Note: The original author is Vitalik Buterin, the co-founder of Ethereum. In this article, he reviewed his experience in participating in the prediction market and summarized the four major problems facing the current prediction market. Vitalik also predicts that in the next few years, the prediction market will become an increasingly important application of Ethereum, and the 2020 US presidential election is only the beginning, and then there are conditional predictions, decision-making and other applications.
Special thanks to Jeff Coleman, Karl Floersch and Robin Hanson for critical feedback and comments.
Warning: I expressed some political views.
The prediction market is a topic I have been interested in for many years. It allows anyone to place bets on future events and use the probability of these bets as a trusted and neutral source of the predicted probability of these events. This is a fascinating application. . Closely related ideas (such as futarchy) have always interested me because it is an innovative tool that can improve governance and decision-making. As Augur, Omen, and the recent PolyMarket (note: these three applications are all built on Ethereum) have shown, prediction markets are also a fascinating application of blockchain.
During the 2020 US presidential election, it seems that the prediction market has finally entered people’s attention. These blockchain-based markets have grown from close to zero in 2016 to millions of dollars in trading volume in 2020. As someone who is very interested in seeing Ethereum applications cross the gap and be widely adopted, this certainly arouses my interest. In the beginning, I tended to simply observe rather than participate in person: I am not an expert in American electoral politics, so why should I expect my views to be more correct than those of others who are already trading? But in my Twitter circle, I see a lot of very smart people I respect. They think the market is actually irrational. If I can, I should get involved and bet against them. In the end, I was convinced.
I decided to do an experiment on the blockchain application I helped create: I bought NTRUMP worth $2,000 on Augur on August 1st (if Trump loses, holding this token will earn 1 US dollars). At the time, I didn’t know that my position would eventually increase to $308,249 and earn me more than $56,803 in profit. What happened in the next two months proved to be a fascinating case study of social psychology, professional knowledge, arbitrage, and the limits of market efficiency. He was interested in any possibility of economic system design. Of people have an important influence.
Before the election
My first bet on this election was not actually on the blockchain at all.
In July last year, when Kanye announced his candidacy for president, a political theorist whom I usually respect quite immediately claimed on Twitter that he believed this would split the vote against Trump and lead to Trump’s victory. I remember thinking at the time that his view was overconfident, and might even be the result of over-internalizing heuristics (that is, if a view seems smart and distinctive, then it is likely to be correct). So, of course I proposed to bet 200 US dollars. My bet was boring support for Biden’s point of view, and he readily accepted it.
The September election once again caught my attention, this time it was the prediction market that caught my attention. The market gave Trump a nearly 50% chance of winning, but I saw many very smart people in my Twitter circle pointed out that this number seemed too high. This of course led to the familiar “efficient market debate”: if you buy a token for $0.52, when Trump loses, it can give you $1, and Trump’s actual chance of losing It’s much higher, why don’t people just come in and buy tokens until the price goes up even higher? If no one does this, why do you think you are smarter than others?
Before election day, Ne0liberal’s Twitter post did an excellent job, summarizing the accurate situation of the forecast market at that time. In short, the (non-blockchain) prediction market that most people use before 2020 has various restrictions that make it difficult for people to participate with a small amount of cash. The result is that if a very smart individual or professional organization sees the possibility of what they think is wrong, then their ability to push prices in the direction they think is right will be very limited.
The most important restrictions pointed out in the article are:
- The low limit that everyone can bet (less than $1,000);
- High fees (for example, PredictIt charges a 5% withdrawal fee);
This is where I objected to ne0liberal in September: Although the boring old world centralized prediction market may have low limits and high fees, the cryptocurrency market does not! On Augur or Omen, if someone thinks that the price of a result token is too low or too high, there is no limit to how many can be bought and sold. However, the price performance of the prediction market based on blockchain is synchronized with PredictIt. If the market really overestimates Trump because of high fees and low transaction restrictions, which prevents calm traders from surpassing overly optimistic traders, then why does the blockchain market without these problems show the same price? What?
The main response of my Twitter friends to this is that the blockchain-based market is very niche with few participants, especially few people who know much about politics. This seems to make sense, but I am not very confident in this argument. So at the time I bet that Trump would lose for $2,000 but did not place any further bets.
Then the election took place. In the initial panic, Trump first won more seats than we expected, but Biden eventually became the winner. As far as I know, whether the election itself verifies or refutes the efficiency of the prediction market is a topic that is easy to explain. On the one hand, through the application of the standard Bayesian formula, I should lower my confidence in the prediction market, at least relative to Nate Silver. The prediction market gave Biden a 60% chance, and Nate Silver gave Biden a 90% chance. Given that Biden really won, it proves that I live in a world where Nate gives more correct answers.
