tim Stenc on Bloomberg Radio. Well, President Trump may have misread voters. Prediction marks, however, did not. Cali is out with some research today that says the platform saw more than $2.5 billion dollars in trading volume for its midterm markets. midterms markets up from 1.3 billion for the 2024 election cycle. And then our own Joe Weisenthal, the co-host of Odd Lots, writing in his newsletter today that quote, "The only thing I want to say about the election is that I think the prediction markets were huge winners last night for two reasons.

The first is that unim ambiguously there was no better way to understand the state of the race in real time than following the prediction markets. Beyond that, he says the prediction markets move decisively towards the Dems in the final days of the race. Again, that's from our own Joe Weisenthal of OddLots.

Here with more and an analysis of the accuracy of the markets on Kali is Nicole Kagan, head of research at KHI. She joins us here in the Bloomberg Business Week studio. Welcome. Welcome. Uh, how accurate were the prediction markets?

Yeah, well, thank you so much for having me. Uh, well, the answer to that question is extremely accurate. So, of the 487 races that had been processed by 6:00 this morning, if we look at what the markets were saying before the closure of polls, Kelsey was directionally accurate in 479 of those 487. So, we're talking about accuracy in the realm of about 98 plus% across the elections we saw last night.

How did the accuracy change as you got closer to the election?

Yeah, so accuracy increases as a function of time to the election. you know, as more information comes out about the candidates, as more polling comes out, as more alternative data is revealed, traders may change their positions. And I think one thing that we saw last night that was really interesting and isn't included in this accuracy release, but is an interesting thing to look at is just how the closure of polls in certain states in Kentucky closing first influenced the way that traders were trading other markets.

So once we saw polls close, once we saw early voting results come in, other voting results come in and come in very blue, we saw traders reposition in other markets and begin to reposition in the House and the Senate races well.

That's called making a trade with more information. Right.

Exactly right. Exactly right. Yeah.

There were some markets that moved very um in a very big way,

very late. So, I'm thinking of of South Carolina, for example, or um

the Texas governor's race, for example.

Um what was unique about those races that caused such big movements to happen so late? And I don't recall if they were after the actual polls closed in in those ones, but it really flipped.

Well, they moved really quickly. The flip in South Carolina for Andrews happened within 5 minutes of polls closing in the state. So you actually saw the market response move decisively very very early right as the votes were beginning to be counted in the state of South Carolina. And so you saw traders reposition and ultimately correctly call that race well before networks were calling it. I think part of it is you know as you saw this early vote come in as and as you gained information from the way that other states had been voting it bled into a little bit the perception of what might happen there as well.

Yeah. Kind of an understanding of the environment right and how voters were doing it. So, how different were the bets? I don't know, do you go a month out, two months versus as you got really close to the elections?

Yeah. Well, so the trades that we saw on our platform in the terms of the accuracy that we were talking about earlier increased from about 95% to about 98% in the months leading up to the election to the date of the election itself to just before poll.

So 2 or 3 months ago, they were like 95% accurate.

Exactly. Right. And we know that when we're comparing to polling, you know, both averages that single out the best rated polling and also polling that it just takes all polling as a simple average. We were outperforming polling 3 months ahead of the election with the directional accuracy of the calls we were making.

But you're pulling on data from those polls, right? Like what what is

I mean I guess the the traders are pulling on are Yeah. They're the ones who are taking all the information in the available universe and and and making that a decision based on that. So essentially it should be an aggregate and correct me if I'm wrong. It should be an aggregate of all the information out there. Right.

That's exactly right. And I think one thing that we'll see in research that is to follow is you know which pieces of information did traders seem to pay the most attention to? So were there sort certain endorsements? Were there certain polls that came out that meaningfully move trader perception? And were there others that they ignored? And if they did ignore them, why did they seem to ignore them? So going into the midterms, there was there was some concern that there was a miss with Brazil's election recently or Kashi and the prediction other prediction markets didn't quite get the Israeli election right recently.

What what's the distinction between those? Was it a volume issue? Was it who was placing the trades and where they were based in the world? Like what was unique about those that that weren't necessarily so accurate, but these were?

Well, it's a good question and I think it fundamentally comes down to what the markets represent, you know? So earlier I spoke a little bit about accuracy which is probably one of the more naive ways of viewing our markets. It's the simplest way to understand them. If you round up from 50%, you know, do you end up with the correct candidate or not? But really what markets are providing to you is a probability of an event happening, right?

And so for the Brazilian election, what we saw was that there was an 80% chance of Flavia Bolsinaro winning that election according to the Kashi markets in advance of that election happening and a 20% chance for Lula. But what that would mean, you know, to a person living their daily life is like you might expect something to happen one of five days of the week that you go to work. So it's actually really not that unlikely a scenario that occurs.

And we know from the research that we've done in the past that if you take kind of over 2 million markets over the context of the entire history of the exchange and you look at, you know, if we say something happens 60% of the time, does it actually happen 60% of the time? The answer is yes. You know, we are very very good at forecasting in those probabilistic terms. And so taking just one example like what happened in Brazil is not quite the best way of representing what should really be viewed in a large sample context.

I am curious. We're talking with Nicole Kagan, head of research at Kowi. So Nicole, what are you and your team after a big event like this? What do you do and go back and kind of look at the variety of trades, the accuracy of those or those that were like what do you do after an event like this?

Well, lots of things. It's very exciting for us. I think um it's like a databased Christmas for us because we're looking through our data and we're thinking, you know, who is trading each of these races? Is it people that are within that state? Is it people outside of that state? When are they choosing to trade those races? So, are they coming in quite early? Well, we had called races quite early. So, for example, um markets had Terico favored from more or less uh you know, weeks ago, August onwards.

Um, you know, you might ask questions as to why markets were so confident in that outcome so early, what kind of inform