You're saying that unattractive, unperformant fund managers have nothing to recommend them, and so that makes the correlation negative because they're censored

You're saying that unattractive, unperformant fund managers have nothing to recommend them, and so that makes the correlation negative because they're censored from the sample? Otherwise it would be roughly flat, if they didn't let the unattractive bad managers go? Could be. But then, it's still a reasonable heuristic, because you might find that unattractive managers are uniformly pretty good or better, and attractive ones are 50-50. The best manager might be one of the really really good looking ones, but overall there is less selection pressure on attractive managers with respect to performance. (Halo effect?)