I have a question on the reward system of Number.ai:
Let’s assume that I create a good performing model with steady high CORR value and share it publicly!
If 100 people copy my model and submit the same predictions then it means
the hedge fund pays out 100x rewards
no added value is created by copying the same model
Can sharing code cause problems for the hedge fund?
Or did I totally misunderstand something?
It is unlikely someone will have the exact same model as yours
There is value in having more stake value on the same model by potentially many people, namely a sign of confidence in the model (e.g. the example model). Conversely if the stakes are wrong, they will burn.
Sharing of good quality code is benefitial to the community in that people make better models and the meta model improves, as long as it does not lead to hyperparameter-level duplications.
Yes, the staking mechanism takes care of this to some degree. The ego of participants also helps (wanting to accomplish something on their own). There is no perfect system, but the current one is doing a decent job of self-regulating.
If the hedge fund doesn’t do well, there will be no hedge fund, and thus no community either. If the hedge fund does great, does it help us beyond the fact that the tournament keeps existing? Maybe. (I’m not sure that necessarily leads to rising token value.) But if we are the engine that keeps it doing great, I think they’ll want to incentivize us to stick around.
One incentive can be considered is that the fund buyback the NMR using x% of their performance fees and redistributed those NMR to metamodel contributors.