[dharmon]

I think this posts misses the most important piece, the real secret sauce, which is how do you sift through massive streams of data and separate the signals from the noise? And how do you do so continuously so that as soon as an edge evaporates you don't keep trading it?

I think the key insight they made here is much less sophisticated than many think. The usual guess is that all these math geniuses have some magic statistical models that tell them the answer, but I don't think that's it. There are known, good ways to detect these things, and most (if not all) hedge funds are aware of them.

I think the magic is in the systems engineering they have done. It is a system which is able to evaluate the quality of a signal as it would be traded. Traditionally, quants come up with models that they then backtest to "prove" before doing live trading. A lot of models that look great on paper, or on historic data, fall flat in real-world trading. Hedge funds spend significant time and resources on quality back-testing data and systems, and I think Renaissance has been able to take this to a whole new level.

This is all mainly a guess on my part, but based on the book The Man Who Solved the Market, which alludes to this system without going into details (obviously). It is much less exciting than some super sophisticated ML models or what-not that people imagine is the source of their success.

This post focuses on the leverage, which is great for goosing the returns, but isn't the whole story. Put another way, if you could magically be gifted some part of Ren, which you rather have their special leverage arrangements, or the signal vs. noise oracle?

[fdgsdfogijq]

No I think the magic is in the signal detection. Basically, there is only so much alpha in the market. RenTech gets to the signal first and exploits it before everyone else. Thats what makes this so hard. If they all know the same math to figure out where the current signals are, its about who is willing to take the risk of getting into the position before its clear its really a signal.

I used to work on investing using credit card data (which rentech uses). All the hedge funds can access the daily credit card data at the same time. The question is, where there is a huge amount of noise in some pattern in the data, the risk is still high entering a position on it. The fund that models that risk best and says, "when the unique number of credit card spenders at this company goes up, I buy", they make the most.

[dharmon]

I think we are saying the same thing, but I am just elaborating on what "signal detection" means. The usual approaches at worst just don't work, and at best don't scale very well.

I am saying that (I believe) RenTech has taken a very holistic approach to signal detection and maintenance, rather than the very academic approach that used to be the norm. And even if a signal is found, they have to be weaved into the trading system gradually, and eventually removed when they no longer work. This is a very challenging problem and they are very good at it.

[matheweis]

Other snippets that have been accumulating over time suggest you are correct about the algorithm(s) being relatively simple.

First hand accounts suggest they are for the most part using simple linear regression; see this comment from a few years ago: https://news.ycombinator.com/item?id=19065226

[want2takearide]

They literally hire the smartest people they can find for this, e.g. David Donoho [1]

[1] https://en.wikipedia.org/wiki/David_Donoho

[matheweis]

From the interview linked in my other comment here, they’re not likely doing anything particularly special. relevant quotes:

> Now we have some of the smartest people around, working in our hedge fund, we have string theorists we recruited from Harvard, and they're doing simple regression.

> the smarter you are the less likely you are to make a stupid mistake. And that's why I think you often need smart people who appear to be doing something technically very easy

> we had 7 Phd's just cleaning data and organizing the databases