Data point in favor of this article's thesis: The markets became significantly more optimistic about the Democrats' chances in the midterms immediately after Nate Silver first released his model.
Maybe outcome wgering in prediction markets coud even dicourage horserace polling entirely pushing it toward issue polling: what do people think about capital gains taxation, deficit reduction, assistance to Ukraine, trade wars, etc.
Event betting platforms can provide data points, but as Matt ultimately concludes there are limits to what they can provide, and I don't think I had my prior on that shifted considerably with a whole article. I much prefer a robust model like Nate Silver's that's starting with the polls and then adding fundamentals and other factors that tend to correlate. And Nate, despite his gambling reputation, doesn't use event betting data in his model, because he's fearful of recursion problems if the gamblers make bets based on his model, which I'm sure some do.
I think it's better to read event gambling more as a confirmation of trends than a revelation. It's paywalled, but this latest Nate article [https://www.natesilver.net/p/expert-ratings-are-ignoring-signs] hardly mentions event gambling more than just a confirmation, and instead argues that the expert ratings are being too *bearish* for Democrats.
".. this appreciation [of predictin markets] exists awkwardly alongside my concerns about the social impact of widespread legalized sports gambling."
Except in edge cases where it is not clear what is what, there is nothing "awkward" about his at all. Prediction markets are information aggregators about real world events. Sports betting and other kinds of gambling are not about real events and really should be heavily (progressively with the aounts wagered) taxed to discourage it.
Sports are very much real events. Someone has to orchestrate them, of course, but that's true for elections as well. I feel like what you're looking for are events completely orchestrated by casinos.
'There’s enough publicly available information and skilled traders to make election markets well calibrated. And theoretically the markets could generate even more useful information by spurring people to invest in high-quality polling. After all, one huge problem with public polling is that if you tell a media outlet you could make their horse-race poll slightly more accurate by tripling the cost, it’s not really clear why you’d spend the money on that. A gambler seeking a private edge might, and his bets could convey new information to the public.'
I'm sorry but I don't understand the logic of this paragraph. Why would a gambler 'seeking a private edge' lead to better public polling, aka 'convey[ing] new information to the public'?
The actual problem that prediction markets have when it comes to elections is that Nate Silver has made a living out of building an extraordinarily successful model which is available at trivial expense to the general public. Consequently there is very little 'private' information which can provide an edge; even most people working on campaigns will have no more reliable data on their chances than Silver's models provide. So it's really unclear to me what prediction markets add in this field.
No, the idea is that (in the world where election markets were efficient) the gambler's activities would convey new information to the public *by moving the prediction market prices*, even if the public couldn't see the polling data that caused them to do that.
"The main exception, predictable from our general knowledge of human psychology, is a small bias toward long shots."
I wouldn't assume this. The observed bias could easily be a result of the way these markets price contracts and fees, which distort the risks and profits for bets at different probabilities.
Agreed, like given the time value of money there’s no point in betting 98% on a sure thing that pays out in a year for instance.
If anything human psychology seems biased *against* the long shots (like in the Hong race, which I did think markets were way too low on the chance of an upset, and the others mentioned in the article).
>>>... or are the polls just massively oversampling Democrats?
Yes - the polls are massively oversampling Democrats, like they always have in the Trump Era.
I looked at the crosstabs of of the seemingly good national generic ballot polls last week, and their sample was even in terms of percent registered Dems and registered Republicans. But the last midterm electorate in 2022 was ~R+5 by raw voter registration, and Republicans have dominated voter registration in the last four years. So if all pollsters are doing this, they are systematically undercounting the Republican share of the electorate ... and perhaps they are doing this because the lowest-trust, hardest-to-poll voters are overwhelmingly pro-Trump. And to the extent that prediction markets are following polls, they are ALSO massively overstating Dems' chances.
Fundamentally, Trump carried ~240 House seats on the new maps, and won the marginal Senate seat by double digits. I don't know how you look at the issue trust and net favorability data on Dems and conclude that THIS party is going to carry ~25 Trump-won House seats, to say nothing of Trump +15 Senate seats. The electorate in the marginal seats has broad and deep concerns about voting for Dems, and we have done far too little to assuage them.
I strongly oppose cuts to SNAP and Medicaid, so it makes me really sad that Republicans are going to win - but that's what's coming.
Why do you care about SNAP and Medicaid? There's more important considerations.
I'm looking at the most relevant number of all: the price of gasoline.
Edit: I'm still not convinced you aren't some weird psy-op; your participation in the comments is largely confined to "the Dems is gonna lose" but since that seems like a really stupid psy-op to run, I'm agog.
Hold on, if this was true, the Republicans would've massively outperformed their polling in 2022, and that just didn't happen. We should expect the out party to dominate new registrations, due to thermostatic public reaction. Polling has done relatively well in the last couple midterm cycles, compared to the Presidential cycles. Polls have repeatedly found that Democrats are far more motivated to vote this year, compared to Republicans, hence the sampling, and the portion of voters that don't answer polls and do show up when Trump is on the ballot don't have Trump on the ballot.
