Prediction Markets: When Do They Work?

Ok, having finished both articles, a few thoughts from somebody who has been involved for years in the subsidized prediction service Long Bets: http://longbets.org/

Broadly, I think he's right. Subsidized prediction contexts could provide real social benefit. It can force a lot of attention and thought around topics that don't get enough consideration. I would love to see, for example, a bunch of VCs covering the costs on a prediction market for the tech sector. Have a bunch of standard running questions, and let the VCs ask any question they want.

I think he is 100% correct that having well-defined bets is hugely important. We got our initial bets by careful, personal assistance to get people with contradictory opinions to agree on clear, testable criteria. We thought once we got going people would just do that on their own, and we were wrong. Sometimes it happens, but pretty rarely.

I wonder about the extent to which his emphasis on speed is about a particular kind of participant (e.g., sports gamblers and day traders). Most of the people I know who invest work at the months-to-years timescale, so I suspect there's an untapped cognitive surplus from people like that which could be applied to longer-term questions.

I also think his notion of subsidizing market makers (who are obliged to provide liquidity) is a good one. I used to work for a market-maker and it was not a bad business to be in. His no-fees market makes that trickier, but I don't think it's unresolvable.

One subsidy option that I'm surprised is not here: guarantee a minimum purse for people on the winning side of a question. Somebody is going to show up to try to get the free money. And then somebody smarter is going to say, "I'd bet I can get a piece of that." I know it wouldn't take much for me to place some bets on some current hot issues. E.g., blockchains, Bitcoin, Uber, scooters, VR, AR.

For political, other major events or sports betting, normal brokers are more efficient. They have spread something like 5% or less. Financial markets can be even more efficient. Existing betting markets typically take something like 10% for the same bet. There is little reason to go to them unless there are some rare arbitration opportunities.

I think these new prediction markets should differentiate from traditional betting markets in some meaningful way.

One possible idea: marketing the skill. Turn prediction market into expert market. People would invest money to experts (individuals, syndicates even algorithmic syndicates).

Rough idea: trading is anonymous until the underlying is settled. Then the market participants get a cumulative score they can use to sell their experience and get investors. The system would show the expected value form investors, quantified risk and uncertainty. Individual investors and investor syndicates can start with very little cash and demonstrate their skill. Complete nobodies could become renowned experts over time and sell their expertise.

This would could be very good way to provide value in emerging markets. Some Kenyan or Vietnamese market or political analyst could emerge from nothing over few years and sell their expertise to bigger players without PowerPoints, marketing skills and connections.

I used to be excited about prediction markets for a while, but I've since accepted that, except for some niche applications, they won't be terribly useful and possibly even dangerous.

I was ultimately convinced by NNT's arguments, that a) prediction is the wrong thing to focus on (robustness or "anti-fragility" to wrong predictions is) and b) humans aren't good at predicting [the outcomes of complex systems] one way or another. PMs may be better than expert panels, but reality doesn't award extra points for being marginally less wrong.

Prediction markets are also called "information aggregation markets" and in that capacity they may be useful. Assuming all relevant information is out there, but doesn't properly propagate upwards a hierarchy for "behavioural" reasons (e.g. nobody wanting to be the face of bad news), then an anonymous (!) market could be the way for information in the low ranks to move where it is needed. In this case, there is hardly any "prediction" involved, and I assume it could work for that reason.

This is basically unreadable without the previous article in the series. It's linked at the top, but this is it: https://thezvi.wordpress.com/2018/07/26/prediction-markets-w...

The article shows an excellent understanding of gambling and financial markets, but it's let down by the author's surprise about the terms of the "ETH over 500" bet.

This is a variation of a quanto future, which is well known and well understood in derivative pricing circles (see Wikipedia link). If you can price ETH/USD options, you can price this. Possibly it should be called a binary quanto.

https://en.wikipedia.org/wiki/Quanto

"Insider trading of securities is illegal....The problem is that it drives people away.

This is an interesting observation in light of some conversations I've had with 20-somethings who have abandoned the stock market in favor of cryptocurrencies.

