by Doug Coulter » Wed May 07, 2014 12:49 pm
Yeah, i'm pretty old-skool as a programmer. ASM, C, C++, and now perl, since it does what I mean with the least work (for me). I'd have used that to suck up the data stream and parse the XML, since there's a perl module (actually, a few) on CPAN that do that.
The issue is finding a realtime stream that can be afforded. Interesting the timing in the TD stopping theirs - just before the book revealing the HFT junk came out (we real traders all knew anyway, but we're not the world). I've been watching the Nanex stuff on ZH for a long time - it's pretty crazy at the "fantasecond" speeds. Trades sometimes get timestamped before the bid and ask...
Now, I don't need that - and in fact you can exploit it sometimes - a little flash crash can get you a buy at a real good price, and recover in seconds - and you sell perhaps the next day (so don't get day-trader flags that cut down your other privileges).
But if it was huge, like the flash-crash everyone knows about, and either Goldman or JP Morgan lost money - they reverse all the trades, and you lose. So you never set a limit order over X% below the market, so if you get it, the trade won't be "busted". Never set a trailing stop - it's red meat to those guys, they see it, jam the price down at an il-liquid moment, buy your stuff cheap before things recover, you lose. EG the other side of that same trade.
I think a fairly (well, conceptually anyway) simple combo of a couple heuristics with linear and nonlinear (neural net) prediction will do just fine and beat the market like a red headed stepchild nearly all the time. You are after all, now trading against some really stupid programmers, and their human owners (which is why "it's not different this time" - humans will still pull the plug on the machines if things look wrong). Look, it was only about 16 months ago these super-slick college grads who couldn't get a real job in their field learned how to program a stairstep function.
I on the other hand, literally wrote the book on digital signal processing. I go up against them, it's their ass that's grass, end of story. The reality is that when you have several of these guys messing around - their own algo's aren't actually stable in the Bode plot sense, but when you have 5-10 things with gain and time delays hooked together - eg, the markets - you get an oscillator unless they are all far more clever than they actually are. So, not that difficult to time, perhaps.
But again, you need the real time feed, and you need that and history for at least 100 or so stocks worth trading....and at a price that doesn't eat all your profits if you don't have zillions to trade with. Anybody who finds that for sale at a reasonable price is automatically my hero.
Posting as just me, not as the forum owner. Everything I say is "in my opinion" and YMMV -- which should go for everyone without saying.