Information Is Trapped
Imagine you have done research on Apple’s hardware manufacturing. After investigating for some time, you know that iPhone sales in China will exceed all expectations, and you think that Apple is about to skyrocket. But today, the best thing you can do with this information is to buy Apple stock.
In reality, the vast majority of information is difficult or impossible to express as a trade. If you download a trading application on your iPhone and buy shares of Apple, you’re exposed to all kinds of risks that lie outside the information you have. Your prediction about iPhone sales in China might turn out to be exactly right but you could still lose money on your trade. Currency fluctuations in the US dollar and Chinese Yuan could move against you. The US stock market may go down as a whole and crash your Apple shares along with it. Even if your information is great, there are many ways you could lose money on your trade.
So you decide not to buy Apple shares, and your information never reaches the market. That’s bad for you, because you miss out on a rare and valuable opportunity. And it’s bad for the world, because it misses out on your information and a more efficient stock market.
Instead of trading 100 shares of Apple on Robinhood, you’d much rather sell your prediction to a hedge fund who can get maximum value from your information. Hedge funds can hedge out the market risk, currency risk and other risks by making clever offsetting trades to get full value from the information. They can also trade your prediction with diversification — combining it with other predictions they have. Your own personal trade might only have a small expected return with a lot of volatility and risk. But your prediction might work fantastically inside a hedge fund’s diversified, low-volatility portfolio.
In short, your information is much more valuable to a hedge fund than it is to you. And it’s better for the world for the hedge fund to have that information instead of you because a hedge fund with $500 million dollars can make much larger and smarter trades than you can with $5,000 in savings and a Robinhood account.
If it’s so clearly better for individuals to sell information to hedge funds rather than trade themselves, where is the great data marketplace for predictions?
It doesn’t exist. And it never could have existed until now.
Asymmetric Information And Market Collapse
You’re a hedge fund manager. You get an email that says, “over the last two years I have predicted Google’s daily stock price with 70% accuracy, and I’m willing to sell you my next prediction for $10,000”. You calculate that this next prediction, if its right, can make your fund an expected $1 million in one month. Do you buy the prediction?
There are just too many problems with credibility to want to buy the next prediction. Did this person get lucky? Are they simply lying? If their past performance is real, where is the guarantee that it will continue? The seller of the predictions knows their quality but the buyer has asymmetric information. There is simply no way for the buyer to verify the quality of the predictions.
Because of these questions around the credibility of the seller’s claims and the quality of his predictions, the people with genuinely predictive data cannot find a market for buyers of their information. No one believes what they have to say. The asymmetric information between the buyer and seller leads to what’s called market collapse. The seller’s information is trapped because no market can form around this data because no buyer can assess the quality (see Nobel Prize winner George Akerlof’s work for more on market collapse).
The only solution right now for sellers of predictive information is to signal with brands or third parties in some way: “I used to work at Goldman Sachs, over the last two years I have predicted Google’s daily stock price with 70% accuracy, and I’m willing to sell you my next prediction for $10,000. You can see I also went Harvard if you add me to your professional network on LinkedIn.”
Now you have to put some faith in LinkedIn, Harvard and Goldman Sachs. This kind of third party signaling with brands might work in some small way but it is not nearly enough to give a potential buyer of the prediction any kind of peace of mind when it comes to assessing the credibility of bold claims such as being able to predict Google with 70% accuracy.
But maybe there is a technological solution for quality assessment. Maybe there is a third party that everyone trusts (or rather no one needs to trust). That technology is blockchain. Imagine a claim like this:
“Over the last two years I have predicted Google’s daily stock price with 70% accuracy, you can verify this because I sent every prediction to the Ethereum blockchain ahead of time and no one can go back in time and change a blockchain. I also placed a large stake of $100,000 on these predictions on the blockchain. I’m willing to sell you my next prediction for $10,000 in cryptocurrency, and if you are unhappy with it for whatever reason, you can destroy my stake by destroying $10 worth of my stake for every $1 you destroy of your own cryptocurrency.”
Today we are excited to announce a new protocol: an unstoppable, peer to peer, decentralized data marketplace for predictions. It will let individuals sell predictions about the world that everyone will trust, and make promises about their future performance that anyone can enforce. Using the blockchain, we have a solution to the market collapse problem. We are making a radical new kind of information market, and it’s called Erasure.
The Erasure Protocol
“The past was erased, the erasure was forgotten, the lie became the truth.”
― George Orwell, 1984
Erasure is a new decentralized data marketplace. It allows anybody to upload predictions, stake them with cryptocurrency, build a track record that everyone can verify, and earn money.