But on the other hand, you can make a case where the prediction market better estimates the edge of victory. The median of Nate’s probability distribution is approximately 370 out of 538 votes for Biden:
The Trump market does not give a probability distribution, but if you have to guess the probability distribution from the statistics “40% chance of Trump winning”, you might give Biden about 300 votes. The actual result is 306. Therefore, in my opinion, the net score of the forecast market Vs Nate is ambiguous.
After the election
But what I couldn’t imagine at the time was that the election itself was just the beginning. A few days after the election, Biden was declared a winner by major organizations and even a few foreign governments. As expected, Trump posed various legal challenges to the election results, but these challenges quickly failed. But in that more than a month, the price of NTRUMP token has been maintained at 85 cents!
At first, people seemed to have reason to speculate that Trump had a 15% chance of overturning the election results. After all, he appointed three Supreme Court justices, and the partisanship intensified. However, in the next three weeks, it became more and more obvious that the challenge was failing, and Trump’s hope of victory seemed to me diminishing day by day, but the price of NTRUMP tokens did not change (in fact, It even briefly dropped to around $0.82). On December 11, more than five weeks after the general election, the Supreme Court unanimously rejected Trump’s attempt to overthrow the vote, and the price of NTRUMP finally rose…to $0.88.
In November, I was finally convinced that the market skeptics were right, so I plunged into it, betting on Trump and losing. This decision is not about money, after all, after only two months, I can make a lot of money by holding Dogecoin, but I chose to donate them to GiveDirectly. More precisely, I participate in the experiment not just as an observer, but as an active participant, which helps to improve my personal understanding of why others have not bought NTRUMP tokens before me.
I purchased my NTRUMP tokens on Catnip, which is a front-end user interface that combines the Augur prediction market with the Balancer constant function market maker. To date, Catnip is the simplest interface for these transactions, and I think this greatly promotes Augur’s usability.
There are two ways to bet Trump through Catnip and lose:
- Use DAI to buy NTRUMP tokens directly on Catnip;
- Use Foundry to access the Augur function, which allows you to convert 1 DAI to 1 NTRUMP + 1 YTRUMP + 1 ITRUMP (here “I” stands for “invalid”, which will be described in detail later), and then sell YTRUMP on Catnip;
At first, I only knew the first option. But then I found that Balancer provides YTRUMP with more liquidity, so I switched to the second option.
There is another problem: I don’t have any DAI, I only have ETH, I can choose to sell my ETH to get DAI, but I don’t want to change my ETH position because of this. It would be a shame if I won the $50,000 bet on Trump, but at the same time lost $500,000 due to ETH price changes. Therefore, I decided to keep my ETH exposure unchanged by opening a collateralized debt position (CDP, now also known as “treasury”) on MakerDAO.
CDP is the way to generate DAI: users deposit their ETH into a smart contract and are allowed to withdraw the newly generated DAI that does not exceed 2/3 of the value of the invested ETH. When they return the same amount of DAI, plus an additional interest (Currently 3.5%), they can withdraw their pledged ETH. If the value of the ETH collateral you deposited drops below 150% of the value of the borrowed DAI, anyone can enter and “liquidate” the vault, sell your collateralized ETH to buy back DAI, and charge you a high amount fine. Therefore, in the case of sudden price fluctuations, it is a good idea to have a high mortgage rate. For every dollar I withdraw, there is more than $3 in ETH in the CDP.
The following figure shows my entire operation process.
I’ve done it many times. Catnip’s slippage means that I usually can only make a transaction of about 5,000 to 10,000 US dollars at a time, so that the price will not become disadvantageous to me (when I skip Foundry and buy NTRUMP directly with DAI At the time, the limit was close to $1,000). Two months later, I accumulated more than 367,000 NTRUMP tokens.
Why don’t others do this?
Before I entered, I had four main assumptions about why so few people buy U.S. dollars for 85 cents:
- Worrying that the Augur smart contract will go wrong or Trump’s supporters will manipulate the oracle (the holders of Augur REP tokens bet on one or more result votes through their tokens) to make it return wrong results;
- Capital cost: To purchase these tokens, you must lock the funds for more than two months, which will prevent you from using these funds or conducting other profitable transactions during this period;
- The technology is too complicated, it is not suitable for everyone to participate;
- In fact, there are far fewer people who are really motivated to seize a strange opportunity than I thought, even if it is obvious.