For a couple decades now, the message has been "you can't beat the stock market. You're competing against top mathematicians who do this for a living and colocate boxes in the exchange so they can execute trades faster than you. Why even try when you're just going to be the sucker for some finance professional? Go invest in index funds instead." (Note that this advice is quite self-serving for a fund manager: it justifies the huge fees they charge. It's also quite self-serving for index funds, which compete on price.)

It seems like the youngest generation - the ones just leaving college now - took this to heart and figured "Well, if there's no point for us even trying in the stock market, let's just throw the whole thing away and invent a new asset class", cheerfully ignoring that they are also throwing out all the regulatory protections that protect against insider trading.

Even with all the SEC's insider trading laws, the existence of vast computer systems capable of crunching terabytes of data and executing a trade in microseconds now creates such an information imbalance that people are leaving the market anyway.

It makes me wonder if all markets have a natural death. As they get more efficient, liquidity dries up. Fewer and fewer people bother playing when they know that someone else is just going to get rich off them. Eventually the winner becomes like the kid on the playground who insists on winning every game; all the other kids refuse to play with him, and he's left shooting hoops perfectly alone. Meanwhile they go off and create a new game that is fresh enough that new entrants might actually have a chance - at least until somebody becomes really good at that, and the cycle repeats itself.

Always worth a read in these discussions. Hanson has a prediction market business, Consensus Point, so there’s at least two sides to the issue: convincing specific customers to care vs the general applicability.

https://www.cato-unbound.org/2011/07/13/robin-hanson/who-car...

Hanson is not naive about the social and political barriers. This goes far beyond prediction markets, extending to all kinds of uses of forecasting and statistics in industry where political gatekeepers want to control the implications of models, rather than caring about objective truth first and mollifying it towards business concerns second.

Eventually an anonymous prediction market will take off, and that will be the end of insider trading enforcement. People in the know will bet on the stock price via the prediction market, and the market will react based on those bets.

For anyone interested in this field should take a look at the work of Ulrik William Nash. I had the pleasure of completing a course led by Ulrik as well as writing an internal paper with him as my supervisor:

His main paper on wisdom of crowds and skew, which is rigorous (free): https://journals.plos.org/plosone/article?id=10.1371/journal...

Short presentation: https://www.youtube.com/watch?v=yqTbzSHi-UQ

The paper above is worth reading for anyone interested in the utilization of collective wisdom.

Seemed like an interesting article until halfway through the 1st paragraph, (on mobile) a full screen ad locked down my whole screen and caused me to close the broawer tab. Booo

>The biggest market there, by far, is on whether Ether will trade above $500 at the end of the year. This is an interesting market because Augur bets are made in Ether. So even though the market (as of last time I checked) says it’s 74% percent to be trading above $500 and it’s currently $480 (it’s currently Thursday, July 26, and I’m not going to go back and keep updating these numbers). When I first saw this the market was at 63%, which seemed to me like a complete steal. Now it’s at 74%, which seems more reasonable

Hi. My job is to essentially figure out reasonable probability distributions for securities on which to price derivatives contracts. Certainly, due to some securities having negative skewedness, it's common for a stock to have more than a 50% chance of going up. I've seen it as high as 65%, but the counterpoint to that is that there's usually very small probability that it goes a lot higher.

I would love to know what probability distribution this guy is using to think that the probability of going up over $20 being 74% is reasonable.

Sounds a bit like a change of numeraire he's talking about there wrt things being prices in Ethereum. https://www.math.nyu.edu/~alberts/spring07/Lecture3.pdf

What about a market that brings together buyers and sellers to bet on how companies are doing? A share price is really a function of how we expect a company to do in the future, but what if we bet on specific metrics? For example, I expect Company A to hit $100 in gross revenue this quarter, so I'd like to bet on it - as opposed to the several variables that affect share prices. The share price may reflect the Company's guidance - but revenues it expects in the future, but not really a reflection of history. This could also be a bet on other metrics - Snapchat number of users, etc. etc. I'd argue that this can be a more rational betting market. There are several metrics that aren't 100% correlated with stock prices.

