Numerai‘s Erasure is one of the most exciting developments in the Ethereum ecosystem I have seen in a while. Along with Augur and Maker, money/finance 2.0 continues to be the most compelling use case for decentralized blockchains. https://t.co/5R8QFBFQAG
Three naive questions after reading the blog post:
Can buyers be individuals?
Do buyers need to precommit to the conditions under which they'll grief? If not, what's their actual incentive? The blog post offers a game-theoretic justification for post hoc griefing: "it may indeed be rational for the buyer to grief the seller if doing so increases the likelihood that the seller will improve the quality of their subsequent predictions." I don't get it. If a good model has been underperforming due to chance or variance, wouldn't results-based griefing push it toward conformity and hence mediocre expected performance? If the underperformance is an artefact of the model not being as good as its track record, how does griefing change the fact that you picked a lemon?
If Erasure works and NMR becomes the staking token for all kinds of high value predictions, does that decrease or increase the risk of a regulator targeting the Numerai team for securities fraud or facilitating illegal betting? Honest protocol developers have been spared such actions to date but you never know.
(b) create a protocol for other hedge funds (oh, and the wider utility of others using the protocol will increase demand for the token).
I think (a) is the much more interesting problem and (b) has whiffs of of protocol-fomo -- like they saw Augur and feel left out of the protocol game.
I guess the hedge fund business plan is not working out for them. For anyone who's worked at a decent quant trading shop the success of (a) seems implausible but at least an interesting idea to try. Creating a protocol (and stressing this wider usage might increase the value of the token) has been overdone.
>I think (a) is the much more interesting problem and (b) has whiffs of of protocol-fomo -- like they saw Augur and feel left out of the protocol game.
Numerai is more interesting than an open source data marketplace that never existed before? Why? Sure Numerai has cool videos and Erasure is just Richard & Fred talking to the camera, but if you understand the importance of a working, successful data marketplace, it's way bigger than Numerai.
Also what do you mean about "whiffs of protocol fomo?" Numerai's master plan was to decentralize everything at some point. That's what they're doing here. [https://medium.com/numerai/numerais-master-plan-1a00f133dba9](https://medium.com/numerai/numerais-master-plan-1a00f133dba9)
>I guess the hedge fund business plan is not working out for them.
How do you know that? There's no publicly available data. As I pointed out above, the plan was to decentralize everything anyways. So the team was going to do something like this for NMR no matter what.
>Creating a protocol (and stressing this wider usage might increase the value of the token) has been overdone.
That's like saying, "we shouldn't create a social network or instant messenger because MySpace and AIM already did" lol. If we went with your logic, Facebook and Whatsapp never would've existed. Plus that's not even specific towards a decentralized data marketplace. No one has built one of those successfully yet.
By the way if Erasure works, it'll become incredibly important as a resource for AI. That's the bigger, longer term goal.
I don't understand the logic behind Griefing, what rational actor would wish to further lose money to punish someone?
And once the ratio for someone gets high enough his competitors may wish to grief him potentially, the whole concept just seems weird. Numerai could use a game theory expert.
Oh the whole thing is in murky legal waters, you can be damn sure no hedge fund would want to grief anyone just for that reason.
Something that ought to be enough to fight against a Sybil attack is to just commit as much currency to an entity and have it be locked for some period of time (this does constitute skin in the game because you are opening yourself to crypto currency volatility at minimum). And make as many predictions as possible as often as possible - thereby reducing the probability you are just shooting darts at random, perhaps a score for buyers to decide if it's high enough for their taste.
My thoughts exactly. It would also give the hedge fund plausible deniability of insider trading since they wouldn't need to know -how- the predictions were made, only that they -worked-.
Related: I'd be curious how they ensure a prediction is released to a buyer atomically after payment without requiring third-party trust. It seems that someone would still have to be trusted to carry out this part of the protocol, with the buyer agreeing that they did in fact receive their prediction (either side could cheat the other.) The problem above could be solved with MPC or trusted computing. But its not perfect.
There is another problem I see with this too. The hedge fund is trusted not to destroy their stake... just because. A malicious hedge fund could be black mailing someone providing information for a cut in their stake or some other reason. So will that part be autonomous or run by humans? A lot of crucial info seems missing in this post.