These four points are reasonable. Smart contract attack is the biggest risk factor. Augur oracles have never been tested in such a controversial environment. The cost of capital is real, although it is easier to bet on the prediction market than on the stock market (because you know the price will never exceed $1, but locking in capital is to compete with other profitable opportunities in the crypto market of). In addition, trading in dapp is technically complicated, so people naturally have a certain degree of fear.
And my actual experience in entering this financial field and observing the evolution of market prices have taught me a lot about these assumptions.
Fear of smart contract vulnerabilities
At first, I thought that “fear of smart contract vulnerabilities” must be the main concern. But over time, I became more and more convinced that this may not be a dominant factor. One way is to compare the prices of YTRUMP and ITRUMP.
ITRUMP stands for “invalid Trump”, where “invalid” refers to the outcome of an event that will be triggered under certain special circumstances: when the description of the event is not clear, when the market resolves it, and when the outcome of the event is not yet clear , When the market is unethical (such as assassinating the market), and other similar situations. In this market, the price of ITRUMP has remained below US$0.02. If someone wants to make a profit by attacking the market, it will be more profitable for them not to buy YTRUMP for $0.15, but to buy ITRUMP for $0.02. If they buy a large amount of ITRUMP, they can force the “invalid” result to actually trigger, which can get 50 times the return. Therefore, if you are worried about being attacked, then buying ITRUMP is by far the most reasonable choice. However, few people choose to do so.
Of course, another way to rule out concerns about smart contract vulnerabilities is that in every encryption application except for prediction markets (such as Compound, various income farming programs), people are surprisingly surprised by smart contract risks. disregard. If people are willing to invest their money in various risky and untested plans (even if they only promise 5-8% annual returns), why are they suddenly becoming overly cautious here?
The cost of capital (that is, the inconvenience and opportunity cost of locking up large amounts of money) is a challenge, and I recognize this more clearly than ever. From the perspective of Augur alone, I need to lock in 308,249 DAI for two months to get a profit of $56,803, which is about 175% of the annualized profit. From the current point of view, even with the various high-yield flows in the summer of 2020 Compared with sexual mining activities, this benefit is also considerable. But when you consider that I need to operate on MakerDAO, the situation will get worse. Because when I want to maintain my ETH position, I need to obtain DAI through CDP, and using CDP safely requires a mortgage ratio of more than 3 times. Therefore, the total amount of funds I actually need to lock in is about one million dollars.
So now, this rate of return does not look very favorable. Moreover, if you consider the possibility of possible hacking or truly unprecedented political events, then the attractiveness of participating in it will be greatly reduced.
But even so, assuming that the locked capital is 3 times, the possibility of Augur contract being attacked is 3% (I bought ITRUMP to hedge the risk), which can reduce the risk neutralization rate to about 35%, and if you consider the real People’s perception of risk, this ratio will be even lower. This transaction is still very attractive, but on the other hand, it seems very understandable now that such a number is attractive to those participants who experience 100 times fluctuations in the cryptocurrency market not enough.
On the other hand, Trump’s supporters did not face these challenges: they only put in $60,000 and cancelled my $308,249 bet (because of the cost, I won less than this). When the probability is close to 0 or 1, as is the case here, the game is very unbalanced, which is beneficial to those who try to push the probability away from the extreme value. This not only explains Trump’s situation, it is also the reason why various popular niche candidates who have no real chance of winning often get a winning probability of as high as 5%.
At first I tried to buy NTRUMP on Augur, but a technical glitch in the user interface prevented me from placing an order directly on Augur (other people I talked to didn’t have this problem… I’m still not sure what happened).
Catnip’s UI is much simpler and works very well. However, automated market makers like Balancer (and Uniswap) are most suitable for smaller transactions, and for larger transactions, slippage is very high. This is a good epitome of the broader “AMM vs. order book” debate: AMM is more convenient, but order books are indeed more effective for large transactions. Uniswap v3 is introducing an AMM design with better capital efficiency, and we will see if it can improve the situation.
There are other technical complications, but fortunately, they all seem to be easy to solve. There is no reason why an interface like Catnip cannot integrate the “DAI->Foundry->sell YTRUMP” path into a contract so that you can purchase NTRUMP tokens in this way in a single transaction. In fact, the interface can even check the price and liquidity properties of the “DAI->NTRUMP” path and the “DAI->Foundry->sell YTRUMP” path, and automatically provide you with better transactions. Even extracting DAI from MakerDAO CDP can be included in this path. My conclusion here is optimistic: the problem of technical complexity is the real obstacle to people’s reluctance to adopt prediction markets, but as technology advances, using them will become much easier.
lack of confidence
Now, we have the final possibility: many people (especially smart people) suffer from excessive humility, so it is easy to conclude that if no one else takes any action, then there must be a good reason why The action is not worth taking.
