“Understanding Jane Street: Strong Language”, Byrne Hobart2022-08-01 (Haskell, economics)⁠:

…Jane Street is not the only big prop trading firm, and not the only firm that thinks this way and tries to hire from the same pool. They’re worth focusing on for 3 reasons: first, the combination of secrecy about some things (and more on this below) plus aggressive recruiting means that they have a bigger surface area for outside research than most firms; second, a few of their decisions, like systematically buying black swan insurance and using the obscure OCaml language, make them an interesting outlier; and third, it’s hard to argue with success: Jane Street earned $6.3b in the first half of 2020, up more than 10× from the year before.

…At a firm focused on automation, that means thinking carefully about programming languages.

Ironically, “thinking carefully” is not the origin story of Jane Street’s highly idiosyncratic decision to use the OCaml language. The origin story is that they had a crufty system built on Excel, and hired a part-time researcher to build some analytical systems. That researcher, Yaron Minsky, chose OCaml because he liked it, and because he didn’t expect anyone else to have to maintain it afterwards. But then he decided to stick around to run a research group that used OCaml, and a few years later convinced the rest of the company to move to OCaml, too.

Programming language choices lead to endless debates, and in fact one of the reasons to use a popular language is that if everyone slightly hates it, at least you’ll avoid factions where one group loves it and one loathes it. One solution to the political problem is to make an esoteric choice early on, and then filter hires in part by whether or not they think it’s a bad one. That can be tricky, because a company can end up wedded to a subpar technology, but Jane Street has found a novel solution to that, too; they contribute lots of open-source libraries, and help keep the language alive and up-to-date. The more people there are who can throw together side projects in OCaml, or who are writing other libraries for their own convenience, the better-off Jane Street is. Other companies can free-ride on this somewhat, but there are two likely cases here: first, if they’re not prop traders, it’s good for Jane Street to increase the population of OCaml-fluent programmers and the number of contributors to new libraries. Second, if they are prop traders, Jane Street can probably outbid them for talent, since returns in systematic trading tend to compound.⁴

…Using an unusual language is also a way for a firm to stay truly technical, when the temptation is to drift in the direction of a more traditional finance culture. A sufficiently technical firm can find ways to appropriately compensate people who do boring but necessary work like fine-tuning and refactoring old code. Even if they don’t get a performance improvement, reducing future maintenance costs is valuable, and when traders, researchers, and developers are all somewhat fungible, you can pay someone appropriately for cleaning up an existing program, eliminating some redundancy, and making the code shorter and more readable. That’s right: they can reverse the classic finance formula and pay a big bonus after someone obliterates a bunch of lines.

One of the meta questions to ask about an automated trading firm is: what do they want to ensure is the most reliable piece of code they have? Yaron Minsky has answered this in an interview: “We have lots of ways of turning things off, including a literal physical big red button.” There are many ways to avoid blowups, but the future contains plenty of unprecedented events, and the only way to deal with them is to find a way to stop doing anything else until it’s clear what to do next. And this is a case where readability, clarity, and reason-about-ability are all critically important: as anyone who has worked for a big government agency, corporation, or other bureaucracy knows, the hardest feature to engineer is an off-switch.