Actually existing prediction markets?

by Douglas_Knight2 min read2nd Sep 201520 comments


Personal Blog

What public prediction markets exist in the world today? Have you used one recently?

What attributes do they have that should make us trust them or not, such as liquidity and transaction costs? Do they distort the tails? Which are usable by Americans?

This post is just a request for information. I don’t have much to say.

Intrade used to be the dominant market, but it is gone, opening up this question. The most popular question on prediction markets has been the US Presidential election. If a prediction market wants to get off the ground, it should start with this question. Since the campaign is gearing up, markets that hope to fill the vacuum should exist right now, hence this post.

Many sports bookies give odds on the election. Bookmakers are not technically prediction markets, but they are awfully close and I think the difference is not so important, though maybe they are less likely to provide historical data. They may well be the most liquid and accurate sources of odds. But the fact that they concentrate on sports is important. It means that they are less likely to expand into other forms of prediction and less likely to be available to Americans. I suspect that there are too many covering the election for an exhaustive list to be interesting, but feel free point to point out interesting ones, such as the most liquid, most accessible to Americans, or with the most extensive coverage of non-sports events.

Betting is illegal in America. This is rarely enforced directly against individuals, but often creates difficulty depositing money or using the sites. I don’t think that they usually run into problems if they avoid sports and finance. In particular, Intrade was spun off of a sports bookie specifically to reach Americans.

Here are a few comments on Wikipedia’s list. It seems to be using a strict market criterion, so it includes two sports sites just because they are structured as markets. Worse, it might exclude bookies that I would like to know about. Not counting cryptocurrency markets (which I would like to hear about), it appears that there are no serious money prediction markets. The closest is New Zealand-based iPredict, which is limited to a total deposit of US$6000, and it takes a 18 months to build up to that. The venerable Iowa Electronic Markets (restricted to federal elections) and the young NZ PredictIt have even smaller limits, in return for explicit legality in America. It includes two play money markets: Microsoft and Hypermind. Finally, it mentions the defunct play-money Scicast, most notable for its different topic: science and technology. Hypermind and Scicast came out of the IARPA contest. Not on the list, I should mention PredictionBook, which is close to being a play-money prediction market, but tuned in different directions, both in terms of the feedback it provides to participants and the way it encourages a proliferation of questions.

Update: In the previous paragraph, I discarded two sports bookies from Wikipedia's list. I did so because I thought that they had very little non-sports offerings, but in both cases I did a poor job of navigating them and underestimated the numbers. Smarkets still seems too small to be interesting, but Betfair does have solid political offerings and is rightfully at the top of the list.

As of March 2016 my recommendations are:

  • Betfair (see also) is the best real market.
  • PredictIt is open to Americans
  • Metaculus is a play money market
  • Good Judgment Project (Tetlock) is another
  • PredictionBook is useful for complementary purposes, such as recording one's predictions, especially ones that do not correspond to questions on other sites. It is not so useful for integrating other people's predictions or scoring accuracy.
Personal Blog


20 comments, sorted by Highlighting new comments since Today at 3:40 AM
New Comment

Truthcoin and its cousin Augur deserve a mention, even though neither is actually operational yet. (They're decentralized prediction markets on a blockchain.)

Idea Futures is still running (play money), but is functionally nearly dead and has very low liquidity. Once upon a time it was the best option for play money markets, and quite good.

Fairlay is a half-decent (though centralized) crypto market, though it's structured as a "betting market" and has no way to sell back predictions at a profit or a loss (you can place later predictions to hedge your risk equivalently, but you end up tying up a lot of money). Liquidity is bad.

BitBet runs some sort of weird time-weighted pari-mutuel system that I don't like, and has a lot of complaints about shady operations (eg very, very bad customer support that results in people losing money due to interface mistakes), but they often have actual liquidity.

As far as I can discern, the current state of things is abysmal, but I'm pretty optimistic about Truthcoin (less so about Augur in some ways, but optimistic there too).

Worth noting that PredictIt's odds should give you pause: there's money on the table betting "No" on all the presidential candidates, and I find it concerning that they can't interest anyone in arbitraging that away.

The primary reason why Predictit can't interest anyone in arbitraging is that they don't take "netting" into account when they determine your maximum possible loss. Arbitrage therefore requires you to tie up way to much money, such that it will rarely be a worthwhile investment

At the well-functioning prediction market Intrade, you were required to cover only your maximum possible loss across a mutually exclusive set of contracts (such as a list of people of whom at most one will be elected president). In contrast, PredictIt binds up your funds based on your maximum possible loss at any given contract.

As an example: Suppose buyers are offering to buy one contract on each of 10 possible candidates, at 15 dollars per contract. In other words, there is 50 dollars of free money on the table. On Intrade, you could sell a contract on each, and bind up 100 dollars for a year. This secures you a return of 50% In contrast, on PredictIt you would have to bind up 850 dollars for a year to secure the 50 dollars; which dramatically reduces the effective return to 6%

PredictIt also has a fees structure that discourages arbitrage (by taking cuts on the profits, and calculating the profits on the basis of each individual contract).