Eliezer Yudkowsky elaborated on this in the second half of his excellent book “Insufficient Equilibrium”. He believes that too many people overuse “modest epistemology” and we should act according to the results of our reasoning even if the results indicate Most people are irrational, lazy or wrong in certain things. When I read these chapters for the first time, I was not convinced. It seemed that Eliezer was just too arrogant, but after experiencing this experience, I saw some of his wisdom.
This is not the first time I have witnessed the advantages of believing in my own reasoning. When I first started working on Ethereum, I was plagued by fear at the beginning. I was worried that this project was doomed to fail for some reason. I infer that a fully programmable smart contract blockchain is obviously a great improvement over before, and there must be many people who have thought of it before me. So I fully expected that once I published this idea, many very smart cryptographers would tell me why things like Ethereum are simply impossible. However, no one has ever done this.
Of course, not everyone suffers from excessive modesty. Many people who predicted Trump’s victory in the election can be said to have been fooled by their excessive rebellious psychology. Ethereum benefited from the suppression of my own humility and fear when I was young, but there are many other projects that can benefit from more knowledge-based humility and avoid failure.
However, in my opinion, as the famous Yeats famously said, “The best people lack self-confidence, and the worst people are passionate.” This is more true than ever. In my opinion, Xiang Quan The society disseminates the message that the solution is simply to trust the existing output of the society, whether these outputs are in the form of academic institutions, the media, the government or the market, this does not seem to be the solution. All these institutions are able to function precisely because some people think they are not working, or at least some people think they may be wrong at some point.
Witnessing the importance of the cost of capital and its interaction with risk is also an important evidence for judging a system like Futarchy. Futarchy and “decision markets” are usually an important and potentially very useful social application of prediction markets. The prediction of who will become the next president is slightly more accurate and has little social value. However, conditional prediction has a lot of social value: if we do a, how likely is it that it will bring some good things X, and if we do B, how big is the chance? Conditional predictions are important because they not only satisfy our curiosity, but also help us make decisions.
Although election prediction markets are far less useful than conditional predictions, they help reveal an important question: how resistant are they to manipulation, or even just prejudice and wrong views? We can answer this question by observing how difficult it is for arbitrage: The assumptions that predict the market’s current probability (in your opinion) are wrong (maybe due to insufficiently informed traders or obvious manipulation attempts). By setting things up correctly, how much impact can you make and how much profit can you make?
Let’s start with a concrete example. Suppose we are trying to use the prediction market to choose between decision A and decision B, where each decision has the possibility of achieving some desired result. Suppose you think that decision A has a 50% chance of achieving the goal, and decision B has a 45% chance. However, the market (wrong in your opinion) believes that the chance of decision B is 55% and the chance of decision A is 40%.
Suppose you are a small participant, then your personal bet will not affect the result, only when a lot of people bet together. How much should you bet?
The standard theory here relies on Kelly’s formula. Essentially, you should take action to maximize the expected logarithm of the asset. In this case, we can solve the resulting equation. Suppose you invest part of your funds to buy A tokens at a price of $0.4. From your perspective, the new logarithmic wealth you expect is:
The first item is a 50% probability (from your point of view) that the bet will pay off, and the part of your investment has increased by 2.5 times (because you bought US dollars for 40 cents). The second item is that there is a 50% chance that the bet has no return, and you lose part of the bet. We can use calculus to find a way to maximize this. For lazy people, you can use WolframAlpha. The answer here is: r = 1/6. If someone else buys and the price of A in the market rises to 47% (B drops to 48%), we can redo the calculation for the last trader who will flip the market to make it correctly biased towards A:
Here, the expected logarithmic wealth maximization r value is only 0.0566. The conclusion is clear: when the decision is close and there is a lot of noise, it turns out that it only makes sense to invest a small amount of money in the market. This is assuming rationality. Most people invest less in uncertain gambling than Kelly’s formula says. The cost of capital is even higher. However, if the attacker really wants to force the result B to pass for personal reasons, they can use all the funds to buy the token. In general, the game is easy to favor offensive teams above 20:1.
