Hacker News Reader

Banking != HFT.

I've worked at both a bank and an HFT firm, so I think I have some perspective you might find interesting.

To make a broad generalization from my experience, "Traders" at a bank are mostly non-technical Excel junkies. It seems like in your experience they also went by the title "Programmer", which I agree is a gross misnomer. Besides that, I think the most important thing to realize is that at banks, non-technical people run the show. The technical people are at the disposal of their non-technical overlords, and thus they spend their every waking hour implementing terrible specs like "do {some incredibly specific task}, and make it fast." Even worse, they are far down on the ladder of responsibility; a "banking programmer" has no feel for the big picture. This all culminates in what you describe: highly specialized/one-off/non-generic piece-of-shit code. But whatever, their MBA overlords don't care about the means, just the ends.

To make another broad generalization, "Traders" at an HFT firm are predominantly go-getter CS majors from top schools. Thus, if they have an idea, they implement it themselves. Any delegation of duty is done with strict specs, and bugs or terrible design in the finished product will get noticed. Incentives abound for writing good code: scalability/genericness is rewarded (have something that averages $1k trading symbol XYZ every day? You're a millionaire if it generalizes to many symbols), speed is rewarded (just the nature of the game), and bugs should lose you money (otherwise they are just unknowingly optimized around).

tldr: banking is like you describe, but there is plenty of quality in the HFT world.

There is a lot of crappy developers that's true.

If you look at the serious pieces of infra inside the front offices for pricing and risk however, there are some smart people that have done some good software.

Quite often however those projects are degraded as stewardship of these projects is inconsistent, and long tenure is not always there, so you don't get continuity in terms of engineering talent.

Frequently this is due to the fact that IT is treated as a cost rather than an enabler of profit, and salary increases do not keep pace with the market worth of fulltime employees - leading to job hopping.

For the non-managerial part of the financial developer's career, it is often a better trade to be an independent contractor on a good daily rate with tax breaks, than take the chance upon fluctuating market conditions and bonuses.

As software developer bonuses are not tied to material outcomes like the front office but are generally more qualitatively judged, this leads to insecurity.

The developers that hang around for many years and end up being the stewards of these good systems that slowly rot, are quite often the ones that are incurious about contracting or moving, or have no better options.

We are routinely picking the top few percentile from the best colleges - the banks don't lack talent - but what they lack is management with good engineering discipline, and good projects sometimes turn bad.

For the professional developer who is actually skilled, this provides unbounded opportunities - it is often easier to make a really big impact with relatively little effort by pruning an overrun garden than rebuilding a very large amount of function points from scratch.

Finally, it is incorrect to say that version control systems are not used. Simply not the case.

It depends on the bank, at some banks front-office developer bonuses are pay out of the desk bonus pool rather than a tech bonus pool.

It is rare these days in my experience - it creates an appropriate separation of concerns to have the management alignment distinct (that can be interpreted to be a Sarbannes Oxley requirement.)

There can be budgeting for specific desk profits in the relative bonus pools decided in IT, but bonus makes up a smaller percentage of developer total comp generally unless you are approaching management.

If you are desk dev and hack Excel and Access for the front office that can help your case in regards bonuses, but this isn't true software development of course.

This is from my experience in London, I guess NYC might be different due to regulatory concerns.

I am speaking of my last 15 years in London working in derivatives/front office IT, and interviewing for many jobs in front office IT. It is very exceptional to find banks that will organise IT to report into the front office structure - this doesn't occur, and I almost never hear of such a thing when I see job specs.

It's not all of IT, but it's not unusual for quant dev type roles to report to the front-office (either singly or as part of a dual-reporting structure). The reporting doesn't have to be direct for dev bonuses to be tied to PnL either, what I've seen happen at some places is that a certain percentage of the desk bonus pool is assigned to the development team and the development team management than decide how to split that pool among the developers.

Go to efinancialcareers and search for "PnL bonus" if you want to see some examples of job specs mentioning it.

So true. Finance attracts the most incompetent egomaniacs possible. I love getting lectures about how to design software from someone who has written:

    if x == 0:
        return "Monday";
    elif x == 1:
        return "Tuesday";
    ...

