don't want to take away from any of hft's future answers, but thought i could give you a real quick answer. hft trades for an hft firm, so the idea that they'd use non-hft technology like deltix, doesn't really make too much sense, but even if we were to assume they would consider a more hft-centric solution like redline, you would have to really wonder how competitive they are. think about it for a moment, in any industry, would you really outsource your primary ip? as far as cep, that's not going to happen in an hft execution engine. think non-portable, use/gateway-specific, with very low level processor intrinsics and highly tuned kernels and/or pure hardware solutions, vs any kind of an abstraction layer like cep.
We use our own execution engine because everything is customized down to very very exchange-specific interactions. It's probably less a matter of latency (though that is also a factor, though you could argue some bare-bones 3rd party solutions can also be as fast or faster, depending the level of complexity it offers). I'm not familiar with the term CEP, but from looking up its definition, I'd say no. The execution engine really just translates exchange protocol into internal protocol, without making many assumptions or predictions. The more complex event processing takes place at the trading model level. Our developers have been doing this for a long time. I'd estimate about 1-2 man-months to develop an execution engine for each exchange. I'd rather not say how many infrastructure developers we have, but multiply those man-months by how many exchanges we have to add/maintain regularly and that's how we determine our staffing requirements. Though keep in mind much of the work is also customizing and improving current connectivity in addition to new connectivity requirements (both new exchanges we're expanding to and current ones that roll out new API's). Ballpark, I'd say each infrastructure developer spends 1/2 of their time firefighting and fixing current connectivity, and 1/2 on new connectivity or enhancement.
Another view point. http://www.bloomberg.com/news/2013-11-26/speed-traders-meet-nightmare-on-elm-street-with-nanex.html
Hft, You use AI (SVM / NN or another) technique for predict the market movement or you use essentially the speed / latency to win $ ?
Are there any hfts trading strategies that are higher profit per contract, as in getting closer to half a tick? Or are all legitimate edges being ground down to the 0.1 tick range?
among all the strategies you mentioned earlier, what would be the best HFT trading strategy that has never failed, and most new HFT comers trade? i.e. is it inter-exchange trading?
hft, Do HTF shops usually trading through a broker, or do they just buy a seat on the exchange? Thanks!
There is some AI involved in some strategies - it varies a lot across all of the strategies we run. You need both some smarts and speed in all strategies, but more of one or the other depending on what you're trying to do.
There are some that come to mind that makes handfuls of ticks each time, but that's primarily due to the tick size and liquidity of the market (metals on EBS for instance). Might only get filled every now and then but makes a handful of ticks each time. To say that all edges are down to X ticks is too broad of a generalization.
"Never failed" is a bold statement. I'd say the most reliable strategies that mature HFT firms run are inter-exchange arbitrage and spot-futures basis spreading, at least from a MM standpoint. Not sure what market-takers count on as their bread and butter, but I'd venture to say inter-exchange super short-term momentum/mispricing plays.