What precisely are the differences between various quant hedge funds?

[dog]

I am struggling to determine what precisely the differences are between the various firms in the quant hedge fund space, and how strategies/research activities differ in general between firms, if at all.

For example, I have a fair understanding of the activities of market makers like JS/Optiver/Jump etc. but when it comes to companies like DEShaw/GSA Capital/Quant arms of multi managers, I struggle to understand how each firm differentiates themselves. It's not clear for example whether GSA does any HFT anymore following the spin out of XTX.

How is it possible that a multi manager like Balyasny has comparable expertise to a quant-only fund? It seems to me that only firms with a real technical edge can compete in this space-what exactly is it that these firms are doing/do differently. Does a small pod in Balyasny really have a competing chance with quant-only firms: I suggest the answer is yes (as they exist), therefore the next question is what space do they operate in, and how does it differ from others?

Information I get from recruiters/previous talks with employees of various levels are things like 'stat arb', 'average time of holding', 'read Cliff Asness'.

Interview preparation guides are not very insightful either, as from whichever firm that provides the guide it basically says 'learn anything from maths/computer science/stats' without any real guidance as to how these theoretical skills are applied. For example, an interview guide might suggest you learn how to do the LU decomposition of a matrix, but let's face it, how many of you would actually know how to do that? Virtually no-one, you would look it up if you needed to do it by hand. No one has done LU decomposition since 1st year undergrad...it's not that you are intellectually incapable of doing it.

Do the firms differentiate themselves by use of data? Which firms use market data only? Which firms are using alternative data extensively (G-Research)? It's not clear.

And then I have read recent papers from Winton saying 'trend following doesn't work anymore'- what are they doing then?!?!?!?

 

[Seignioraged]

Shame you didn't get an answer earlier. I have no insider info, but as an industry participant what I have inferred from talking with other participants, mainly longer term hedge funds, no HFT.

Large hedge funds make quant work in two ways, pod model ( Millenium , Balyasny ) of "pool" model (AQR, Bridgewater).

Former is straight forward, you are a cog in the model expected to deliver trades, you get plenty of support in terms of infrastructure OUTSIDE of alpha generation. Ie, you will have most of the data accessable, maybe even cleaned, trading is not on you etc. This leads to diversification of general quant ideas by overlapping implementations and gives a chance for very unique approaches.

Pool model aims at the organization having a collective knowledge of investing process, instead of focusing on alpha individually focus is on refinements of certain objectives (ie better vol forecasting) that can be reused by others. General quant ideas are distilled through an iterative process over years and years.

Pods are way more competitive vs pools that are (on paper at least) more cooperative.

In finance and tech it's common of having a discrepancy of what you need to know to get the job vs to do the job. Former being theoretically more complex while latter being more practical. Data cleaning is such an example, you only gain experience by doing it.

As for new grads, I have hard time seeing anyone straight from school being productive alone in a pod (as a part of one with great mentors maybe). What masters programs teach you is knowledge that is widely known, available and someone bothered to write it down (getting rich on a book was easier than getting rich on the concept). Most serious places are ahead of that "forefront of knowledge" in the domain they are active within. Imo this applies to majority of PhDs as well, while usually really sharp theoretically, lack in real world experience. Academia values complexity while industry values simplicity. With that said intuition of what might work, why things are not working etc is what's valuable and education prepares you for that.

Once you get up to scale (Aum>500mUSD), relevant data is relatively cheap, quant has exhausted pretty much all of it, Jim Simons said in some interview that they don't have better data than everyone else anymore for example. Alternative data is more expensive and way less scalable, say car sales data, gives relevant information on 10-20 stocks only.

Quant funds at scale (not HFT, not MMs or prop shops) cap out at ~Sharpe 1 ( no Sharpe hacking with options). Assuming strategies are truly uncorrelated and deliver 0.3 Sharpe you only need 11 of them to get to 1. Contrast this with the few dozens of "new factors" discovered just last month and posted in various journals, often uncorrelated to everything and over 1 Sharpe after trading costs.

Why quant is hard is the same reason why predicting markets is hard, signal-to-noise is extremely poor and in general there are no incentives from any participants to improve this "signal".