With 6 million NMR in circulation and only 1 million NMR currently staked (https://numer.ai/nmr), there’s plenty of supply right now for any stakers who want NMR. So I’m not convinced that adding more gamers will push the value higher. Indeed, over the last year the total amount staked has only gone up (implying increased NMR demand) but the price of NMR has gone down (on average).
We’ve already got the world’s best data scientists, and they’re doing a great job at the problem they’ve been incentivized to solve. I think that adding another bunch of similar models made with similar techniques will only:
- lower the payout factor,
- negligibly change the metamodel, and
- have zero impact on NMR value given the high circulating supply.
I know this opinion flies in the face of various CoE proposals aimed at increasing participation, but I personally think we should try and reduce participation. Specifically, we should look to reduce the participation of folks who are only focused on CORR. Why? Richard’s comments during the last fireside chat implied that having a ton of models maximizing CORR isn’t helping the hedge fund make the kinds of revolutionary investment decisions they seek to make. But since those players still get paid, we have misaligned incentives. Result: the meta model is flooded with similar models, payout factor goes down, no one is happy.
An easy way to re-align these incentives would be eliminating pure CORR payouts. You might think I’m crazy but guess what? It’s already happening thanks to the diminishing payout factor and instability of NMR. Right now we’re all getting 0.45 * CORR. And that factor is dropping. At the prospects of earning a 2% return with a 4% CORR model but then losing 10% due to the wildly dynamic NMR price (and of course having that stake locked up for 30 days so you can’t react to NMR price changes), you might pull your money from this game.
Numerai could nudge incentives in the right direction by removing payout factor multiplication on MMC stakes. Pay MMC stakers in full, aka:
return = payoutFactor * corr + mmcMultipler * mmc instead of
return = payoutFactor * (corr + mmcMultiplier * mmc).
Another option would be to make a new MMC that directly helps the hedge fund make smarter decisions, one that is so tempting returns-wise (3x??) that folks who really want to maximize returns will make models that maximize this new MMC. This is of course what they talked about at the last fireside chat and partially/prematurely rolled out as TC version 1 earlier this year. I eagerly await the next iteration.
In the meantime, if I were the hedge fund, I’d slowly gobble up circulating NMR for the treasury. Spend a percentage of hedge fund earnings each month in an effort to keep the army of data scientists in the game. After all, NMR is the gas that fuels this particular race. Numerai counts on it having value so that they can incentivize data scientists, and data scientists count on it having value in order to put their time and energy into earning it, otherwise they’ll find other ways to make income. But with so much NMR circulating, third parties have a greater impact on NMR value right now than those for whom it counts most.
These kind of changes have to be more impactful than talking a few more data scientists into joining and staking a few hundred bucks each.