Hedge funds and investment firms thrive on information asymmetry. Having more information (and therefore, insights) than the rest. But how does this affect decentralized hedge funds and traditional hedge funds? In other words, how do you reconcile the fact that “value” is captured because of information asymmetry and decentralized hedge funds will greatly reduce information asymmetry?
Do decentralized hedge funds make sense? In fact, does it ever make sense to have decentralization when there is information asymmetry?
Machine learning competitions are susceptible to intentional overfitting. Numerai proposes Numeraire, a new cryptographic token that can be used in a novel auction mechanism to make overfitting economically irrational. The auction mechanism leads to equilibrium bidding behavior that reveals rational data scientists’ confidence in their models’ ability to perform well on new data. The auction mechanism also yields natural arguments for the economic value of a Numeraire token. […]
We propose a new system for data scientists to communicate their beliefs about the quality of their models. Data scientists will compete in the new tournament by staking a new crypto-token, Numeraire (NMR), on their predictions. The auction mechanism for resolving these stakes will reward correct predictions of a model’s ability to perform well on new data. With Numeraire, data scientists will now be able to express their confidence in their models’ live performance. Their expressions of confidence help us to emphasize the right models and improve the performance of our hedge fund.
If I were a data scientist, and I knew that my model performs well on new data, why would I share this publicly with Numerai to profit THEIR hedge fund? Why wouldn’t I use my model to make profits for myself? I am pretty sure the economic incentives I earn from submitting my model and getting the Numerai token are lower than the profits I can earn independently.
If my thinking is flawed, I’d LOVE for someone to help me understand where my reasoning is incorrect.
I’m wondering if the participants may value peer review (and/or competition) over profit? That would make it make sense
As an aside, I always thought of hedge funds as having more of a regulatory asymmetry than an informational; investors have to be accredited, and then the fund can enter/exit positions way faster than mutual funds, they can day trade, employ riskier strategies, use stock tips and employ complex derivative tools and leverage. Don’t know this from experience though, just reading over the years.
I think you’re right about the fact that hedge funds have more of a regulatory asymmetry advantage (for example HFTs can trade as much as they like and they pay for faster connections to give them an edge). However, the original debate was sparked about the information asymmetry advantage VCs use, and they also claim better results:
Importantly, venture capital is the lifeblood of major, fundamental innovation, the key to substantive economic growth. Think about the big business success stories of the past dozen years – Microsoft, Alphabet, Facebook, etc. All were spawned by venture capital investment . . . and there are many more that have not yet gone public and are still operating on their own as well as ventures that were acquired by established companies who recognized and valued their growth potential. As a result, aggregate returns for the asset class [venture capital] have historically been outstanding, and there’s no sign that will change anytime soon.
Disclaimer: I haven’t actually done much research myself into whether VC funds perform better overall, but I have heard this mentioned a few times
In order for the data scientist to profit from their models, we need to assume that the data scientists face no restrictions which can prevent them from acting on their models. In particular, the data scientists:
Can independently identify relevant patterns (even if they did not engage with Numerai).
Numerai provides the scientists with the data which they subsequently use to develop models. The data is not available freely. Further, hedge funds can source models from different data scientists, and can identify patterns that are not known to data scientists independently.
Are aware of how to use their models to make trading decisions
The data generally lack labels, so the scientists are only tasked with identifying patterns between unknown variables - they do not know what these variables mean, so cannot act on them.
Have access to sufficient capital
A hedge funds would perhaps have access to more capital compared to the average data scientist
I thought that the interview between Richard Craib (Founder of Numerai) with Laura Shin provided a lot more clarity on how Numerai operates, especially when compared to their whitepaper:
"…usually it’s a really bad idea to give away your data because it’s like you’re giving away your edge, and so the trick with Numerai was that we gave away our data but in this totally obfuscated form. So, people could do machine learning on it, but they had no idea what it represented. So, if you download the Numerai data, it’s just 50 features. It’s feature 1, feature 2, feature 3.
You have no idea what feature 1 means or feature 2 means, and you’re just modeling a binary target, 1 or 0, so it’s this huge grid of numbers between 1 and 0, and it turns out if you…you can still do machine learning on that. You can still find patterns even though you don’t know what you’re modeling, and so this kind of aligns incentives, like the problem is, with sharing data, is that people could run off and start their own hedge funds. By doing it this way, you’re not only making it impossible to steal the data, but you’re making this incentive to, once you build a model, you should submit it, submit the predictions to Numerai."
P.S. I believe Numerai tokens are governance tokens which offer the data scientists opportunity to submit models in the future - they are used to control the quality of the models being submitted, not to incentivise scientists to participate in the platform. If a model outperforms the rest, the model owners get compensated in some asset which holds value (the compensation has been denominated in both fiat and crypto in the past), and additionally in Numerai tokens (which offers them the opportunity to submit models in the future, and therfore ability to access future returns)
Additional note:_ Richard Craib has recently founded another decentralised data marketplace called Erasure, where data scientists can develop models based on data they have sourced themselves. This partly lifts restriction 1 and 2. Hedge funds can still have access to more information collectively - both economic and skills based. This was partly validated by Seth Stephens-Davidowitz, a data scientist at who (without success) tried to use exploit data freely available on the internet to generate alpha - he summarises his findings in the book, “Everybody Lies”. I do feel you there is enough to make this an interesting experiment to follow.
You bring up an interesting point: asymmetry can be regulatory vs informational. To play devil’s advocate, are accredited investors really “regulated” in the literal sense? I’ve looked into the accreditation process over the years, and the requirements per the SEC are for the most part financial:
An accredited investor, in the context of a natural person,
includes anyone who:
- earned income that exceeded $200,000 (or $300,000
together with a spouse) in each of the prior two years,
and reasonably expects the same for the
current year, OR
- has a net worth over $1 million, either alone or
together with a spouse (excluding the value of the
person’s primary residence).
This is one of (many) frustrations that have led to my increased interest in alternative systems and decentralized structures. The amount of money required to be accredited is massive (note that the minimum net worth can’t include primary residence real estate). There’s no institutional knowledge or experience, other than going through a program or boot camp, that is part of the accreditation process. Money does not necessarily = expertise, but the accreditation process seems to imply as such.
Perhaps this is another reason why decentralized hedge funds do make sense, if we look at the “information” investors/hedge funds are providing more as offshoots of “access.” It allows information flow that isn’t reliant on substantial pre-existing revenue (i.e. the net worth requirements, which can be met via any means, not simply investing successes) to have equal value.