Frequently people on this board will cite data from online betting markets as indicative of how the primary race is going, in the same way that one would cite polls.
Nate Silver says that these markets are worthless and don’t provide useful information; he doesn’t use them at all in his forecasting model. Based on some of the cites I’ve seen, that seems spot on; I recall seeing that these markets gave Bernie Sanders a ridiculous 70% chance of winning the Iowa caucuses, and prior to last night’s debate had Bloomberg at an absurdly high 30% to win the nomination. It seems logically that these markets shouldn’t be very reflective of reality, since they’re basically like polls that only reach people who spend a lot of time online.
So is there any scientific argument to be made that these markets are worth paying attention to? Is there evidence that the markets have consistently been more accurate predictors than polls?
Can I get a cite for this? This seems like an incredible thing for him to say. I did a quick search on election markets and Nate Silver and got this article he wrote in 2012.
At the time, 538 was leaning more toward Romney than the betting markets. Silver surmised that the betting markets might be more predictive because they can take his models into account before betting. Silver says that election markets are tough to beat.
Polling models are based on old data. The polls have to be taken earlier. But they’re more systematic and more comprehensive. Betting markets have a lot of noise, but they’re in real time. Some of the noise includes people betting long or short, big players that skew the numbers, the herd mentality, etc.
They don’t necessarily serve the same function. For instance, on the night of the debate, the betting markets had an immediate reaction to the Bloomberg debate. Nate Silver refused [timestamped] to give a forecast on what would happen to Bloomberg’s polling despite his insistence that Bloomberg had a bad debate–“I thought he was pretty terrible… . . He sounds like he hasn’t talked to a normal American in 10 years.” Despite that, he wouldn’t give much of a prediction on how the voting public would react. He gave a range from the debate doesn’t affect Bloomberg at all to it was campaign ending for him. But he wouldn’t give odds on which was more likely.
The betting markets are forecasting in real time. Nate Silver doesn’t do that. If he was a prognosticator, he’s not very good at it. In this video in Jan. 2019, he’s asked to pick who the Dem nominee will be. He picked Kamala Harris. In Oct. 2019, he then picked Bernie Sanders but he wasn’t too enthused about him.
Well, you’ve found a more comprehensive accounting of his thinking. I was thinking of a comment he made in a recent chat that “betting markets are stupid”. My general impression is that his opinion has evolved more in that direction since 2012; IIRC, most of the comments I’ve seen about markets on 538 recently are of a similar tone.
But yes, polls are more systematic and more comprehensive. It takes about a week to collect and analyze the data. So unless there’s less than a week until the election, why wouldn’t you just wait for the polls? What use is it to have “real time” information of dubious accuracy?
I’m still interested in seeing a source for this. I searched for your quote with Nate Silver and didn’t find anything. I did find this video where he validates the betting markets’ take on Elizabeth Warren back in June 2019. He wasn’t denigrating them. He was validating them. At the time, he said that if he were to put money down, he’d put it on Elizabeth Warren over Bernie Sanders, which is what the betting markets were showing.
It’s another source of data. I haven’t seen anyone else say they were of dubious accuracy, including Nate Silver. All forecasts have some bias in them that doesn’t make them 100% accurate. That report in my last post shows that the accuracy of select betting markets, filtering for bias was more accurate than Silvers’ model. They were quicker to predict Obama’s win than Silvers at the time.
That may be an offhand comment not fully reflective of this thoughts on the subject. But I’m pretty sure Nate would argue that his model is significantly better now than it was in 2008.
Thanks for the link. In context, the rest of the staff were discussing Bloomberg’s debate performance and what it meant. No one seemed to know what to make of it in terms of how voters would react. Nate obviously went looking to see what other people were thinking and some evidence of that is in the betting markets. Nate seemed to want to make sure that everyone knew this wasn’t some kind of definitive statement. But he must have thought that it wasn’t completely useless or he wouldn’t have taken the time to look it up.
As to whether his model is better, we won’t know until the polls come out and he puts them in his model and then wait for another election. Even he wanted to get a better idea of the mood of the electorate before then.
Another question might be: Why are people enamored of pollings? In late July 2011, the Zogby polled nationwide and found that Michelle Bachmann was first among Republican voters, with Herman Cain 2nd! A month earlier, Sarah Palin was first in a Reuters poll. As late as October, IBOPE Zogby had Cain polling at 45%, way ahead of Romney at 21%.
And of course prediction markets may take into consideration future possibilities (e.g. health relapse) unreflected in pollee opinions. Also, isn’t the idolization of Nate Silver rather exaggerated? He’s good at statistical modeling, but so are others. If he deserved the pedestal Dopers place him on, I think he’d be working for Renaissance Tech!
Anyway, OP, in light of post #2, do you wish to revise parts of your #1?
Polls have their flaws but they are directly measuring what people are thinking. Betting markets measure what some people think other people are thinking – based on polls and their gut. Why wouldn’t I trust polls more?
Why do people pay attention to betting markets?
Because that’s where the money is.
My uneducated guess would be a betting market “stock” is about as predictive as a poll with the same number of participants. Because a market is a type of poll, after all. Of course, questionnaire polls aren’t that predictive in general–only carefully controlled scientific questionnaires have any utility.
Markets are self-selecting polls, which degrades their usefulness, but participants are risking real money to try to be accurate. I’m not sure if where that puts them for predictiveness, but if Nate Silver doesn’t throw them into his mix, I’m guessing low utility.
