Topic: [WHITEPAPER] Decentralized Bitcoin Prediction Markets - page 3. (Read 26332 times)

Ah good, thanks sportscliche for posting those.

From what I've glanced through of that extremely lengthy discussion, I finally would add to my comparison to say:
1] Truthcoin (in attempting to directly model a traditional prediction market) makes an extremely direct attempt to ensure that the winning state of a market will sell for 1 and losing states sell for 0. BitShares does not attempt to ever directly influence the price, and you have to hope that market participants give you the price you want.
2] Creating a new market appears to be much, much easier in Truthcoin (designed to do this all day, liquidity guaranteed) vs BitShares (apparently requires a whole new blockchain and new participants).
3] Finally, BitShares can force you to sell at the current market price, which seems unstable and manipulable. Truthcoin cannot force any trade.

It is very important for this community not to duplicate work, so it is good (and not a derailment of the thread) to have these discussions. However, I now feel that I've made a strong case for Truthcoin being an extremely separate and valuable project.

Bitshares was discussed at length on these forums when it was proposed last fall.  The questions in this thread have been raised before:

https://bitcointalksearch.org/topic/m.3214589
https://bitcointalksearch.org/topic/m.3036037
https://bitcointalksearch.org/topic/m.3205395
https://bitcointalksearch.org/topic/m.3282870
https://bitcointalksearch.org/topic/m.4621908
https://bitcointalksearch.org/topic/m.3610353

You can read through these and decide for yourself if they've been adequately answered.

Thanks for getting this additional info.

Regarding quote 1: I personally don't find a string of text "BitHillaryWinsIn2016" to be a convincing explanation. My understanding of BitShares is that it relies on infinite term arbitrage. However, if Hillary wins, who would then buy my shares from me for 1$? Who will trade them at all? Future value = 0, present value = 0. Is it not that simple?

I admit my low interest in exploring BitShares, as I was only trying to answer 'What is the difference [between Truthcoin and] the new BitShares approach?'. If anyone feels I could do more to answer THAT question please let me know. I am still of the opinion that BitShares is not a prediction market, defined as a place where people buy securities which are worth a certain value should an event occur, and zero otherwise. BitShares doesn't even seem to limit prices to a (0,1) range, but perhaps I have misunderstood this.

Regarding quote 2: I am (for the time being) personally more than willing to spend my valuable time answering ANY questions people have about the work I published. However, this is an entirely different matter from attempting to convince everyone of the merits of the idea. I do not expect everyone to agree, certainly not at first.

Reinterpreting that comment as the question "Doesn't the fact that the system use voting to establish consensus make it fundamentally flawed?": Firstly, the system does not use 'voting' to establish consensus, the consensus algorithm uses a SVD-modified weighed-vote for coin-owners only. Coin owners have the highest ROI when future trading is maximized, a proof-of-stake system. The most significant multivariate-outlier-voter loses the most coins in the following period, and has the lowest relative influence within the current period. When voters lose in this way, honest voters gain, so every single actor has an incentive a] to vote honestly, and b] to lie about their voting intentions, discouraging cartels. The (required) accumulation of several votes into a Ballot is also a powerful decentralizing factor.

Figure 3 (page 28) best disproves the claim that 'only voting' is used. In it, the weighted votes for contracts C2 and C3 are split between Yes and No 50%-50%, yet the mechanism determines that C2 was 'Yes' and C3 was 'No'. This is because of SVD.

I doubt that Mr. Larimer seriously read and comprehended the paper in the 3.5 hours which separate those posts. His answers paint him as disinterested (which is fine by me).

This was years ago (2004?). It became hard to get money in and out of my account in Ireland and the contracts became less interesting. Originally, if you followed the news and did some homework you could outguess the market as a whole and reap some profit. It seemed to me that over time the contracts became more difficult to predict and celebrity based. I didn't know about their recent troubles. Pitty. I'll offer my defense of prediction markets as penance for my derailment.

I would love to see a good implementation of a prediction market in bitcoin. I think it could serve a use beyond speculation. It could be a new important media source, even an everyday app for making decisions. In the famous story about the discovery of the "wisdom of the crowd", Francis Galton finds the best guess as to the weight of an ox is a large sampling of guesses. What is less emphasized is that the guesses were actually bets. By adding a scalable risk and reward people are encouraged to act in their own best interest thus doing better research and mitigating personal bias. In aggregate the results reflect more than say a poll. A poll is a better gauge of what people are willing to say than what their internal best guess is.
In the future I could easily see monetized crowd sourcing for all sorts of decisions. Some people would pay a lot of money for best guesses now. Intrade was a goto for many in the press during elections. Anyway, I'd be interested in it.

