I was thinking about simply limiting the universe for scoring: if I understood correctly, when Numerai One wants to long a stock in Japan, it has to short something within Japan to build a safe LS position. If this is correct, local signal (may work only in Japan) may be still useful for Numerai, and calculating corr or mmc only using Japan stocks should be possible.
I understand that for Numerai things are easier if users predict all the stocks, and if missing, the median fill is performed. Maybe in the end Numerai does not need such ‘local signal’ I proposed.
But still, how about giving feedback about localized scores (e.g., score for only US stocks etc) to users?
If, for example, there is a guy who is extremely good at predicting US stock returns but very bad at doing so for Asian stocks, he may not earn much in Signals at the end of the day. He may leave. But if those localized scores are given, he can realize his weakness in his portfolio, and he may be able to find someone who is good at Asian stocks but not good at US ones in the rocketchat. Then they can start to collab. They can now submit great signals covering most of the universe and earn a lot in Signals, and Numerai can also benefit from them and improve its meta model.
…Yeah I guess the last half of my argument may be a fairly tale, but in my book giving feedback about localized scores seems to be a step to improve Signals. If everyone is using yfinance or something similar that Numerai already has, I am afraid that Signals may end up with being an impoverished version of the Tournament.