Admittedly I have no idea how something like that could be quantified, and the study of economics has been trying to figure it out for decades. But if YC Research is looking to make maximally impactful investments, it seems like studying investing itself would make sense. There's a high risk of not producing anything useful, but also a huge upside if you get just a little better at figuring out who to fund. YC Research is in a position to bring together people from the worlds of business, machine learning and economics who might be able to make a dent in the problem.
Practically, it seems like navigating the financial investment process is the core skill of YCombinator, so you'd already be pretty far down the learning curve. And you don't have to convince people to adopt your ideas right away, you can try out whatever ideas your research produces simply by making investments. Better capital allocation of all forms could have a really big impact on society at large, if the right systems can be developed.