[jawns]

Prediction markets and bookmakers really strongly expected a Clinton win in 2016. Irish bookies were so sure Trump would lose that one firm began paying out to those who bet Clinton even before the election.

And we all know how that turned out.

Nov. 7, 2016:

"Hillary Clinton’s odds of winning the presidency rose from 78% last week to 91% Monday before Election Day, according to CNN’s Political Prediction Market."

https://www.cnn.com/2016/11/07/politics/political-prediction...

[johnfn]

You can go look at the old Predictit market if you like - it's still around[1]. You can also check election betting odds[2]. Clinton hovered around 60-80 in the week or two before the election.

Even if it were 90%, I'm not sure that's an indictment of prediction markets. A probability of 90% means that 10% of the time, the other thing happens. It's a bit like saying "wow, it's cold today; I guess global warming isn't real." You'd want to to an analysis across many many prediction markets in order to see if they're generally accurate. That already exists, it's here if you'd like to look at it[3].

[1]: https://www.predictit.org/markets/detail/1234/Who-will-win-t...

[2]: https://electionbettingodds.com/WIN_chart_maxim_lott_john_st...

[3]: https://www.metaculus.com/questions/track-record/

[not2b]

We have no way of exploring the multiverse or running the same election more than once, so there isn't any evidence that the payout from a prediction market matches the probability of something happening. It's just collecting the predictions of people who are willing to bet money, weighted by how much they spend, and calling that the prediction. I don't think rich people are necessarily better than not-so-rich people in predicting something not related to their expertise, but prediction markets are weighted according to how much was bet.

[johnfn]

Is this a response to the metaculus track record, which is an aggregate of the accurate of all predictions ever made on the platform, and shows that predictions fall close to accurate probabilities? Which part of the track record do you find fault with?

[not2b]

I looked at it.

Their FAQ says that they aren't a prediction market, and they operate differently. People aren't placing bets, so it isn't weighted in favor of the wealthy. They weight people based on past prediction success.

So, maybe their approach works better.

See https://www.metaculus.com/help/faq/

[aidenn0]

#3 makes it look like things predicted as "very likely not" (10-20%) and "very likely so" (80-90%) tend to be biased towards the contrarian view. Does Metaculus have limits that make it unprofitable to make bets things that would push them outside the window? Anecdotally I've seen quotes from prediction marktets that show a surprisingly high chance for things that are incredibly improbable (to the point that I would call them "impossible" in casual conversation).

[johnfn]

I've been snagged by a few markets that were in the high 90s or low 10s and looked like free money, but ended up resolving the other way. It's usually either 1) there was some subtlety in how the market was written that causes it to resolve unexpectedly or 2) a true inversion - it just seems like human nature to under-predict how often they actually happen.

[aidenn0]

In this case the markets are overpredicting inversions; there's at most a very small difference between how often markets predicted at 5% and markets predicted at 20% resolve as "no." Same for the 75%-95% range.

[flashback2199]

This is true and is a great counterexample. Not too sure why it's being downvoted..

[blibble]

brexit was similar

I remember several people IRL saying "well the market thinks it's a 90% chance of remain so I'm not worried"

[johnfn]

Same response I gave the other guy - you can go look up the market here rather than conjecturing [1]. It was about 25% the day before. In any case, it seems very silly to discount prediction markets because the probability didn't work the way you expected it to in a single event. The whole point of probability is that it's probabilistic!

[1]: https://www.predictit.org/markets/detail/1413/Will-the-UK-vo...

[blibble]

> In any case, it seems very silly to discount prediction markets because the probability didn't work the way you expected it to in a single event.

prediction markets participants are idiots with no underlying knowledge betting according to their bias

they are no different to the "markets" created at the dog track

[johnfn]

If that were true then the results of prediction markets wouldn't closely track the actual results that happen, as the link I provided shows that they do. Also, you could be making tons of money right now! How's that going?

[blibble]

> Also, you could be making tons of money right now! How's that going?

as someone who has spent most of their career working on trading floors watching professional traders do their jobs

I'm perfectly happy staying the hell away from prediction "markets", thank you very much

[johnfn]

I've heard this argument before: you're fully capable of making tons of money, but not particularly interested in proving it for some reason.