Commit to every prediction ahead of time
In 2013, an open source project called Proof of Existence was launched. Proof of Existence allowed a user to timestamp a document using the Bitcoin blockchain. You could add your document to the Bitcoin blockchain, and thereby prove that the document existed at that time. Rob Wile from Business Insider wrote that Proof Of Existence is “perhaps the most straightforward example of a post-Bitcoin service using Satoshi’s blockchain”.
On Erasure, you need to timestamp every prediction you upload, much like Proof of Existence.
Erasure users will submit thousands of predictions; for example, a quant model for global equities might produce 10,000 stock predictions per day. It is too expensive to add all this data to Ethereum, so instead the data is uploaded to IPFS. IPFS is a decentralized file storage network, where Erasure stores each data feed. Erasure sends the address of these stored files, called a hash, to the Ethereum blockchain to get timestamped. By submitting just the the IPFS hash to Ethereum, the whole file of predictions gets a timestamp. (Think of the hash as like a URL pointing to a file containing all of the predictions on IPFS.)
This process does one very important thing, it commits your prediction to the Ethereum blockchain at a certain time so you cannot alter your prediction later. If you submitted two years of daily Google stock price predictions to Erasure and you were 70% accurate, the whole world would be able to see that and check it for themselves (say against Yahoo! Finance data). Everyone would agree that your historical predictions occurred ahead of time, because all of your predictions are timestamped on Erasure. Now people can believe your claims because they don’t need to trust you.
Reveal predictions later
While exposing your predictions to the world would earn you credibility, it also means people could just use them without paying you. You somehow need to conceal your predictions while simultaneously committing to them. So on Erasure, every prediction is submitted using a commit and reveal scheme. You can commit to your prediction now and get it timestamped but you only have to reveal what your prediction was later when it is no longer valuable.
Because everyone’s historical predictions are revealed but everyone’s most recent predictions are concealed, anyone can verify how good everyone’s prediction feed has been historically without ever being able to use the predictions. On Erasure, you can prove the quality of your predictions without giving anything valuable away.
On Erasure, buyers can assess which prediction feeds have the highest quality and choose to buy the feed, which gives them a special key to see all the most recent predictions before anyone else — and therefore can trade on that information before anyone else.
“The usual touchstone of whether what someone asserts is mere persuasion or at least a subjective conviction, i.e., firm belief, is betting. Often someone pronounces his propositions with such confident and inflexible defiance that he seems to have entirely laid aside all concern for error. A bet disconcerts him. Sometimes he reveals that he is persuaded enough for one ducat but not for ten. For he would happily bet one, but at 10 he suddenly becomes aware of what he had not previously noticed, namely that it is quite possible that he has erred.”
— Immanuel Kant, Critique of Pure Reason
Suppose you’re a hedge fund manager and you find a prediction feed on Erasure which really does have 70% accuracy predicting the daily price of Google. Are you comfortable buying the feed? Because the feed is on Erasure, you know for sure that every prediction was made ahead of time with a tamper proof timestamp to prove it but you don’t know whether what you’re seeing is a result of luck.
Maybe the creator of this prediction feed has created 100,000 other feeds and you’re looking at the only one that got lucky. Maybe 50,000 of his other feeds had less than 50% accuracy on Google, and only a handful had more than 60% accuracy. He’s trying to sell you his feed which has 70% accuracy but that 70% accuracy is a result of pure luck. You don’t want to buy a prediction feed that simply got lucky with its historical predictions; the luck is very unlikely to continue.
Erasure needs some kind of resistence to this problem to give buyers more quality assurance. To prevent the 100,000 prediction feed exploit, a kind of Sybil attack, Erasure uses staking. If every prediction on Erasure required some cryptocurrency to be staked at risk, then sellers of predictions couldn’t afford to make 100,000 prediction feeds in order to try to get lucky. So on Erasure, every prediction is staked with cryptocurrency. By staking cryptocurrency, the seller of a prediction feed puts something at risk if things go wrong (“skin in the game”) producing incentive alignment and higher quality feeds.
Allow buyers discretionary recourse
You’re a hedge fund manager, you find the Google feed with 70% accuracy and you see that a $10,000 stake was made on the feed at the time that the first predictions were submitted. You are more confident than ever about it’s quality. But you still can’t be sure that the 70% accuracy will continue, and really no one can be sure about that including the seller but it does appear that this prediction feed has the potential to be worth trading on. But you do want some kind of recourse if things go wrong. If the feed accuracy ends up being 50% you’re going to be upset you bought it. You’ll lose money paying the seller for the prediction, and you’ll lose money trading as well. So Erasure has something called “griefing”. Griefing lets you destroy the seller’s stake by destroying some of your own.