Maybe the polls are off, but Republicans picking up 20ish seats suggests that polls are so off, that you can make a ton of money off this, or you're missing something. I think that you're wildly overweighting new registrations.
It is fascinating to observe that predictions in a field are unrelated to knowledge of a field, forecasting is a completely separate skill where independence from the field is useful. It should completely change the internal functioning of government and other institutions work, but it doesn't seem to have fed through.
“predictions in a field are unrelated to knowledge of a field, forecasting is a completely separate skill where independence from the field is useful”
It’s worthwhile to separate out two different kinds of great forecasters:
- quant-style forecasters, who get “alpha” from meticulously tracking down research information nobody else had access to (eg using satellite data to track the distribution of cows in an area, to predict the fair value of milk on futures markets)
- Tetlock-style superforecasters, who basically do a more advanced version of the outside view. Pick the most “natural” reference class available for the event in question, then anchor on the historical base rate for that event, with maybe a few minor adjustments on the side.
These are quite different and require vastly differing levels of subject matter expertise. They also plug each other’s gaps in many domains where either of them alone fails.
Of course, there can also be domains (like AI) where superforecasters constantly get their asses kicked and underestimate the pace of progress time after time after time, but that’s precisely because they lack the “inside view” understanding of how the tech is sui generis.
"...predictions in a field are unrelated to knowledge of a field"
I'm not sure that's true. We know that specialist forecasters beat domain experts with no particular skill or training in forecasting. But I think for any given level of forecasting ability, more knowledge of the field will always be helpful.
But I agree that forecasting ability is massively undervalued in government.
> What method did the Trump whale use to predict US polls?
> Theo used a polling approach he called the "neighbour effect". Instead of asking pollsters whom they intended to vote for, he asked them whom they believed their neighbour would vote for.
> The reason for this approach was that most people may feel reluctant to reveal their own political leanings but are more open to guessing the political preferences of those around them. This approach also relieved respondents from the pressure of sharing their own views and could be seen as a light-hearted exercise.
Hire a bunch of opinion pollsters with a better method of polling to do private surveys for you -> bet on results -> profit.
Data point in favor of this article's thesis: The markets became significantly more optimistic about the Democrats' chances in the midterms immediately after Nate Silver first released his model.
You’re allowed to just wait and see what happens.
Maybe outcome wgering in prediction markets coud even dicourage horserace polling entirely pushing it toward issue polling: what do people think about capital gains taxation, deficit reduction, assistance to Ukraine, trade wars, etc.
A shame for us here that few people care about these things to that kind of detail.
I found this article to be quite weird.
Event betting platforms can provide data points, but as Matt ultimately concludes there are limits to what they can provide, and I don't think I had my prior on that shifted considerably with a whole article. I much prefer a robust model like Nate Silver's that's starting with the polls and then adding fundamentals and other factors that tend to correlate. And Nate, despite his gambling reputation, doesn't use event betting data in his model, because he's fearful of recursion problems if the gamblers make bets based on his model, which I'm sure some do.
I think it's better to read event gambling more as a confirmation of trends than a revelation. It's paywalled, but this latest Nate article [https://www.natesilver.net/p/expert-ratings-are-ignoring-signs] hardly mentions event gambling more than just a confirmation, and instead argues that the expert ratings are being too *bearish* for Democrats.
If this ditch the Trump wild farragoes have driven us into isn't a setup for a Blue Wave, then I don't know what would be.
".. this appreciation [of predictin markets] exists awkwardly alongside my concerns about the social impact of widespread legalized sports gambling."
Except in edge cases where it is not clear what is what, there is nothing "awkward" about his at all. Prediction markets are information aggregators about real world events. Sports betting and other kinds of gambling are not about real events and really should be heavily (progressively with the aounts wagered) taxed to discourage it.
Sports are very much real events. Someone has to orchestrate them, of course, but that's true for elections as well. I feel like what you're looking for are events completely orchestrated by casinos.
'There’s enough publicly available information and skilled traders to make election markets well calibrated. And theoretically the markets could generate even more useful information by spurring people to invest in high-quality polling. After all, one huge problem with public polling is that if you tell a media outlet you could make their horse-race poll slightly more accurate by tripling the cost, it’s not really clear why you’d spend the money on that. A gambler seeking a private edge might, and his bets could convey new information to the public.'
I'm sorry but I don't understand the logic of this paragraph. Why would a gambler 'seeking a private edge' lead to better public polling, aka 'convey[ing] new information to the public'?
The actual problem that prediction markets have when it comes to elections is that Nate Silver has made a living out of building an extraordinarily successful model which is available at trivial expense to the general public. Consequently there is very little 'private' information which can provide an edge; even most people working on campaigns will have no more reliable data on their chances than Silver's models provide. So it's really unclear to me what prediction markets add in this field.
No, the idea is that (in the world where election markets were efficient) the gambler's activities would convey new information to the public *by moving the prediction market prices*, even if the public couldn't see the polling data that caused them to do that.
