LΞPak

If you can find an archive of the NuclearPhynance forums, there was a really good long running thread there about RennTech. Most of the general consensus was around: 1) They’re not really playing HFT games. You can see that they don’t have a footprint in things you’d expect HFTs to have (eg captive broker-dealer) 2) Medallion is mostly trading short term stat arb time frames. Probably just above HFT time frames, where portfolio management becomes a lot more relevant and HFTs don’t want to hold risk. 3) This short horizon stat arb is very heavily loaded on t-costs, because the magnitudes of moves is small. 4) RennTech has *a lot* of alphas all in one place. Any of these alphas alone is only marginally profitable because of t-costs. But stack many orthogonal signals and you start consistently triggering trades. The sum is more than the parts, and there are returns to scale. 5) Medallion seems to also collect a lot of weird anomalies, that just work but aren’t really easily explainable. These anomalies tend to persist for a long time because they’re not intuitive. 6) Their employees tend to be really loyal. Most trading firms have high turnover, hiring people to bring in more strats, but also a lot of people constantly leaving, leaking strays. RennTech really has neither. So they have higher returns to “weird alpha”, because quants are leaving for competitors and leaking the secret sauce. 7) Despite all the PhDs, most of the modeling is probably simple. Not much more beyond regression. But they have a very good pipeline with a lot of data, making it easy for researchers to find new signals, then turning that around to a huge number of markets with very efficient trading. Again, returns to scale and breadth are probably the real moat

I have the better question - I say you N random variables with correlation p between each pairs; what is the maximum and minimum for the correlation p?

Loi is going to ruin your productivity