I wanted to ask/discuss a little bit about how Numerai compares to previous attempts to crowdsource trading strategy, Quantopian (and their demise ) comes to my mind when thinking about this. For those who don’t know quantopian was a platform to develop trading strategies, it provided a research interface, stock market data, and some interesting tools for backtesting ( zipline, alphalens, pyfolio ) the company was initially born with the idea of crowdsourcing strategies and reward the authors of the most original and profitable strategy. The company shut down last year (acquired by Robinhood I think) they were ultimately not able to monetize hundreds of strategies and most strategies were just overfitting.
It seems to be hard to capitalize crowdsourced strategies, quantopian guys tried for years and they couldn’t really crack it in the end.
I’d like to hear what your opinions on this, I’m not suggesting the exact same thing will happen to numerai, but If would be interesting to discuss the differences and similarities between the two approaches and perhaps identify potential pitfalls and learn from the past mistakes ( without ovefitting of course)
“Those whose programs survived a meticulous screening could have them included in a hedge fund Quantopian ran, and get a cut of their strategies’ profits.” https://www.bloomberg.com/news/articles/2020-12-16/quant-trading-platform-quantopian-closes-down, it seems like the Numerai staking system may give them a leg up, in that your model must perform for long time periods, not just on the validation enough for them to pay you some and include your model in their trades. Putting money on your submissions means you wont overfit them(if you do, you will lose NMR and stop submitting after some time) which seems like it may have been an issue in quantopia as you mentioned
Interesting, I asked myself the same question. I “played” with Quantopian for many years, when it was alive. I have noticed few key differences with numerai, although I cannot say they are enough to guarantee a success. I do not want to start a long analysis, so I just want say the main difference is the data: Quantopian didn’t offer the same high quality data as numerai and in a difficult task such as predicting financial market, this is a key point.
I recently wrote this on twitter, related to this subject.
i tried kaggle and quantopian, but stuck it out at numerai because i didn’t have to fork over code or write about my models. funny thing is though i’m eager to share a lot of my secrets. disclosure being my choice, and immediate earn/burn feedback made all the difference for me
But you don’t have to write about your model or fork any code on kaggle - which is advised if you want to learn.
I come across kaggle/numerai comparisons on multiple occasions, which is nonsensical to me since the two platforms offer different service.