The way that it forms the consensus on the prices of assets and thereby how to collateralize these assets is based on the same underlying premise as any prediction market.

Just from glancing at it now, it seems that BitShares would be for a continuous price/time-series on an asset that we expect to never be removed from popular exchanges such as the DJIA, Oil, Gold, whereas Truthcoin is for binary (settling in finite time at 0 or 1) events which do not have to exist anywhere else.

From what I can tell, BitShares is actually not a prediction market. How I can use BitShares to bet on the Election of Hillary Clinton as US President in 2016? How do I collect money if I am correct? etc.

Instead it appears (and claims) to be a fully collateralized bank for paper assets, using a custom currency they can suck away from you. Debt seems to be created with every transaction so it will be interesting to see how stable these markets are.

You're quite welcome.

Some opinionated yet informative links:
http://www.forbes.com/sites/timworstall/2013/03/11/intrade-closes-its-doors-but-why-is-this-another-mf-global/
http://www.foxbusiness.com/on-air/stossel/blog/2012/11/26/government-crushes-innovative-online-prediction-market

sorry for offtopic, but why is intrade currently deactivated?
You must have information about this when you were cleaning up on contracts there.


(shoot me a pm if you don´t want to derail the thread further)

what is the difference to new bitshares approach? - read their (new) whitepaper and for it sounds like a very smart system of implementing prediction markets

Thanks for doing this. Prediction markets are fun. I used to clean up on contracts at http://www.intrade.com/v4/home/
If you are a news junkie and regularly can see what is happening early on, you might like this a lot.

Surely you mean "can't elsewhere gain"?


The incentive structure for Authors (the people who create a prediction market) is that they pay fees up front, and get (1/2) of 1% of the total market lifetime trading volume. So any markets which are failing just will stop being made, and new markets will be tried by entrepreneurs who believe in them.

Nonetheless, if markets were created with low or even ZERO volume, these markets still have permanent liquidity thanks to LMSR (a clever but almost-completely-overlooked PM invention). One can always make a trade at the current market price, changing it to their preferred price (and probabilistic expectation), and profit from that trade (if they are right), even if they are the ONLY trader. So if you had insight you could still use Truthcoin to profit from that insight, in addition to speculating elsewhere. In fact (if someone else created the market) you'd have to be very lazy not to make a profitable trade, as using Bitcoin is far easier than opening an InTrade account. You could even make a 'billion dollar' trade, but that trade would itself cost you roughly a billion dollars as it would crash the price of your state into 1 almost instantaneously.

So even in this case the PM would aggregate all insights, obvious or otherwise.

If other people really want your info, they can make a very liquid market via donations ("Amping Beta") or dominance assurance contract, allowing you to make reap large profits on your large trades. Of course, to make a billion dollars in profit, you'd need to hope that other people donated at least that much, which they would only do if they cared about what you had to say.

I'm sorry for not being clear.  My point was that if I have any non-obvious insight into how Fed policy will shape the future, I can already make a billion dollar bet in other markets.  I don't think it's realistic to expect Truthcoin to do anything but aggregate obvious insights regarding future events that bettors can elsewhere gain as much exposure as they want to.

The only markets on InTrade that received any volume were the election markets.  There wasn't any other simple way for a bettor to gain exposure to those events.  InTrade got shut down for pissing off the CFTC by offering prediction markets on the price of commodities.  Those markets received no volume.

Hey I want to answer this question but I really don't understand your second sentence.

You might read another paper I placed into the docs folder, https://github.com/psztorc/Truthcoin/tree/master/docs , called Combinatorial Binary Prediction Markets.pdf, and if you look at Example 3 you can see how you can use these PMs to bet on the effect FED policy will have on something. If you make it to all the way to Example 6 you might see exactly how you can reshape the entire distribution of, for example, inflation outcomes, based on, for example, a FED policy decision. If your forecast is an improvement over the existing forecast, (and depending on how lucky you are), you can expect to profit.

If you have extra-special insight, you could actually create this market yourself, set the probability to what you think it is and bait other people into voting against you. On InTrade and predictious, you cannot create a new prediction market (although you can request one), and in fact I believe they are very slow and un-entrepeneurial in providing what people want.

Even with a large majority of attackers the Nash Equilibrium is a collapse into honesty, but only for True/False I'm afraid. You can find the 'voting strategy' section of the paper on page 11 in which I discuss all the bad things I could think of.

I applaud your creativity, but who is deciding the outcome of "On February 20th, Feed Y will be more trustworthy than Feed X,"? Presumably people would compare Feed X and Y to what-they-decided-to-use, which is probably just another corruptible feed.

