The CoE Numerai Newsletter- Edition #13
Aug 1st "Fireside Edition" Q3 2022
CoE Sponsored Numerai Meetup in London 7-16-22. 360 shot taken by ai_joe
London CoE Sponsored Meetup and Hackathon Was A Success!
The CoE conducted its first community meetup in London on July 16th and it went great! We want to thank ai_joe for all the hard work of finding the location, setting it up, getting the food and making this entire process completely painless. We had around 40 attendees! We also want to thank all the speakers for their presentations. We look forward to locking down our next location soon!
Meetup Goals
Demystifying Numerai - many top data scientists out there are just not too sure about the crypto elements and/or the tournament format. Speaking with and listening to long-term participants can be an effective way to build trust. Some participants learned to trust Numerai after watching Office Hours with Arbitrage.
Knowledge sharing and brainstorming - although we may not share all our secret sauce, meetups are great opportunities to bounce ideas off each other and may lead to new models with crazily high TC.
Face-to-face discussions between Numeratis and Numerai team - a lot more than just asking the team “wen scores” in person, these events can provide a quick and direct feedback loop for both participants and the Numerai team.
We have created a github link with access to all presentations with video files to come soon
https://github.com/councilofelders/meetups
If you signed up on site to receive a POAP though email make sure you complete the process of claiming to a non-custodial wallet such as metamask. If you claimed direct to a wallet you are on the list to claim a free Numerai NFTee soon!
CoE Election Results Are In
In late June, jrAi decided to step down from the CoE. A Numerbay poll was created for the nominees.
bor (bor1)
jefferythewind
yifanxie (yxbot)
It was an extremely close race with bor taking the lead to secure the open position.
We are very excited to welcome bor to the CoE and look forward to the future
Here is a comment from bor during the voting of things he is planning on working on:
The transition to TC staking is probably the largest challenge we have right now as a modeling community. I would be interested in getting people sponsored that have ideas how to help people in tackling the TC challenge that with sponsoring from the CoE or with the process of retroactive reimbursements.
In general I am interested in getting people to share more about their models, and I tend to be quite forthcoming when people ask myself (if only because I am running something so cumbersome and tailor-made, it’s not easy to copy). And there is a smallish group as well that shares things in the datascience channel. If I can prod more people into sharing, all the better.
-bor
Official Numerai Announcements
On July 8th Numerai announced that Signals Staking on TC will be enabled for round 324 on Jul 9. Signals users will be able to stake 0x (default), 0.5x, 1x and 2x TC multipliers. Staking on MMC will be disabled starting with round 324 - any prior MMC multiplier settings will be ignored.
ONE MILLION NMR
July 22nd 2022: > 1M NMR is now staked on both tournaments
Numerai Supreme
On July 6th the first Numerai Supreme trades hit the global markets
Numerai Supreme is a more aggressive fund and uses the same signals as the first fund but with different optimizer settings.
They have so far raised 70M(signatures to come to lock these funds in the coming weeks)
Richard Craib Moves to NYC
On July 28th 2022 Richard Craib posted to Twitter that he has decided to move to New York to deepen Numerai’s relationship with global capital. He continued in a thread on the reasoning behind the decision.
Numerai has just passed $100m in AUM and we’re just 2 months away from a 3 year track record (http://numerai.fund). These are important milestones in the left-brain, dispassionate, beautiful world of quantitative finance.
Investors in quant funds don’t have good reasons to rush to invest in new funds before they’ve built a track record. This is why they start small and stay small until they prove themselves. For example, Voleon was <$50m from 2008-2013 and then $6bn by 2019.
Since Numerai already has the best data science talent (think ~40 Kaggle Grandmasters and >4000 models staking an all time high of >1 million NMR), there isn’t much better use of my time than to be in NYC, the quant capital of the world capturing capital for Numerai’s funds.
We wish him the best and plan to conduct a CoE sponsored meetup in NYC soon!
CoE Wallet Transactions
July 2nd -21 NMR Newsletter
July 3rd -34 NMR Numerbay
July 4th -180 NMR CoE Travel Stipend
July 19th -66 NMR Numerbay
July 19th -50 NMR London Meetup setup costs
July 19th -0.14 ETH Gas Reimbursement
July 23rd V3 pool fee collection
Current Holdings:
2,013.11346049 NMR
5.70347215 WETH
1.74446409 ETH
Numerbay Updates
2022-06-19 — 2022-07-03:
Added TC metrics for Signals listings
Updated community dashboards links
Other bug fixes and improvements (See list of closed issues)
2022-07-03 — 2022-07-17:
Added support for change of auto-submit model after order confirmation
Added email notifications for failed auto-submissions
Fixed bug causing decryption failures
Other bug fixes and improvements (See list of closed issues)
2022-07-17 — 2022-07-31:
Added seller upload reminder button (rate limit once per 30 minutes)
Added webhook trigger on round open and muted when round is closed (docs)
Added more sort options and filters
Other bug fixes and improvements (See list of closed issues)
Memes of the Month
by correlator
by keno
Numerai Fireside Chat Q3
Numerai conducted its quarterly fireside on July 7th 2022
Here is a recap of the fireside below with answers to slido questions:
Numerai Supreme is now trading
Daily trading with Automated submissions coming soon
Worked on the Stock Universe. They are going to increase the size of the universe. The number of stocks in signals will increase by about 10-15%. Also hundreds will be added to classic. Corr has been seen to increase in testing as well as sharpe.
They are currently researching new targets. There is a target to use to score corr but there are many other targets. We should consider learning on different targets. They are researching adding more targets. They have a lot of ideas.
