On the website https://25iq.com/2014/07/09/a-dozen-things-ive-learned-from-jim-simons/ (mirror), James Simons cited “We have three criteria. If it’s publicly traded, liquid and amenable to modeling, we trade it.” Why liquid?
Liquid => low transactions costs. When you are trying to take advantage of small anomalies you need transaction costs to be low.
Simons does not believe there are big price inefficiencies; he does not think you can buy for 10 something that is worth 20 tomorrow. He thinks there are many small anomalies where a security is mispriced by a fraction of a percent. To profit from this you need: to combine many securities (so the randomness washes out) and to concentrate on securities with low transaction costs (if an illiquid security is "10 bid 11 ask" that is a 10% round trip transaction cost, no way you can overcome that with a fraction of a percent per month price appreciation).
A market being liquid also means that the books are deeper. This implies that you can trade larger volumes without moving the market prices as much.
A liquid market also allows you to reverse your position quickly and avoid being stuck with unwanted assets.
When the market for a security is highly liquid, a large, sophisticated investor can assume that his presence in the market for the security does not affect liquidity. When the market for a security is illiquid, a large, sophisticated investor must consider how his presence in the market for the security affects liquidity. To understand this intuition, start with a standard Kyle (1985) model, and suppose that an additional informed investor is deciding whether to trade the risky asset given an exogenous internal cost of capital.
RenTech is large and sophisticated, and likely has a short-term investment horizon and a high IRR. If its presence in the market for the security affects liquidity, it might not be able to exit positions easily, and the returns process is likely to be less "amenable to modeling."
A third interesting reason why the markets should be liquid is volume of data.
If the models being used demand a lot of data to have a small error (as is the case in Deep Learning models), then liquid markets give you a large enough dataset to train your models.