I actually think this simple idea could be very useful though. If we say "Feed X to be 'accurate' through the month of March 2014?", we can at least audit a continuous feed. But what if the feed is mostly correct, with one wrong value out of 1000, or one missing/censored value. Is the whole feed wrong?

tl;dr A little of both. Truthcoin designed such that voters know what to vote without consulting any external source at all. Authors should not create high-search-cost-Decisions as they risk loss of funds.

INCENTIVE COMMENTS - Why centralization might not matter
I am of the same opinion of gwern that bad data will be ignored.

Firstly, I state in the whitepaper that these markets are only appropriate in certain conditions, one being low search costs (and I don't consider relying on an automatic feed to be actually 'searching for the information' at all, as the automation literally avoids doing just that, which implies that searching for the info was not worth the voter's time [ ie high cost] ). Much more on this later.

To the main issue: in order to prevent people from lying about reality, people vote in a simple binary True/False way but are actually punished continuously, such that the most deviant person is punished more (much more) than the second-most-deviant, and so on. Therefore actually every voter has incentive to actually lie to each other, including tricking rival voters into using a broken feed, hoping to fool someone into voting incorrectly. For this reason voters may anticipate that >51% of other voters have noticed and caught the error in the data source, and so themselves correct the error.

Additionally, voting involves investment. To vote, one one needs to buy some coins (or forgo the opportunity cost of selling preowned coins), and yet one only realizes the full benefit of this investment in the future, in the form of collected trading fees. "Screwing up" crashes the trust of the system, meaning low (or zero) trading thus ruining the profitability of the investment. For this reason, voters might vote against a bad data feed even if they believed that it would likely win >51% of the voting power. If the bad data successfully alters the network, all coins might be worthless, so 'risking' them on an honest vote (and just hoping and praying to grab the critical 51%) isn't much of a risk at all. Quite the reverse, actually.

So really it is a nice mix: traders are paranoid enough not to trust each other (which might lead to herding as you indicate), and yet greedy enough to cooperate to promote the value proposition of the service. It is quite a bit like Bitcoin mining this way, where miner-competition secures the network, yet the mining pools are ready-and-able to hold these emergency cooperation-meetings to solve Bitcoin problems.

Finally, different Truthcoin branches can change the decision fee magnitudes. Branches with higher decision fees are more profitable, and voters would have more to lose by disappointing Traders, and so more to gain by 'double-checking'. I once worked in marketing dept for an upscale trading firm, every single email, pamphlet, website edit was literally triple-checked by the hawk-eyed lawyers / grammar PhDs of Compliance. I literally mean three times by experts. Internal Auditing would regularly force someone to drop everything and spend a week preparing for what the firm would do IF we got 'externally audited' for the accuracy of our published claims (which, as far as I could tell, never even happened). Creating a new branch is (theoretically) very easy, so if some voter-failure is inevitable, competition should at least produce exactly an efficient level of failure.

SCOPE - Avoiding Centralization
Now my discussion of the scope of this project. I regret not specifying this in the paper as clearly it really is quite important.

Unfortunately, (and I mean it), I could not find a magical solution to everyone's prediction market problems. However, I focused on a design that should be able to accomplish "the easiest" PMs, like "Will Hillary Clinton win the US Presidency in 2016?", impending wars, critical regulatory outcomes, disasters, or breakthrough technological/biomedical research.  Items that, given the widespread nature of the info, and social benefits, we really """should""" have. I'm not a very emotional guy but I literally experienced sadness and disappointment, that we as a society (including Silicon Valley, scientists, philanthropists) simply did not care enough about progress to construct PMs ("institutions that actually work"). Bitcoin isn't perfect. For a while it was really hard to use, it still is to most people. 1 hour confirmation time? Soft forks? Having to spend days downloading and verifying the entire transaction history? legal ambiguity, crazy exchanges, network analysis of the blockchain, malleability, it goes on and on.

Yet Bitcoin restricted its focus to what it could accomplish, and so despite abuse it's still alive and kicking. I also placed restrictions on Truthcoin, but mine are easier to misunderstand. For example, early in the paper I note that the voters are instructed to vote .5 (the "information unavailable" vote [ recall that this vote punishes the Authors who created this Decision ]), even when the information IS available if THEY also DECIDE that the search costs of that info are prohibitively high (as in 'more than a google search')! I expect this attempt to generally fail, with voters doing a lot of extra research-work in the hope of not upsetting any Authors/Traders, thus improving "the service" but in the process opening up a can of systemic risk.

So Truthcoin was designed with a focus on after-the-fact outcomes of which everyone is already AWARE (as in, already thinking about RIGHT NOW, such as the identity of the US President). Because different humans can be aware of different things, the 'branching' capability allows us to cheat this limitation. "Truthcoin - Main" might be full of hobbyists who just google the latest DJIA, but "Truthcoin - Serious Finance - US CRUDE Wednesdays" might be a niche market with restrictions and incentives such that only professional oil speculators would decided to own those coins, and so they have absolutely no need to rely on any outside source because the information they provide to their branch every Wednesday evening was what they already spent their entire day doing. This is an extreme example to make the point, I don't actually expect to see a branch of this kind (but whatever the market will support is fine with me).

