From ML in tech to finance
I have years of experience in ML at Google, FB and a couple other places. My experience is applied research, not just tensorflow/pytorch wrangling. I’m beginning to get tired of big tech and want to see what finance has to offer. My background is probably best matched to quant research/algo development, and I’m mainly interested in Two Sigma, Jane Street and Citadel. Is this a good career move, medium to long term? How does TC growth potential look these days? wlb is not a factor I’m looking to optimize for. TC ~700k
Do you hold a PhD? Can you elaborate what it means to do ML beyond tensor libraries. Are you at Google Brain? What orgs in google allow you to do this kind of work
I’d answer but that might dox me. Sorry.
Even more interested now. How to get into Brain without PhD? Internal transfer?
Add Hudson River trading
Thanks
That's a good one...actually great one
The real question!
What level for your TC? FB E6?
I think most of the people on blind from those places are devs rather than quant researchers, which is probably the role you'd be targeting. So I don't know how much specific advice you'll get. In terms of comp, I make significantly more than you with less experience as a dev. I'd expect researchers to make more, so I wouldn't expect comp to be a barrier.
Thanks, this is helpful. Do you know if you hire researchers with my background frequently, or do you mostly target schools?
I don't know much about how we hire researchers, sorry. My guess is that that's the position you'd be most interested in, though. This posting for example: https://www.janestreet.com/join-jane-street/position/4276720002/
DeShaw, Jump, and Voleon might be worth a try.
(Background: I used to work at Citadel and now work at another top quant firm). As for TC growth -- yes, the top quant shops can definitely pay what you're looking for and if you're really good, the TC can grow much higher. However, I suspect in your case you might have to take a pay cut at least for the first year or two. You are switching to a new area with no track record and no domain expertise -- those companies might be willing to take a chance on you but since you're looking for 700k+ then I suspect they might not be willing to pay top dollar until you first prove you can deliver. Part of it is seeing if your skills can transfer (a good "applied researcher" on computer vision is not necessarily going to be good at finding trading signals) but also there are a few cultural differences between FANG and hedge funds and I've seen a huge number of tech people flame out due to cultural mismatch that they weren't willing to adjust to. So yes, it's definitely possible and if things work out you'll be able to get to the comp you're looking for; so I recommend interviewing if you're interested just to see what happens. But I suspect the offer's bonus may be a "target" rather than a "guarantee" and there's a lot that can go wrong. I suspect you might decide it's not worth the risk. Still, go ahead and interview just to see what happens!
Thanks for taking the time to answer. This was very informative. What are some of the cultural differences that you’ve observed ex-faang struggle with?
The biggest cultural differences I've seen are: 1) Different expectations about how fast research needs to pay off. As a quant at Citadel, you'd better have at least one profitable trading signal deployed to production within your first year (and probably much sooner) or else you'll be let go. No excuses. (They are a bit more forgiving about developers though). 2) Compared to tech, there's far more "it just has to be this way, stop asking" in finance. This is due to regulatory requirements, or its too much work to change the system to support your idea, practical experience of your boss, the gut feeling of the CEO, whatever. Some of these rules are for good reason and some aren't; but your boss will have EXTREMELY strong beliefs about what is allowed and what isn't, and I've seen people flame out fighting dumb issues like "If the firm-wide accounting system can't support my brilliant trading strategy, then accounting (and every other team in the company) needs to change their system rather than me dumbing down my model". It's not that you don't have freedom to do your own ideas; but there are more constraints and you have to be willing to live with them. 3) Less on the quant side but common on the dev side -- the old debate about "Should we spend 4 months doing this manually or one year writing a generic solution in case its needed again" has a different break-even point in most finance companies vs most tech companies. (Finance companies usually prefer the former). You will not be allowed to rewrite the system from scratch every couple years. 4) "Failure is OK as long as you learn from it so now we'll pivot". Common attitude in tech, but good lord, don't ever say that at a fund. Yes, everyone knows ideas doesn't always work out and you're allowed to hit a dead end once in a while, but don't ever say "Failure is OK" out loud when your desk is losing money. 5) An interest in the business side is key. So many people are only interested in the math/engineering but there's lots of market micro-structure details in trading (for instance the rules for settlement date of a stock around the time that stock pays dividends) that won't be obvious from a backtest and you have to be interested in understanding those kids of non-technical details and how they affect your strategies. 6) Collarary to #5 -- finance has a much lower signal-to-noise ratio and non-stationary and all kinds of other things that mean you should never 100% trust your model, no matter how good it looks in the backtest. You will need to be the sort of person who's obsessively watching the output results (not just the results of your back test, but the live production trading numbers too). I used to know a trader whose model obviously introduced a bug in the latest release (overnight, it suddenly started losing money nearly every single day) but he kept not even looking at that because his backtest was fine and he kept saying "it's a statistical process so it's expected there will be some periods of underperformance". Amazingly it took over a month for him to admit there might be a bug and rollback to the prior version.
Jeez stay away please. You are doing well and unless you hate what you do please do not go into a slow, extremely bureaucratic industry like Finance. TC *may* be more than what you make but cannot confirm and will vary by company.
Thanks. Do you feel the same about the mainly tech-driven places mentioned above?
I personally know the ex Head of Tech from Citadel and he explained the shit show that they have going.100% stay away. Two Sigma is okay from what I heard but every place has its problems so that may not be too bad. Not sure about Jane Street. I'd say try and get offers and then decide. TS may not be able to beat your current offer afaik.
This thread is so useful! I've another question for the finance people on this thread. Very likely OP has transcended this problem already, but I haven't. One thing I hated about Google ML was the competitiveness - politics, visibility, territory, secrecy, alliances, backstabbing, theft of ideas, etc. I'm curious if anybody can compare how brutal these phenomena are at core Google PA ML roles vs. Jane Street / Two Sigma (be it dev or quant)?
Very useful comment, thank you. So if you find something promising, you can talk about it at lunch without fear that somebody else will take it? Are the incentives set up differently or is it just that there are more ideas than people to try them out? What reinforces/ensures the openness/collaboration?
I haven't worked at other finance firms, but from what I've heard our pay structure is much less eat-what-you-kill than other places, so another desk making more money is likely to increase bonuses for your desk. It's probably partially a culture thing, too. You see senior folks collaborating across teams a lot and follow suit.
Nice job man, yoe?
9ish.
Damn dude that is phenomenal