You can’t weigh betting markets against polls. The markets themselves reflect the polls and are not independent sources of information.
But polls are not be-all-and-end-all. Even a poll-driven guy like Nate Silver needs to make adjustments to accommodate for things like weighing the disparate results of different polls, the “house effect” of certain polls, the reliability of individual pollsters, and so on. In addition, there are other factors like national polls vs state polls. Further, polls reflect the current state of the race and are also impacted by name recognition in the early stages, and it would be a mistake to rely on them too much in the face of other factors which suggest future movement in the polls. Which is why even modelers like Silver incorporate things like fundraising and endorsements in their modeling.
Bottom line is that the prediction of elections is much much more complex than simply looking at a poll. So there’s a lot of room for someone to come in and boil it all down to a number - “as a result of all these various complex factors, we think it all adds up to an X% likelihood of such-and-such happening”. Guys like Silver et al are doing that. Betting markets are even more meta than that, as they incorporate the market’s assessment of the value of Silver’s modeling (and those of others).
The betting markets are a lot like the stock market in this regard. It goes without saying that the betting markets can be wrong, just like it also goes without saying that the stock market could be pricing such-and-such stock wrong. What both the betting and stock markets are is the collective wisdom of people with money on the line as to the likelihood of various events, based on all publically available information. That collective wisdom could well be wrong, but it’s absolutely worth paying attention to.
Those who think the betting markets are poor predictors have the opportunity to “put their money where their mouth is,” by locating mispricings and making favorable bets.
I don’t blame you for not indulging in such speculations — though you’re passing up easy money if you really think the markets are way off! — but many others will bet; this tends to move the prices toward rational estimates.
Why do you have a problem with those poll results? There’s no reason to think they weren’t accurate at the time, and nobody with any sense would have looked at any of those polls and thought the current leader was a lock to win, given how obviously volatile the race was. Betting markets offer a picture of how some people think the race will unfold in the future, but it’s a self-selected group of people who aren’t representative of the general population and don’t have any special expertise or information. If you have evidence that more than a small portion of the people participating in these markets are serious students of political science, that would change my opinion.
Yes, you are correct that Silver uses many different factors, not just raw polling data, in making his models. And as Pleonast pointed out, I’m sure he’d be using betting market data if he felt it would add any value, but he doesn’t.
If he would use it, it would be somewhat circular, since the betting markets themselves reflect his modeling (and that of others).
More generally, the whole point of Silver’s modeling is to provide an independent data-driven analysis. The betting markets are along the lines of “expert opinions”, which Silver does not include. Incorporating betting markets into his modeling would be analogous to including a survey of pundits. That’s not what he’s trying to do. This isn’t to imply that pundit opinion has no value; only that that’s not what Silver is doing.
But polling itself is circular; voters’ decisions on who to support depend enormously on who the polls tell them has a chance of winning. ITSM that market prices are much more akin to objective numerical data like polls and demographic profiles than they are to punditry, so I don’t think he would have a philosophical objection to including them.
That’s not circular, assuming they intend to vote for who they say. If you’re saying that people are lying to pollsters, that’s something else, but if they’re not then it makes no difference what influences their decision.
OK. But both market prices and polls take the opinions of a group of people and translate them into a numerical value. Unlike polls, markets are assessing predicted outcomes rather than preferred outcomes. But like polls and unlike punditry, markets tell us nothing about the reasons people hold the opinions they do. They might influence strategic decisions, but they’re not likely (or intended) to change anyone’s opinion about who should win.
Nate talks about the value of betting on one’s beliefs in his book “The Signal and the Noise” - he may have serious doubts about the “instant response” of a betting market over the course of a debate, but he probably doesn’t dismiss them entirely.
If you think Exxon stock is undervalued, you buy it but are faced with a danger: Maybe you’re correct that Exxon is undervalued but a year later the market still hasn’t figured that out; your money’s been tied up for a year to no avail.
Bets on the Democratic nomination do not have that problem. This summer all the contracts will be worth either $100 or Zero.
If you think Betfair overvalues Bloomberg … Bet against him! If you think Betfair undervalues Biden … Bet on him! If you think Betfair misevaluates these odds but would rather talk about it on a message board than lay money on the line, then … ?
And — key point — whatever your reason for not placing a bet, others are not so constrained. Good opportunities would attract smart money. Google “Efficient Market Hypothesis.”
I didn’t have a “problem.” YOU wrote that polling results were more informative of chances than betting markets; I presented data to disprove that point; now you’re AGREEING with the thrust of MY argument: polls, by themselves, are not predictors. If you’ve come around 180 degrees and now agree with me, say so!!
You claim that Nate Silver, the anointed “Babe Ruth of Forecasting”, doesn’t like prediction markets and get 100% refuted by post #2.
I don’t really know whether Silver’s numbers are more or less accurate than Betfair’s; one reason I quote the latter is they are more accessible to me. (Where are Silver’s numbers? In another thread someone referenced Silver’s simulations of convention scenarios, but didn’t post a link: What is that URL?)
Why do you have a problem with those poll results? There’s no reason to think they weren’t accurate at the time, and nobody with any sense would have looked at any of those polls and thought the current leader was a lock to win, given how obviously volatile the race was. Betting markets offer a picture of how some people think the race will unfold in the future, but it’s a self-selected group of people who aren’t representative of the general population and don’t have any special expertise or information. If you have evidence that more than a small portion of the people participating in these markets are serious students of political science, that would change my opinion.