I'd say a good 50-75% of the work is tricky quantitative programming. Read: prediction, analysis, concurrency, micro optimizations (today I shaved 10-20 seconds of a 15 minute program run time). Probably 10-15% of it fits what you would describe as "far out, complex coding". There is of course grunt work; tax calculations, making our FIX talk to their FIX, etc. There is a lot of frustration; I've had a number of ideas I liked, all of which failed horribly.
I'd say it's roughly equivalent to being an employee at a startup, except that there are only 3 employees, all of whom are within shouting distance.
The most fun is being able to watch the market, and have that epiphany where you all of a sudden have a trading system in mind that you want to try. You furiously code for the next 30 minutes, compile, debug, input parameters, hit enter--- and then have your dreams crushed to find "hmm... well if I traded that, we'd lose 98% in x years." :)
I'd say it's best described as a startup meets online poker. On the startup side, it's a small team, constantly pivoting based on perceived market demands for different investment vehicles, and lots more freedom to fail fast with little ideas. The poker side is where the whole luck factor comes in, dealing with huge swings due to flash crashes and political speeches.
Basically I knew a really smart guy (my boss) and he just happened to be launching this hedge fund just as I was finishing grad school.
I would love to hear more about your work (and yummyfajitas). I work for a company that manages 401(k) plans and haven't had much intellectual stimulation recently. Modeling isn't something we do to a large extent.
What programs/language do you do your modeling in? Where does your data come from for back testing?
Any more info about the frameworks, day to day, etc. would be very interesting.
I'm currently reading a book titled "Algorithmic Trading & DMA" which will hopefully give me a basis of how market systems actually function and I'm hoping to build up from there.
Data can come from numerous places depending on your licensing budget. For an initial analysis you can go to Yahoo! for daily prices; data reliability isn't great but you get what you pay for there. Bloomberg provides great data (assuming you guys have a Bloomberg tutorial), but make sure to read their T&C! They have very strict rules about taking data off machines, etc. Other sources are places like Tickdata.com and similar vendors whose sole purpose is to provide you with clean, reliable data.
As far as frameworks go, to be honest most of the code is created from scratch. My C# framework is around 50,000 lines plus the code for individual models. It still feels like it's only about 5% of what I want it to be. If I were to start over again, I'd probably go with R because so many computational finance people have contributed to it and its graphing/plotting/statistics features are amazing.
Are you in a position to change how your company approaches modeling? We're currently expanding into consulting and services as well. If you want to talk more about that, check my profile and send me an email. We could probably help you guys get started in the area or work with your team to develop custom models.
The book looks like it may be really good. Most of what I've learned about automated execution has been through the long road of trial and error. Most of those topics in the book are important, though it seems based on the ToC that the content may fizzle out just when it gets to the good stuff.