"The main exception, predictable from our general knowledge of human psychology, is a small bias toward long shots."
I wouldn't assume this. The observed bias could easily be a result of the way these markets price contracts and fees, which distort the risks and profits for bets at different probabilities.
Agreed, like given the time value of money there’s no point in betting 98% on a sure thing that pays out in a year for instance.
If anything human psychology seems biased *against* the long shots (like in the Hong race, which I did think markets were way too low on the chance of an upset, and the others mentioned in the article).
>>>... or are the polls just massively oversampling Democrats?
Yes - the polls are massively oversampling Democrats, like they always have in the Trump Era.
I looked at the crosstabs of of the seemingly good national generic ballot polls last week, and their sample was even in terms of percent registered Dems and registered Republicans. But the last midterm electorate in 2022 was ~R+5 by raw voter registration, and Republicans have dominated voter registration in the last four years. So if all pollsters are doing this, they are systematically undercounting the Republican share of the electorate ... and perhaps they are doing this because the lowest-trust, hardest-to-poll voters are overwhelmingly pro-Trump. And to the extent that prediction markets are following polls, they are ALSO massively overstating Dems' chances.
Fundamentally, Trump carried ~240 House seats on the new maps, and won the marginal Senate seat by double digits. I don't know how you look at the issue trust and net favorability data on Dems and conclude that THIS party is going to carry ~25 Trump-won House seats, to say nothing of Trump +15 Senate seats. The electorate in the marginal seats has broad and deep concerns about voting for Dems, and we have done far too little to assuage them.
I strongly oppose cuts to SNAP and Medicaid, so it makes me really sad that Republicans are going to win - but that's what's coming.
Why do you care about SNAP and Medicaid? There's more important considerations.
I'm looking at the most relevant number of all: the price of gasoline.
Edit: I'm still not convinced you aren't some weird psy-op; your participation in the comments is largely confined to "the Dems is gonna lose" but since that seems like a really stupid psy-op to run, I'm agog.
Hold on, if this was true, the Republicans would've massively outperformed their polling in 2022, and that just didn't happen. We should expect the out party to dominate new registrations, due to thermostatic public reaction. Polling has done relatively well in the last couple midterm cycles, compared to the Presidential cycles. Polls have repeatedly found that Democrats are far more motivated to vote this year, compared to Republicans, hence the sampling, and the portion of voters that don't answer polls and do show up when Trump is on the ballot don't have Trump on the ballot.
Maybe the polls are off, but Republicans picking up 20ish seats suggests that polls are so off, that you can make a ton of money off this, or you're missing something. I think that you're wildly overweighting new registrations.
It is fascinating to observe that predictions in a field are unrelated to knowledge of a field, forecasting is a completely separate skill where independence from the field is useful. It should completely change the internal functioning of government and other institutions work, but it doesn't seem to have fed through.
“predictions in a field are unrelated to knowledge of a field, forecasting is a completely separate skill where independence from the field is useful”
It’s worthwhile to separate out two different kinds of great forecasters:
- quant-style forecasters, who get “alpha” from meticulously tracking down research information nobody else had access to (eg using satellite data to track the distribution of cows in an area, to predict the fair value of milk on futures markets)
- Tetlock-style superforecasters, who basically do a more advanced version of the outside view. Pick the most “natural” reference class available for the event in question, then anchor on the historical base rate for that event, with maybe a few minor adjustments on the side.
These are quite different and require vastly differing levels of subject matter expertise. They also plug each other’s gaps in many domains where either of them alone fails.
Of course, there can also be domains (like AI) where superforecasters constantly get their asses kicked and underestimate the pace of progress time after time after time, but that’s precisely because they lack the “inside view” understanding of how the tech is sui generis.
"...predictions in a field are unrelated to knowledge of a field"
I'm not sure that's true. We know that specialist forecasters beat domain experts with no particular skill or training in forecasting. But I think for any given level of forecasting ability, more knowledge of the field will always be helpful.
But I agree that forecasting ability is massively undervalued in government.
It is an empirical question where I don't have the data, my suspicion is that more knowledge makes it harder to take an outside view.
And when the market *is* thick, we can and do see people funding private polls. The 2024 presidential election featured a European "Trump whale" who bet millions on Trump after his own private polling showed Trump with a surprisingly high lead: https://www.business-standard.com/world-news/how-a-french-trader-predicted-trump-s-victory-by-asking-about-the-neighbour-124110701151_1.html
> What method did the Trump whale use to predict US polls?
> Theo used a polling approach he called the "neighbour effect". Instead of asking pollsters whom they intended to vote for, he asked them whom they believed their neighbour would vote for.
> The reason for this approach was that most people may feel reluctant to reveal their own political leanings but are more open to guessing the political preferences of those around them. This approach also relieved respondents from the pressure of sharing their own views and could be seen as a light-hearted exercise.
Hire a bunch of opinion pollsters with a better method of polling to do private surveys for you -> bet on results -> profit.
But it is hard to find out what is the best polling method beforehand.