5 reasons you should check the Numerai Competitions.
And 5 reasons you should skip them.
I’ve been competing in the Numerai Quant competitions ( there are 2 ) for the past 3 years and thought it could be interesting to write a follow up to my previous two posts as a way to catch up…
Previous posts you might like:- Numerai walkthrough: Quantitative Analysis & Machine learning for fun and profit. The practical parts are now outdated, but the introduction explanation is still relevant. - Practical Keras Simple regression for the Numerai Tournament An up to date (late 2020) technical look at setting up a Neural Network for the tournament.
What is it again ?
In a nutshell Numerai is a Hedge Fund with crowdsourced elements in the form of two quant competitions, the main tournament ( numerai ) and the newer ( signals ). The main tournament consists of an obfuscated dataset that represents years of market and financial information and you are tasked with predicting a future date. Signals in its current form is closer to traditional stock picking/portfolio construction in that you submit a list of tickers with short/long ranks and are scored on returns over a shorter weekly period.
You get paid in a volatile cryptocurrency called Numeraire (NMR), but only if you stake and lock some NMR beforehand, so it is strictly pay to play, your stake can also decrease ( get burned ) if your predictions are wrong, so you can also lose money.
A few pros and cons based on my experience :
In Favor.
1.There’s still nothing like it. The longevity of the project (more than 3 years at this point) is perhaps the best indicator of the success of Numerai, starting and maintaining a hedge fund is not easy, double so if you add the challenges of exposing some of the decision making to Data Scientists and Quants along with a cryptocurrency to maintain.
2. The problem while complex is interesting. Financial markets are unpredictable creatures since they encapsulate the economic environment along with capital markets, human behavior and decision making. Predicting them is a career and interest of many, and here you can put your theories and research into some practice.
3.You can learn practical Data Science and/or Quantitative Finance. While the learning curve is steep and you are mostly on your own, it’s probably the next best thing to being a junior quant at some bank, at least for independent self learners.
4. You can make some money. How much is hard to say since you need to provide your own capital to start, there’s also the ever changing payout rules and NMR volatility, so don’t quit your job just yet.
5. The community. There’s a varied, knowledgeable and helpful group of users that for the most part make up for the lack of support and sometimes even contribute code and tips for the competition.
Against
1.Opinionated approach. This is not wallstreetbets where you pick a long /short or derivative position and yolo your paycheck, while the metrics your are scored on change they usually represent some excess return, so if the S&P gave you 10% this year you are scored or penalized for doing better/worse than this 10%, if you think returns are returns no matter how you got them this will probably drive you crazy.
2.Is it a fair competition ? Hard to say, scoring is a black box so you have to trust Numerai in that respect, people working for Numerai can also stake and submit models, some of the highest staking and performing models are from Numerai’s own staff, the same staff that scores your models.
3. Users can be seen as treated poorly. There is a long list of negatives that users are subjected to, from abysmal documentation and support to draconian tournament rules and decisions, some of this comes from having a crowdsourced, sybil resistant competition, but can feel mean and needless at times.
4. Changes and difficulty. Some weeks you will have to work the whole week or weekend because they dropped breaking changes at the last minute or had some technical problem, there’s also diminishing returns with an ever higher risk profile and you are given no protection from black swan events.
5. Founder, team and community issues. No easy way to say this and your experience may well vary… political, tone deaf and misleading statements are common all around, so if you were looking for a less toxic place than Wall Street this might not be it.
The competition: I believe it is still early days for these type of projects, here's a couple that are comparable...Quantopian: Recently closed it's doors but had fairly deep tooling (research/backtesting/data) but a weird profit sharing scheme, you can find some of the tools here : ziplineKaggle: The original DS competition, but no Quant or finance focus.Haven't tried the following so can't vouch either way but worth looking into...QuantConnectQuantiacs
Do I recommend it ?
Yes if you are curious about the intersection of Quant Finance and Data Science and want to dip your toes (mind the steep learning curve and the rough edges though). As a way to make money and as a startup/project in general, you will have to draw your own conclusion, I hope this helps.
Thanks for reading !
