Hacker News Reader

Comment by Animats | Hacker News Reader

Numerai combines two things - a prediction competition, and yet another complicated and obfuscated crypto coin scheme.

The prediction competition is like the Netflix Prize [1]. They give you a data set and a prediction goal, you submit an algorithm, and there's a winner. OK, although if you can do well at that, you're wasting your time giving your results to Numerai. You're competing for a total payout of $15K/month.

Then, for no particularly good reason, this is combined with some smart contract / crypto coin scheme. Other than for PR purposes, it's not clear what the gain is. Everybody is dealing with Numerai; it's not a many-to-many system. It's more like airline reward points.

The hedge fund itself doesn't use Etherium; it's completely traditional.[2]

[1] https://en.wikipedia.org/wiki/Netflix_Prize [2] https://www.forbes.com/sites/laurashin/2017/02/21/this-is-th...

View on HN

You don't submit an algorithm, you only submit your predictions. Your algorithm to generate your model predictions is not known to Numerai.

The data you work with from Nunerai is obfuscated, so you can't take it directly to the market.

It is an interesting no-risk way to evaluate models/algorithms. A lot of people are there to learn on a tough problem.

Edit: The reason for the coin is to allow data scientists to give confidence information about their models. There is an additional payout in USD associated with NMR staking (currently an extra ~$4000/mo for the most consistent models).

And you find that a compelling reason to tokenize it? I don't.

Very compelling. Without staking the hedge fund doesn't have any knowledge of how confident you are in your estimates. If you can put the value of a coin up against your confidence estimates then you are giving that information in a very real and binding way.

Do you have another way data scientists can give estimates of their model's ability to do better than chance and be held accountable for that estimate? Additional prize pool based on giving good reliability estimates seems like an excellent way to do it.

The tokens are not the value of the numerai fund itself. They are merely a tool for the staking process.

Yes...I remember a few years ago, with one of the algorithm testing sites, a group basically just submitted 20-30 pseudo-random predictions. One of them ended up winning, but it wasn't like the team actually had a useful algorithm, they were just playing the system because betting was free.

why not stake with bitcoin or ether?

Because they can't mint and award it to users. NMR is only earned by performing well.

theyre listing it on poloniex FYI

If anyone is interested. Software Engineering Daily did a podcast about Numerai not too long ago, with one of their employees.

"Then, for no particularly good reason, this is combined with some smart contract / crypto coin scheme. Other than for PR purposes, it's not clear what the gain is."

Oh, it is entirely clear what it gains.

Cryto currency backed equity is just free money that you don't have to pay back if things go south.

There are basically zero protections for the "investors". Numerai could just walk away with the money and they'd probably get away with it.

Why wouldn't you turn down free money that the market is throwing at you?

> Then, for no particularly good reason, this is combined with some smart contract / crypto coin scheme

This single line, appended to pretty much anything, describes the latest crowdfunding fad of cryptocurrency or token ICOs.

> Although if you can do well at that, you're wasting your time giving the results to Numerai.

This is the fundamental break down in proposals like this (Numerai, Quantopian, etc).

For anyone who can consistently generate alpha, why bother with a crowdsourced model? They can just go join any one of the hedge funds that will pay half to several million each year in compensation for that acumen.

Numerai doesn't have the model, just predictions. The users don't have the real data to be able to trade their own predictions.

Sure. But that doesn't meaningfully negate my point. In fact it strengthens it - if you can generate alpha on the strange "homomorphically encrypted" data they've got, you can do it in the actual market with transparent data and more control far more profitably.

I'm not saying someone is going to do well and leave with the data. I'm saying anyone who can do well should probably just participate in the real market.

And as another point, Numerai lacks entire categories of data outside the conventional "core financial" that is useful for finding alpha.

> if you can generate alpha on the strange "homomorphically encrypted" data they've got, you can do it in the actual market with transparent data and more control far more profitably.

Sure. Now go get "transparent data"...oh wait, that's also time-consuming (and much less interesting). That's why, if one doesn't have to search for tradable data, it might be interesting to validate a signal/strategy on data that someone with stake says is tradable.

Agreed that if one starts making numerai money with numerai data, one should definitely go get some transparent data and start trading it. Trouble is, now one doesn't know what transparent data one has actually learned, so it's hard to know what to go and spend time on trading.

At that point, though, (I guess this is the theory, at least), numerai presumably know that one has a good signal, and one also knows one has a good signal for something, so both parties know that one's signals are valuable. That should be enough to get one a job at a place that does have good, transparent data.

Assuming that's a job which you want. Some people want to screw around with a project for a while and this is a good opportunity to test your skills on something interesting and real.

Worst case scenario you get some "free" money and some other people you don't know get rather more money out of it.

2 counterpoints: 1. buying data is expensive 2. running a fund is a completely different skill than building predictive models (back-office, compliance, etc.)

I would guess the crypto scheme is to cash out with ICO - those have been raising ridiculous amounts of money lately, for projects that sound like a joke.

My Mom suggested I start a coin where each coin is backed/tethered to one of the thousands of Beanie Babies in her closet. I told her the majority of cryptocurrency traders are too young to get the joke so it wouldn't get much traction. Then again Dogecoin started as an obscure joke and it now has $300Million+ market cap.

There is no ICO. Coin goes directly to competing data scientists. And that coin can be used to increase earnings of the data scientist depending on the consistency of their model.

How is the coin value jumpstarted, then? Who do the scientists sell the tokens to for real money?

People have been going to the slack asking for data scientist to sell them their tokens. Bidding was up to around $20/coin.

The main point, though, is that data scientists don't have to sell the coin. They can stake coin against their models to earn additional USD. If their model performs better than chance, then they get their NMR back to stake the next round. If they are also competitive wrt confidence in their consistency then they also get some additional prize money that is only available to staking participants.

If they are overconfident and do not do better than chance then the coin they staked is destroyed.

But there was an initial sale of the coins right?

All these cryto ico things are premined or centrally allocated in a way where they get a cut. If you have access to the printing press, then you make all the money, no matter what happens with the currency.

Sure, maybe they gave away some of the coins to datascientists, but I am sure they kept a healthly cut for themselves.

No initial ICO. Only way NMR is earned is competing.

Reply on news.ycombinator.com