CONCLUSION
For these reasons, my expectation is that point 5) will actually not occur, because although lazy voters might (conveniently and efficiently) "auto-pre-fill" their Ballots, I expect them to be so paranoid as to at least glance over it manually to see if someone has tricked them into casting a vote indicating that John McCain was elected president.

For further thought: one idea I had to increase reliability at the expense of speed, would be to have multiple rounds of voting, where (for each Decision) the "newest" round is more influential than all previous rounds. Then, the truth would have to deliberately and slowly build up to some set parameter of "certainty", with attacks delaying this build up but not preventing it. As feeds would (we hope) eventually become correct, attackers would have to be constantly attacking the validity of the vote system. I concluded that that was overkill for the basic scheme but it might be appropriate for a branch.

------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Now a rambling comment for EE about my failed attempts to create a decentralized time-series feed:

I get the feeling that you may be comparing Truthcoin and RealityKeys a little too directly (not through any fault of yours). As I indicated on Twitter, I feel the services are so different that they can hardly be competitors, with RealityKeys being a low-cost high-volume service and Truthcoin being high-cost low-volume (although with branching who knows). For starters the information in a data feed is generally not in anyone's awareness, because we've passed the job completely to robots.

Specifically, I doubt that this service will be used for 'feeds' at all as the cost would be prohibitive. Think about it: a time series needs maybe 10 partitions per time unit to guarantee any useful information. If you wanted a daily feed, that's 300 per month for one variable. I expected Voters to do about 100-200 Votes per month per "branch" (as in, in TOTAL).

"Why not have the voters just vote on a level, such as 4, 70, 8, instead of just on TRUE FALSE?" Well I'm glad you asked.
1] I exploit .5 as a maximally-uninformed outcome in 1 vs 0 land, whereas this is impossible with continuous variables because the mean, median, and mode a) are just estimates of 'maximally-informed' and not logical requirements for 'maximally-uninformed, b) can all be gamed, and c) are so flexible that I lacked the ability to even determine the entire attack space.
2] Confusion over rounding, units, language. Is 3.0104 different from 3.01040013? Which is better? What if the contract says "rounded to the 6th decimal place", but the 2-6th decimal places are zero?
3] PCA actually assumes that all the variables are normalized, implying a measurable mean and sd. What happens to those numbers if some lunatic with 1% of the voting power types in 6 quadrillion?
4] True / False harmonizes the units, which are identical [0,1] range, preventing the incentive scheme from varying with a shift in the average-unit-variability of Decisions.
..and MORE.

I did derive this insane way of doing a continuous ts variable "feed", which ended up using regular expressions to constantly search the marketplace for a set of markets worded in an pre-standardized way and determine the relevant range in which a continuous variable was expected to exist. Then I tried to make it so that entrepreneurs would always have an incentive to create-and-trade-in markets on a log scale as close to the real-world level as possible such that the range would shrink to an acceptable level of clarity. I could not figure out a way to make this viable. In fact I am going to have nightmares tonight just re-living this explanation.

There are two "solutions" to the feed problem: consensus/diversification, and incentivization.

The first involves either spreading your bets among different feed operators, or betting on a (trustless) "average" of feeds. If averaging, ideally the algorithm should weigh feeds somehow (e.g. by volume or transaction number), and creating a bet should have a nontrivial cost (to prevent fake volume).

The second involves incentivizing operators to report values "close" to the consensus. This could come naturally in the form of participants choosing one operator over another (we assume in this case that feed operators are "fungible", even though that isn't the exact definition), or users betting on a consensus value, with operators that report values "closer" to the consensus getting a larger share of the bet fees. As feeds ostensibly apply to actual truths about the world (as opposed to predictions, estimates, etc), I don't think the "lemming problem" is an issue here. On the other hand, I'm not sure how resistant this scheme is to cartels, but for any easily verifiable value (GDP, stock market index, temperature), collusion and cheating should be easily detected.

Not speaking for TruthCoin here but any prediction market, especially one that can be used anonymously, should incentivize you to bet on your insight in the hope of making money. When you bet this will move the price (at least somewhat) which will make that price a better indicator of Fed policy or whatever than it would have been without your participation.

Let's say I'm an actor with a non-obvious insight into the future effects of Fed policy.

How does Truthcoin incentivize me to add my insight to the market?  My status as someone with non-obvious insight into the future effects of Fed policy means my opportunity cost is rather high.