Here's an article that shows some hft orders at the, shall we say, sub-atomic level: http://www.theatlantic.com/science/...ts-the-tracks-of-bizarre-robot-traders/60829/ Anyone care to attempt to describe the strategies being used by these programs?
Those bear no resemblance to what I think of as HFT, which is usually just automated market making based on some sort of pricing model and mean revision.
I absolutely believe this is HFT, and exactly the market-making/mean reversion referred to above. They're sitting some std dev away from the current price, hoping to catch large volume orders that clear the books and subsequently take out multiple stop orders. There's no reason "market-making" has to imply living off of the ask/bid spread; you're offering backstop for the 2+ sigma moves that come from liquidity mismatches. And I agree with the Nuclear Phynance explanation. These are just the results of a dynamic system, nothing especially nefarious about it.
they're flooding symbols with ridiculous amounts of quotes knowing their competitors will have to process them, giving them an information and speed advantage because they won't be processing them. they shouldn't be at it for long though. exchanges really can't support that type of abuse and will catch up to them eventually.
heech, 5k quotes on a single issue per second? you need to reread the article. there's no mean reversion mm strat in existence that has a legitmate reason for doing that. only reason for doing that is if they're trying to outfox competition by doing what i described above. it's basically the finance version of a dos attack and exchanges really should ban members doing it.
You lost me. It'll take them more cycles submitting these orders, handling resulting order state... than it will take for other market participants to simply filter out orders that are far from the NBBO. I don't see how you can possibly gain any advantage from doing this.
Either testing for big market moving orders or just screwing with the competition. Either answer seems to be within the realm of possibility to me.
Asynchronous threading? Some strategies seek out liquidity like you say, which requires looking deeper than the NBBO. Typically its hard to filter out noise like that since broker API's expose events that fire whenever there is ANY change to the books, be it price or volume listed. So the "victims" have to handle it. So those price sawtooth patterns that are deep off the NBBO and only list 100 shares are nefarious. The ones that list big volume deep off the NBBO are cock teases, and/or potentially legitimately trying to fish for liquidity seekers, index arbitrageurs, etc. I can imagine how frustrating it must be for developers who have to try to capture that liquidity, only to watch the pattern change or move further away as they hit the NBBO to try and catch it.
hehe, yea, i'm sure they're doing everything from a single thread on one machine. given the sophistication and regularity of the attack, i would put money on them having boxes dedicated to doing this. the advantage lies in the fact that they compete in the subsecond arena, and if they can get their competition to waste ms processing bs quotes, then they get an advantage.
They would have to process their own quotes too. I bet filtering them would take about as much time as just doing some numerical calculations with them.