I’m new to this concept. Can someone help explain how prediction markets are used for good? Given that this is a zero sum game, isn’t this just facilitating the transfer of money between the winner and the loser.

I hooe you welcome some critique:

> Remember, if you can’t spot the sucker in your first half hour at the table, then you are the sucker.

given there is no insider information in this case, this holds true to the stock market too.

> There’s a lot of interest in what the odds are, but the volumes traded are quite thin, so much so that it is in the interest of partisans to trade in order to move the price and thus change the political narrative.

true when these markets are only a couple of bookmakers. Also true when there is a goobalized easy access anonymous system without a bookmaker's fee? Not so sure.

sports betting and commodities futures could be considered types 'prediction markets' that seem to work pretty well. The reason why prediction markets have not caught on is due to regulation and that most markets are tiny and prone to manipulation and low liquidity.I also think that insurance acts as a counter-party instead of having to use a market.

If prediction markets are of any value, couldn't you set one up to answer this question?

Ugh. I'm torn when I see things like this. I sincerely like the rationalist community, but I never quite engaged with it. And posts like this seem to confirm that it was probably a good decision on my part.

The author identifies 3 successful instances of what he wants to exist, and then fails to ask the question of "how do they accomplish what I want to?"

In particular, one of the marketplaces, the financial markets, is several orders of magnitude larger than the others and has a huge vested interest in getting issues like this right. And they largely have.

For instance the capital issues are handled through a broker system, where the aggregation of opposing bets allows for mitigation of risk from margined bets, and the settlement processes, such as daily settlement of price differences in futures.

If you're trying to build something complex, it's probably a bad idea to start from a blank slate. A lot of our existing solutions were hard-won, and it would be a shame to simply discard that understanding. The programming community, and others connected to it, seem to disregard history and existing institutions as obsolete or to be distrusted. For a field that's less than a century old, that's somewhat defensible. Nassim Taleb makes the point in Antifragile that ideas usually have something akin to a constant hazard rate. Something old which is still used is less likely to be replaced than something that has only been around for a short while. The nascency of most of science makes it hard then to determine what is and isn't fungible. But marketplaces are as old as trade. The aspects that we see in them are far less arbitrary than you'd think, even if it's hard to express exactly why they are there. If you're going to work at building another version of something that has been around for so long, you'd be wise to start by analyzing and understanding what is already out there. I think it's the engineering equivalent of Eric Weinstein's policy around talking to people you know to be intelligent. If they seem to be saying something obvious, they are probably saying something subtle. And if they say something that seems wrong, they may instead be saying something counterintuitive. Similarly, if part of an antiquated system that is still in use seems to be useless, needlessly complex, or downright harmful, there is a good chance that you simply don't understand it.

I was really excited last year when I saw some people within lesswrong trying to become more accommodative and appreciative of "hufflepuffs" within the community (https://www.lesswrong.com/posts/DbdP8hD2AcKcdSsgF/project-hu...). I really hope that effort continues and sees fruition. Hangups like this could be easily spotted and remedied by having a contingent of people who are less philosophical and more battle-hardened within your ranks.

Weak. Any proper treatise on this would be mathematical. A comparison of polls versus prediction markets would be appropriate, based on intervals of confidence.

I seem to recall at least one instance where the prediction markets absolutely killed it on the US state elections, so yeah I'd say yes.

Prediction Markets were at one time supposed to be the “secret weapon” of neoliberalism. They would so predict the future, that corporations would become unstoppable, much like the company that hires Ben Affleck in the movie “Paycheck”.

Most laughably, in 2003, Admiral John Poindexter, thought it would be a brilliant idea to launch a “terrorism and coups-d’etat prediction market”.

About 2 months after it launched, he became persona non grata in the Washington defense scene and was forcibly retired from DARPA.

http://www.cnn.com/2003/ALLPOLITICS/07/29/terror.market/

https://en.wikipedia.org/wiki/Policy_Analysis_Market#Proposa...

I would bet there is almost no overlap between the people betting in prediction markets and those who dont screen email/websites/calls (ie, answer polls honestly).