This should be a criminal offense.

Honest question. What is wrong with something like that? I could see it justified by saving the overhead of a complex Date class.

I had the same question. I had these thoughts:

It looks like python, so the semi-colons are unnecessary.

If you want to rid yourself of datetime for whatever reason, implement it as a list, returning name_of_day[x], so that you can start the week on a Sunday(=0) easily if you want to.

Beyond that, I'm stuck too :-(

If you are going this route why not use a switch?

Or just an array.

python have no switch

Two words: mapping structures.

Python ships with a standard datetime class which is powerful enough to represent time, and there is a ubiquitous module, dateutil, which extends it to do nearly anything people need to do with dates regularly.

Additionally, one would normally use an array of names and index it:

  days = ["Mon", "Tue", "Wed", "Thu", "Fri", "Sat", "Sun"]
  that_day_you_wanted = days[x]

It's just inelegant. A better approach would be an array where you return the content at the index.

I have a better one. Project manager replaced enums with strings and instead of the switch statement he put a long list of if-s (without the else-s, too), and a lot of obfuscated crap between the braces.. Guess who had to sort out that mess? Also, the motivation for that horror was related to how those string parameters were visible in intellisense (because they were made static members in a class). Don't even ask.

> I had one project where the damn process used so much ram per request they had to go buy an Azul box just to make it run even at 2 req/sec

I either worked on the same project as you or very closely related one at BSC. The upside of this was I got to play with Azul hardware for a year.

It sounds like you had a bad experience at a single team in a single bank and are trying to extrapolate those experience to a sector that employs hundreds of thousands of developers.

Yes, there are plenty of shitty programmers at investment bank, there are also lots of top-tier ones.

And yes how much money your code makes is more important than how well factored it is. It doesn't matter if your code is the most beautiful code in the world if it doesn't work, shitty code that gets the job done is better. Regardless of whether you're working in finance or on a web app.

Fast C++ is an oxymoron. Real men use C and assembly. And yes, I bootstrapped my own HFT

"And yes, I bootstrapped my own HFT"

I would be interested in hearing your story. Would you like to share some details?

Realized this would be a much larger effort than a single HN reply or a single blog post would justify, so I am starting a blog series on this. First post: http://news.ycombinator.com/item?id=2829116

Thanks! The first post left me hungry for more :D

Would love to hear more about what's involved on bootstrapping an HFT house.

When you're doing performance optimization, C++ is basically C. You restrict yourself to an almost C subset and maybe end up doing some clever stuff with templates.

Maybe it's different in London compared to NY. There's plenty of decent makers in Finance in London. There's people who use all the best practice techniques in software development. People who wrote code you use every day.

> Maybe it's different in London compared to NY

I've seen the kind of guys described in zed's first para, in Finance in London. I don't think it's different, but YMMV.

Yeah, they exist too for sure. My guess is you'd get a fairly typical spread of quality compared to other industries.

some bank sunk millions (and maybe billions) into infrastructure just to run that crap even moderately fast

Why, were they running Ruby on Rails? </snark>

So one takeaway from this gem of a post is that if you don't write open-source code you're a shitty programmer?

I think the point is that anybody can say they write great code if they never have show it to anyone.

I understand that, but not everyone runs around claiming they're awesome seeking validation from Zed Shaw. There are lots of great programmers in every field that involves a computer. There are bad ones too, just like in open-source. To claim that there are no good programmers in any given field is nonsense.

Given this type of code directly makes or loses money, it's effectiveness for it's intended purpose is already directly measured.

Whether it's "good code" or not from a technical perspective, if one guys pile of crap code works the market 10x better than the other guys, who has "better code"?

Funny, I didn't use open source at all in that post. I said, they don't show it to anyone and I meant even other employees at the bank. Some of those dudes guarded their secret sauce like crazy, even when I was asked to help them improve it. When I finally got in to see it, it was a nasty horrible mess.

But hey, good lame-troll attempt.

Flippancy aside I'd say if good coders aren't reviewing your code you have a good chance of being a shitty programmer. Open source is a great way to achieve that but certainly code review is prevalent in other places.