• C++ Programming for Financial Engineering
    Highly recommended by thousands of MFE students. Covers essential C++ topics with applications to financial engineering. Learn more Join!
    Python for Finance with Intro to Data Science
    Gain practical understanding of Python to read, understand, and write professional Python code for your first day on the job. Learn more Join!
    An Intuition-Based Options Primer for FE
    Ideal for entry level positions interviews and graduate studies, specializing in options trading arbitrage and options valuation models. Learn more Join!

Algo trading software went wild, lost $1M in one second

Joined
5/2/06
Messages
11,767
Points
273
Infinium Capital Management confirmed only that it is the company at the center of a six-month probe by CME Group Inc into why its brand new trading program malfunctioned and racked up a million-dollar loss in about a second, just before markets closed on February 3.

The documents, dated March, reveal that Infinium used an algorithm that was less than a day old to execute a "lead/lag" strategy between an exchange-traded fund called United States Oil Fund, which tracks oil prices, and the U.S. crude benchmark future, West Texas Intermediate.

The algorithm was turned on at 2:26:28 p.m. (Eastern) on February 3, less than four minutes before NYMEX closed floor trading and settled oil prices. It immediately started uncontrollably buying oil futures, according to the documents, which include letters from Infinium's lawyer to the regulation unit of CME Group, and cite notes from a company developer.

Infinium placed 2,000 to 3,000 orders per second before its flooded order router "choked" and was "dead in the water" a few seconds later, the developer's notes said. The algorithm was shut down five seconds after it was turned on.

By then, the documents show, the firm had sent 4,612 "buy limit" orders into the market. It quickly offset the position, mostly with large "block" trades in the next few minutes, leaving it with a $1.03-million loss.

http://www.reuters.com/article/idUSTRE67O2QQ20100825
 
Bound to happen in stray instances - since we allow the might of the machine to take over the markets.

Question is, can we look forward to benefit from it?

How 'bout algos that wait on the sidelines, and look out for 'erronous behaviour pattern' and then profit when some wildcat-algo goes loose. It would indeed be challenging to correctly identify this, but it can indeed be a good large university project.
 
if only it was illegal to lose money....

this seems more like a publicity stunt by CME to show that it's tough on algos than anything else.
 
Yes, I did make such a comment, and although I laughed when I saw Andy's post, it didn't really shock me all that much.

One system I did years back included a test that sent every possible input price within the 'sane' range through the system to see what it would do. Although of course reals are uncountably infinite, and even floating point is a quite range, quotes only had 6 significant figures, and outside a defined rang would flatly ignore the numbers.

It found nothing,so was that a waste of time ?
Many would say that it was a waste, and that is why systems are so poorly tested.

Algo systems are in a nether world, because it pays so well, the people who build the systems are typically smarter than average, and because it is a secretive business they are extremely reluctant to let others in to test it, and the idea of passing it to the IT department is likely to be met with laughter.

I'm no lawyer, so I can't say if they broke any laws, but I don't see that they've done anything 'wrong', in the sense that they clearly weren't trying to deceive anyone, and I find it hard to say what the difference between a trader making an error and the same error made by his computer.
 
looks like a joke played by a software development company ... seriously even normal softwares go through a set of standard tests before being implemented for use and for a sector where software malfunction could be as destructive as this , there are levels of testing which make sure there isnt a single major fault left.
 
it's just a testing problem, not an algo trading problem. It didn't go wild, it did exactly what it was programmed to do. Malfunction is an interesting way to put it.
 
i am new to quant invest but my impresion that most of prop shops hate variance as much as i do. are very disciplined when it comes to capital at risk and very disciplined whith their bet sizing. reading this i say lol. i guess edge was so gd that risk realocated capital from other strats.
 
Risk management at prop shops is hardly as stringent as you think might think, unforunately. If it was so important there would not have been so many quant fund implosions. ‘Quant’ Funds Try to Resurrect Computer Investing - NYTimes.com

Once something is successful it gets cranked. Even the 'gurus' at LTCM didn't manage their risk effectively,and failed.

Volatility is a good thing, if you didn't have any, you wouldn't make any money.
 
i definetely agree with you that volitile enviroment present opportunities. but in the context of the p&l distribution of trading strat i personaly wldnt think that its such a gd thing.
 
i definetely agree with you that volitile enviroment present opportunities. but in the context of the p&l distribution of trading strat i personaly wldnt think that its such a gd thing.

Volatility is a statistical arbitrage fund's best friend. The "choppier" the markets, the "more" mean reversion. ;)
 
i definetely agree with you that volitile enviroment present opportunities. but in the context of the p&l distribution of trading strat i personaly wldnt think that its such a gd thing.

Sounds like a poker player :)
 
lol idg. but Joy we are talking about different things here. i would rather flip a coin 1000 times at $1 as 4:1 favorite than once for $1000 as 4:1 favorite for the same price. unfortunately in life bad beats do happen often.
 
looks like a joke played by a software development company ... seriously even normal softwares go through a set of standard tests before being implemented for use...

Sorry, but I found this hilariously naive ;)

I'd extend Dominic's statement and go so far as to say that a vast majority of software isn't extensively tested... regardless of industry. It's easy enough to hunt around on the net for statistics about "bugs found per 1000 lines of code". Be afraid... be very afraid.
 
joel wsnt reffering to success of any particular strategy, but rather once a favorable situation is found how to manage ur bet sizing in order to capitalize on +EV and not get ruined at the same time. in the situation with Infin i am amazed that risk didnt cap capital allocated to that particular strategy and let pile up that much exposure. another interesting case is CFM that ppl compare to rentec of europe. while the firm does stat arb vol arb, manage CTA and their guys publish very hard papers they lost 10% of fund in some shady IL based cash management comp.
http://www.thehedgefundjournal.com/magazine/200905/profile/capital-fund-management.php
 
Yeah I know what you're saying, I just think it's not surprising, most firms don't manage risk properly because they're greedy, and code doesn't get tested because they can't spend the time, and it's probably written by one guy anyway and he is the only one that understands it. I think that goes on most everywhere.
 
Sorry, but I found this hilariously naive ;)

okk .... maybe you need to first know the differences in implementing a software for an active and running stock market and that for a library management system. :|


regardless of industry. .

To be precise, there was a subject in my engg. dedicated to atomic transactions which are necessarily characteristic of some particular industries (banking is one ;).) The requirements of a software vary hugely depending on the industry, time and the costs of testing. It's basically a tradeoff between cost and efficiency.

I find it extremely naive of you to say softwares are tested regardless of industry. :P
 
Back
Top