Of course, in fact, attackers are rarely willing to bet all their funds on one decision. And futarchy is not the only vulnerable mechanism. The stock market is equally vulnerable, and non-market decision-making mechanisms may also be manipulated in various ways by determined wealthy attackers. But in any case, we should be vigilant, do not think that futarchy will push us to a new level of accuracy in decision-making.
Interestingly, the mathematics seems to indicate that futarchy will play the best role when the expected manipulator wants to push the result to an extreme. An example of this might be liability insurance, because someone hoping to improperly obtain insurance will effectively try to reduce the estimated market probability of adverse events to zero.
Can the prediction market get better?
The final question to ask is: Is the prediction market destined to make the same mistake? Just as it judged that Trump’s probability of overturning the election reached 15% in early December, even after the Supreme Court (including the three judges appointed by him) asked Trump to get out, its probability of overturning the election reached 12%. So serious? Surprisingly, I think the prediction market will not repeat the same mistakes. I see some reasons for optimism.
1. The market is naturally selected
First of all, these events gave me a new perspective on how market efficiency and rationality arise. Proponents of market efficiency theory often claim that market efficiency occurs because most participants are rational (or at least rationality is more important than any group of superstitions). This is an axiom. But, on the contrary, we can view what is happening from an evolutionary perspective.
The crypto market is a young ecosystem. Despite Elon’s recent tweets, the ecosystem is still out of touch with the mainstream, and there is not enough expertise in electoral political details. It is difficult for those experts in electoral politics to enter the field of cryptocurrency, and there are many inverse forms of cryptocurrency that are not always correct, especially in the political field. But what happened this year is that in the cryptocurrency space, the capital of prediction market users who correctly expected Biden to win increased by 18%, while the capital of prediction market users who incorrectly predicted Trump’s victory decreased by 100% (or at least Is what they bet on).
Therefore, there will be selective pressure. After ten rounds of such predictions, good predictors will have more capital to bet on, and bad predictors will have less capital to bet on. This does not depend on anyone “becoming smarter” or “learning their lessons” or any other assumptions about human reasoning and learning abilities. It is merely the result of the dynamics of choice. Over time, participants who are good at making the right guesses will dominate the ecosystem.
It is worth noting that the prediction market performs better than the stock market in this respect: the “upstarts” in the stock market often get lucky with a one-time gain of a thousand times, which adds a lot of noise to the signal, but in the prediction market, the price is The limit is between 0 and 1, which limits the impact of any single event.
2. Better participants and better technology
Second, predict that the market itself will improve. The user interface has been greatly improved and will be further improved. The complex operation of MakerDAO->Foundry->Catnip will be abstracted as a transaction. Blockchain expansion technology will be improved to reduce participant fees (ZK rollup Loopring with built-in AMM is already running on the Ethereum mainnet, theoretically predicting that the market can run on it).
Third, the demonstration that we have seen the correct operation of the prediction market will ease the concerns of participants. Users will see that the Augur oracle machine can give correct output even in very controversial situations. People from outside the crypto industry will see that this process is effective and are more inclined to participate. Perhaps even Nate Silver himself will use some DAI and use other prediction markets such as Augur, Omen, Polymarket, etc. to supplement its income in 2022 and beyond.
Fourth, the prediction market technology itself can be improved. The following is a market design suggestion that I put forward. It can improve capital efficiency and bet on many unlikely events at the same time, helping to prevent the unlikely result from obtaining an unreasonably high probability. Other ideas will definitely emerge, and I look forward to seeing more experiments in this direction.
Through the first incredibly direct experiment on prediction markets and how they conflict with the complexity of personal and social psychology, it shows a lot about how market efficiency works in practice and what its limitations are. , And what can be done to improve it.
It also demonstrates the capabilities of the blockchain well. In fact, this is one of the most valuable Ethereum applications in my opinion. Blockchain is often criticized as a speculative toy, except for self-referential games (liquid mining, whose proceeds are usually in other issued tokens), it does not do anything meaningful. Of course, the critics are not aware of the exception, I personally benefit from ENS, and even when all credit card options fail, using ETH for payment can also benefit from it. But in the past few months, we seem to have seen the rapid development of Ethereum applications, which have concrete help to people and interact with the real world, and the prediction market is a key example.
I expect that in the next few years, the prediction market will become an increasingly important application of Ethereum. The 2020 election is just the beginning. The future prediction market will receive more attention. This is not just elections, but also conditional predictions and decision-making. And other applications. If the prediction market operates mathematically in the best way, what kind of amazing promise will it bring, of course, this will continue to conflict with the limits of human reality, and hope that over time, we will be able to understand this more clearly Where can new social technologies provide the greatest value?