I tried to start a discussion about this on the PredictIt forum, but the comment did not make it past moderation. The lack of arbitrage is due to the poorly designed structure of the market. Possibly their hands are tied by the CFTC, I don't think it indicates anything sinister about PredictIt.

This is almost certainly one of those cases where there really is free money lying on the street, but each contract price will have to substantially exceed the interest rate in order to make the investment worth it.

is that they don't allow margin trading

What you are describing is called "netting".

Margin trading is when the broker actually lends you money and the securities you buy with it are the collateral for the loan.

Thank you - Edited. I was convinced that "margin trading" was the lingo that people used at Intrade. Either users of that site had their own non-standard terminology, or my memory is faulty. Either way, I hope it makes more sense now.


Just wanted to bring this old thread up to ask you about this service:

PredictionMarkt — Ethereum-based trading platform for prediction markets

I think it solves the netting issue you are describing with Outcome Bundles, which are described in the short sales section here.

The outcome bundles system is identical to the one used at Iowa Electronic Markets.

What do you think?

Even so, at the moment there are sane interest rates available if you tie up your money that way. It's not just the lack of netting; it's the lack of netting, combined with the small deposit limits, combined with the high withdrawl fees. Fix any of those, and you'd see more arbitrage (I think).

Also, they have a really dumb system where each candidate has both yes and no shares, instead of each election having shares per candidate. Which means there are more different prices than there should be, and no system-enforced rule that the sum of the probabilities = 1.

Also, they have a really dumb system where each candidate has both yes and no shares, instead of each election having shares per candidate. Which means there are more different prices than there should be, and no system-enforced rule that the sum of the probabilities = 1.

Actually, the "yes" and "no" shares are the same contracts: Buying a "yes" contract is exactly the same thing as selling a "no" contract. The best offer for "buy yes" plus the best offer for "sell no" will always equal 1, without requiring arbitrage or any action on the part of the market participants.

For some reason they have chosen a counterintuitive user interface such that these contracts appear to be different from each other, but they are the same.

Yes, I suppose my comment wasn't clear. There are twice as many distinct prices as there should be, not 4x. There should only be one price per candidate (plus an additional price for "other" in many cases). The "buy no" price for a single candidate should be equal to the sum of the "buy yes" prices for all the other candidates, and that relationship should be fully enforced by the exchange.

It is common across prediction markets that the fee structure makes it not worth it to push extreme events to further extremes. Thus unlikely candidates have too much mass and the total adds up to more than 1. But maybe Predictit is even worse for the reasons Anders gives.

You can build systems that preserve sum of probabilities = 1. They'll still see bias away from the extremes, because of fees and because of time value of money. But you can do a lot better than PredictIt. (One thing that helps on the fees side is to make fees go down for trades near the extremes; I argued for that in detail on Augur here.

I made a couple of bets on using bitcoin. It was pretty straightforward and easy to use.

Another bitcoin one is

Any thoughts about British bookies as a prediction market?

I don't know much about individual bookies. I was hoping that other people would talk about them. I added an update to the post saying that Betfair is probably the best existing prediction market. It was on Wikipedia's list, but I excluded it in error.

Here are my general thoughts on bookies:

I do think that bookies are currently the best source today for odds on the US presidential elections. The have the disadvantage of usually being opaque, rather than markets. But that same structure has the advantage of not swinging as wildly when large bets are placed. And my impression, is that even if they are structured as markets, they are less likely to record the past movements of the price, which is valuable to someone who wants to test the quality of prediction markets, rather than using them as black boxes (either for making money or for learning info).

But even though sports bookies are probably better than existing prediction markets, I think that if a prediction market got off the ground, it could be better than a sports bookie who only does it as a hobby. A sports bookie that took it seriously rather than as a hobby could be great, too, and maybe that's the direction Betfair is going, but it's moving slowly. The advantage of doing it separately is getting quasi-legal American participation. I think that the mindset of a prediction market rather than a sports bookie is important. For example, long-term predictions may require different rules than short-term sports bets. If the market views itself as selling information rather than entertainment, it might act differently. It might change the rules to stop encouraging overestimation of low probability events. But it might not, since they are good advertising.

[-][anonymous]5y 2

Gwern has a list here, although I don't believe he's added any of the more recent bitcoin ones.

foresight exchange. It uses no real dollars though. But it exists for quite some time now.

Stock markets. I am using them continuously (if passively).

There are some kind of forecasting tournaments provided by Phillip Tetlock mainly associated with politics issues, however I've found no info on how to enter one. Here is a short introductory course

In my opinion prediction markets are still very raw concept which doesn't grow and spread very well in its current form and needs capital transformation

The web page is here. In previous years they recruited new participants about this time of year, but they don't seem to be doing so this year. They are transitioning off of IARPA funding and much of the page is broken.

This was one of many competitors in the IARPA competition. I didn't include them because their predictions were not public. They couldn't be public, because the whole point was to compare their results. Except Scicast, because it wasn't predicting the same things. Even if you joined GJP, you probably only got aggregated predictions across your team, and maybe not even that.