Numerai thanks degerhan for finding a critical bug after an ETH hard fork.
Slido Questions
How do you want to incentivize participation in daily rounds? If you have full participation in the daily rounds they think we should be paid more. We are working on that solution. As for payout factor there could be higher payout for early migration. They want to help with automation first.
Do you believe TC is in its final form? If not what do you want to change? There's actually subtle things we've changed to the Numerai One optimizer so we actually slightly increased some of our positions. This would change TC but they have not made official changes yet. There is really multiple TCs depending on fund and time horizons so we can really define a new TC adding them together. Maybe we can do something like this soon.
Are there plans to remove core staking? Proposal different payout factors for core and tc? Don't think we have a plan to remove core right now although I think TC is definitely the superior scoring system in terms of aligning the tournament and the hedge fund.
Any plans for numeri etf? So there's a couple things you can't really do very easily with etfs. One is charge performance related fees so you just have to charge management fees. You can't do things like shorting easily as well. We got close and then decided not to do it. We can make more by not making it fully public like an ETF. We care about making money a lot. We cant to have a very durable revenue stream.
How much measurable improvement in the hedge fund has occurred since tc has been introduced? Well we don't ask for people to submit on test data anymore so we dont know for sure but by definition with rewards and scoring its helping push models in the right direction.
Any stance on if v4 data performs better than v3? I don't know for sure I haven't checked lately but I mean obviously we wouldn't have released v4 without testing it. Internal models on v4 performed much better than v3 which is why we released it. We dont have a great way of tracking users who are using the different versions of the data.
In terms of capacity, do you have a growth plan beyond your current target? Could other strategies or even capacity constraints be crowdsourced? When we do test our models on half a billion dollars of assumed aum in a back test we do really well. If you're good at three-month return predictions you will have higher capacity than if you're good at one month so being good at these longer horizons means we can hold stocks longer and therefore have higher capacity. We might start showing three-month corr on the website or three-month tc and that will definitely help capacity too. The other thing is trading every day
Wen account level staking? SOON
Wen 500 million AUM? SOON. Things are going to get easier to get AUM as our 3 year track record is hitting time
Is there a plan for utility for NFTees? Yes there is a secret plan
When will tc staking be available in signals? Should be coming out this next week
Can you tell us about Numerai Supreme? SEE ABOVE
I heard about the upcoming changes on compute will you keep it AWS only or will you also consider other clouds and a provider or provide a general extendable solution? We have the most expertise in AWS. We want to actually make it simpler. We are calling it “compute lite.” It should be extendable to all clouds but right now we are going to focus on AWS
Wen benchmark models? We announced this during Numercon and this is something we really want to get to you all but have not started working on it yet but we have it still on the plans to complete. Its happening.
Will there be an update on the data set in new features, better targets for tc, and daily data data yes updates on data set coming soon? YES
Is there any concrete plan from the team to engage the wider data science and machine learning community? No we don’t have any marketing plans besides what we already have which is just having the highest paying hardest best data science tournament on the web.
Is Numercon 2023 happening? Haven't thought about it but Japan could be cool.
Hey Anson bro when decentralization? I did give a teaser in Numercon. I'm still excited about that vision I think I'm still trying to build towards that vision.
Regarding daily submissions, in the future is more than two hour window from data download to prediction deadline possible? This is something I've been working on a lot which is sort of like mapping out the the timelines of when we get data from our data providers how long it takes us to construct a data set for the tournament how long it takes all the tournament models to run and then how long it takes for us to build the meta model signal optimization and then trade execution. I think the truth is if we want to have the freshest metamodel signal possible the tournament is going to have to be same day turnaround style.
The cost of data computation is bigger than the average salary of third world countries. How do you plan to improve those places using data science? One way is using more compact file formats. If you're not using parquet like please look into using parquet.
Any plans for new tournaments for example daily submissions or different prediction horizons? YES and YES but we did talk about the possibility of using different targets so maybe not fully new tournaments but the possibility of scoring on different timelines and targets etc within the main tournament. We are still thinking about these different things
Any chance compute would feature automatic uploading of predictions to buyers at the Numerbay platform? I think the mandate for us working on the the data api and the submission api is for us just to build a general platform to support you guys to build whatever you want. I think Numerbay has the ability to build that system.
Wen synthetic data? There is a lot of great ideas in the forum posts. Its something ive thought a lot about. We actually haven't done much more internally on the subject so far. I think we will have more targets before synthetic data but its on our agenda.
Wen erasure revival? NEVER
Can we get more model slots please? Yeah I think so. We can get more model slots. We will work on that
Will there ever only be six thousand nftees or could there be more minted in the future? There will only ever be six thousand
Is the data right for starting a similar tournament for crypto? I've often wanted to make a crypto signals not a crypto Numerai but a crypto numerai signals because it's incredibly easy to do. We would probably be like one of the best people in the world very quickly at creating a market neutral crypto hedge fund. We should eventually do that. We are not set up yet to do that. Eventually we should.
How do we measure success? Are we successful? Is this good enough? A lot of hedge funds hide their performance one of the reasons we show our performance against aqr is that they actually show their performance on their website. A lot of other hedge funds dont do that. Basically what I want is a leaderboard, a global leaderboard of hedge funds and then I think we would be very high on it. Our internal goals are a little bit higher than our realized.
Last question:
Hey richard bro, nj sis and team love you all.
Disclaimer: This is not an official Numerai newsletter. It is sponsored by the Numerai CoE, a decentralized autonomous organization. Every effort is made to provide accurate and complete information but there is no claims, promises or guarantees about the accuracy of the contents.