Codenfreude

[hirez]

You know, as J. Random Hacker, one gets used to the self-congratulatory wank that spurts out of Hackernews, anything brogrammer-related and the gak-fuelled 'rockstar' 'coders' who go off to work in banks.

Because I am at heart a gleefully malicious sort of bastard, this made me point and laugh. $440 million spunked away by some shit code gone over-centre and into some nightmare destructive feedback loop.

You know all that malarkey about being all hard and Klingon and testing being for the weak-spirited?

It's a joke, son.

[/Foghorn Leghorn]

On the other hand, one must congratulate the comrade-coder who was a deep cover operative for so long within the capitalist machine! This is how we shall destroy the forces of capital!

Comments

[aidan-skinner.livejournal.com]

50p and a pound of grapes says this was down to one judiciously placed punctuation mark changed at the last minute with hilarious results

[hirez.livejournal.com]

"But I didn't chan... Oh. Shit."

[aidan-skinner.livejournal.com]

I do kind of wonder if the algo wasn't a desk traders excel sheet running on something ridiculous wired right into the market. Never before has the line between columbian marching powder and your pension scheme been so thin.

[hirez.livejournal.com]

I would begin to worry about the sort of Excel compiler that seems to be able to run at 40 trades/sec...

... Actually, that's pretty slow.

Anyway. Some number of months ago, I went poking around for signs of the actual code, and a thing that I found was called Quantlib. It's C++.

(Mind, I don't really know what and exchange platform looks like, so, er, Pontrilas)

[d-floorlandmine.livejournal.com]

Indeed.

[drumiller.livejournal.com]

What you mean brogrammer-linesnorter-howlermonkeys shouldn't be given unfettered access rights to HFTs, just because they had one good quarter of trades on the exchange and wrote a couple clicky-box 'programs' in their 200-levels at Stanford?

Personally I think we should have a human reaction time limit to trades, but I'm a shakefist sort when it comes to how fubar'd the global markets are.

Watch this space for an irony announcement shortly, tho.

[hirez.livejournal.com]

I'm sure I read a thing that showed that your average hedge-fund manager could be replaced with a noisy diode and some perl ruby, since picking stocks randomly demonstrated the same overall performance.

Meanwhile the HFT stuff is just arbitrage. Turbo-nutter-bastard arbitrage, admittedly.

I'm waiting for it to turn into core-wars. Well, I say 'waiting'. I mean 'Oh god fuck off you wankers'

[badnewswade.livejournal.com]

Pretty sure it's already there:

http://www.theatlantic.com/magazine/archive/2010/07/monsters-in-the-market/8122/

At least a few high-frequency traders have learned to make a killing by detecting the more simplistic algo strategies deployed by basic pension funds and mutual funds, buying the next stock the funds plan to buy, and then selling it to them at a higher price.

Then there's this:

http://coffeebreak.hiq.se/2012/03/23/be-quick-or-be-dead-high-frequency-trading-part-2/

Quote stuffing: The idea is basically to gain an edge towards your competitors that are also in the HFT business. By flooding the order book with orders that are quickly cancelled, one can create a situation for the competitors where there is too much information to process and they lose precious time.

Then there's the Timber Hill case, in which two guys from Norway almost did time for "exploiting" a robot:

https://algosandblues.wordpress.com/tag/peder-veiby/

(Spoiler: they got away with paying back the money in the form of a fine).
So yeah, as humans are edged out of the market, eventually just about every short term trader will be using algos.

[quercus.livejournal.com]

Did you catch Dave Cliff on R4 this afternoon, on just this topic?

[hirez.livejournal.com]

Ooer. Which programme?

[quercus.livejournal.com]

World at One, or PM, or one of those.

[christiffer.livejournal.com]

The best guess is the accidentally deployed a performance test harness along with the real trading algo, which quietly sat trading away in the background until somebody noticed...
http://www.ibtimes.com/articles/370872/20120806/knight-capital-happened-nanex.htm

[hirez.livejournal.com]

It's just... I mean, we've all been around when test-data has escaped or test-code accidentally a production system, but what kind of loopy netherworld do you have to live in for something like that to happen?

Anyway. I am becoming fascinated by how all this gubbins glues together.