When a seller stakes their prediction feed on Erasure, they choose what’s called a “griefing factor”. The griefing factor is the degree to which the seller is happy for the buyer to be able to destroy their stake. A griefing factor of “1:10” means that for every $1 a buyer destroys of their own money, $10 of the seller’s stake will get destroyed. So on Erasure, when a buyer starts to buy a live prediction feed, the seller gives them a special right to destroy their stake according to the griefing factor.
Especially agressive buyers might be especially agressive with griefing. For example, a buyer who buys a feed with 70% historical predictive accuracy on Google might start destroying the seller’s stake if the accuracy drops below 65% and other buyers might be more lenient. As the marketplace for predictions develops, sellers will adjust their griefing factors to signal quality or fear. Some sellers might give griefing factors of 1:50 to express confidence in the quality of their predictions others might give griefing factors of 1:1 afraid of meeting aggressive buyers.
But why would a buyer grief at all if he loses money in order to harm the seller? Isn’t it irrational to harm yourself in order to harm others? No, it sometimes is rational especially in the setting of Erasure which is a repeated game. Buyers are not just buying one prediction they are subscribing to a feed of ongoing predictions. If a seller gives a bad prediction, it may indeed be rational for the buyer to grief the seller if doing so increases the likelihood that the seller will improve the quality of their subsequent predictions. Griefing on Erasure is similar to an Ultimatum Game, which in a repeated setting tends to result in fair behavior.
Griefing is the final recourse for buyers on Erasure, and the final element to produce aligned behavior to make it possible for a market for predictions to take form.
Numerai and Erasure
Erasure uses no speculative technology; it is something that can work today. In fact, Numerai has been using a special case of the Erasure protocol for over a year.
Just over one year ago, Numerai released the Numeraire token (NMR), which is used by data scientists to stake stock market predictions submitted to Numerai’s hedge fund. Since then, NMR has become one of the most used utility tokens on Ethereum. In June, there were more NMR stakes in number and dollar value than any other Ethereum application including CryptoKitties.
NMR also worked exactly as designed. The predictions staked with the NMR token dramatically outperform predictions not staked with NMR. In our backtests, predictions staked with NMR have a Sharpe of 2.09 vs Sharpe of 1.66 for unstaked predictions. Skin in the game works.
But NMR has a problem; it is a token designed just for Numerai. Only Numerai data scientists use NMR for staking, only Numerai can buy predictions and destroy stakes, and all prediction datasets need to be uploaded to Numerai’s servers. In a word, NMR is centralized — and that’s bad because it limits NMRs potential. But today we are announcing that we are decentralizing NMR by making NMR the native token of Erasure.
NMR will become the staking token for Erasure, a decentralized marketplace for predictions where any individual can sell any prediction feed to any hedge fund.
Right now, NMR is the token for Numerai. With Erasure, NMR becomes the token for the entire hedge fund industry.
For NMR holders, Erasure is a whole new use case for NMR in a bigger, braver, decentralized future. For the hedge fund industry, Erasure is an unstoppable decentralized data marketplace to buy high quality, trustable new data feeds.
Erasure is a way to free all the trapped information, help it reach the market and improve the world but Erasure could be used for anything from sports predictions to crypto predictions to weather forecasts to whistleblowing. There are no restrictions on the kinds of data feeds you can submit, and no usage can be censored or stopped.
Marketplaces are two-sided but Erasure starts with a big buyer on the buy side (Numerai) and sellers on the sell side (Numerai’s users who have already staked over $1 million worth of NMR in thousands of different stakes across billions of predictions).
But as prediction feeds build verifiable track records on Erasure, traditional hedge funds will have no choice but to participate in the marketplace as well. If quants on Quantopian build long track records predicting the S&P 500 with high returns and low volatility and staked these predictions with large amounts of NMR, traditional hedge funds like Two Sigma will have no choice but to take notice.
Demand from hedge fund’s will lead to more sellers and vice versa. Hedge funds will compete aggressively to buy predictions and over time Erasure will become the highest paying way to sell predictions. So the best, most unusual data sources will be drawn to sell on Erasure thereby attracting even more hedge fund buyers.
We are looking for people who understand that the highest leverage way to make a difference in the world is to work on the invisible things — the systems behind the system. We have some open job positions for people who see that finance is the system behind it all, and that rebooting finance will have a huge impact on every other industry as well. If you’re interested in helping to build Erasure email email@example.com.
If you’re interested in contributing to the project as a developer, data scientist or hedge fund, email us on firstname.lastname@example.org, join our Telegram channel @NMR_Official, or sign up to receive email updates on erasure.xxx.