Don't forget that Goldman has every incentive in the world to hype the importance of the code, so that they can use the FBI to punish this guy. This almost certainly could have been handled with a lawsuit.
Surely they've had people quit to work for competitors in the past. People who knew exactly how their code and current strategies operated. All this guy did was actually copy the code, which is clearly IP infringement, but not necessarily so much more damaging in practical terms.
Don't underestimate the vindictiveness and competitiveness of these financial guys.
I don't know if they want to hype it that much. They are a publicly traded and looking unnecessarily weak could cause a huge unwarranted dip in their market cap (which they have a fiduciary responsibility to prevent).
Program (or quant) trading accounted for 40.4% of all NYSE trading volume during the week of 6/15 - 6/19. Goldman's program accounted for the more of this than any other firm, almost 1.2 billion shares, or 20.6% of all program trades. Ostensibly, this guy made off with the algos responsible for (.206 * .404 =) 8.3% of all NYSE trades. It's quite possible this is the world's most valuable source code. There are lots of people claiming that the code is useless without their infrastructure, but knowing exactly how GS makes their trading decisions allows anyone who holds the information to engage in front-running, or set traps for them by pushing stock prices one way or the other in order to trigger their system to buy or sell.
> Major developing story: Matt Goldstein over at Reuters may have just
> broken a story that could spell doom for if not the entire Goldman
> Sachs program trading group, then at least those who deal with "low
> latency (microseconds) event-driven market data processing,
> strategy, and order submissions." Visions of swirling, gray storm
> clouds over Goldman's SLP and hi-fi traders begin to form.
Here's what's not clear to me... Was the code simply a low-latency platform for making arbitrary automated trading decisions, or did it include trading strategies as well? hachiya's link suggests that the major value was the microsecond latency of the platform. If that's the secret sauce, I would think it could be replicated without resorting to espionage.
It's really both. Data comes in from various sources with low latency (you get the data right as trades happen, not the 15-minute delay you'd get at Yahoo Finance or something), then the system runs the data through various trading models they've developed and makes trades automatically.
Once a competitor gets their hands on your model, you might as well trash it, since they could use it against you; and places like GS spend BILLIONS on their models, so if this guy really got all (or even just some) of their source code out, the firm is going to take a big hit, and it's not just going to be a short term bump in the road either.
Quant trading algorithms, like most stochastic trading approaches, lose their effectiveness over time as the market internalizes them and as knowledge dissipates naturally (not necessarily through leaks). The best of the best individual quant trading algorithms have a half-life that does not exceed one month. And we're talking exceptional here. Also, at any given time many, many algorithms will be running simultaneously; complementing, overriding eachother as necessary.
So there's a lot of churn overall. The code that someone picks up in this way contains a single snapshot of this transient trading strategy, not particularly useful to run on their own. You're right that, what really sucks for Goldman is that someone could find some basic patterns inherent to their trading platform and trade "against" Goldman.
The catch is that this sort of maneuvering happens all the time. Sure it's going to be painful in the near term for the Goldman quant guys, but they don't need to start from scratch necessarily.
I think it can last longer than a couple months. I worked on some relative value trading algorithms in connection with work that a b school prof was working on, which got picked up by some of the major HF players. It continued to work for a couple years before the correlations faded away.
Relative value and high-frequency trading are very different animals. With relative value, you probably don't even know if you have statistically significant results for a couple months; with HF, you have evidence within hours.
A "quant" is a trader who writes his own trading software. It's easier for them to do this than to write a spec for someone else, especially since it's constantly being tweaked and tuned, and the full-time programmers in a bank build tools and infrastructure, they don't trade themselves.
Just for a bit more info...most job postings that I've seen for quant positions generally require at least an M.S. in Finance, if not a a Ph.D. in Finance or another technical field (Math/Physics/Comp. Sci./Engineering). This is so you'll either already be trained in the underlying financial math, or will be able to learn it quickly.
Also, the field is pretty stressful from what I hear. Not that some dev shops aren't the same way though...
Yes, and you would have picked up your programming skills incidentally through studying your main subject. Quants are not software engineers - the programs they write are simply proxies for themselves as traders, able to react instantly and run 24/7.
He doesn't appear to be a quant, based on his linkedin profile. Looks like he had a telecom background, so he was probably working on the architecture of their co-located trading platform. In the automated trading world, co-location means putting your machines at the exchange to reduce the latency of receiving market data and entering orders. For a high frequency strategy (i.e. tens or hundreds of thousands of trades per day), these improvements can really increase profits, and there is a huge arms race under way in industry to get the lowest latency.