I don't think you understand the microstructure of the market. If the market is trading at 10.01 x 10.02 and the market moves to 10.02 x 10.03 then the new queue at 10.02 will not have any GTC orders and has a completely new queue. The strategy with the lowest latency will potentially be the first in line at the new queue at 10.02 (unless the move came from a large order that created the new queue, does happen more frequently in futures market than equities). And I'm sure you've had problems getting first queue position because you're slow. What kind of infrastructure were you using? Did you have a box sitting in a data center close to the exchange using a custom market data feed handler over infiniband maybe even processing things on an fpga or modified linux kernel? If you weren't or don't know what that is then you were too slow and your experience doesn't matter in trying to attest how good a position a low latency strategy is getting in the book.
I know exactly what you're talking about, and personally know traders at TMG and AF here in Chicago who use the order queue position automation strategies (much more so a few years ago than recently; the TMG guys are long gone and the AF group got hurt badly in 2007 and 08). I've also seen them get absolutely smoked in trending markets. The problem is that size turns the market and that size order rests until the substantial balance gets filled. It isn't free money - the market really has to be range-bound for it to be consistent. Paul Rotter used to do it with the Schatz.
Pretty sure you dont know what I'm talking about. Read those papers in my first post in this thread. HFT does not hold positions long enough or in sufficiently large size to get killed in a trending market. What your saying is anectdotal at best whereas those papers are using data provided by the exchanges which no one else has access to.
I have read your papers in detail, and I now realize that we are each talking about two completely seperate topics in the HFT realm. And the literature you cite has no bearing on order queue management, but rather high speed 'lead-lag' strategies where the system participates in bids or offers that trade through and then issues an immediate covering order with a position holding timeframe horizon of about ten to fifteen seconds.
I guess ya need to read a little closer. Pg.12 of "The Flash Crash: The Impact of High Frequency Trading on an Electronic Market" says "HFTs reduce half of their net holdings in 115 seconds". Page 11 of the same paper says "High Frequency Traders are consistently profitable although they never accumulate a large net position." Page 14. "First, HFTs seem to anticipate price changes (in either direction) and trade aggressively to profit from it." So the people that are taking the other side of their trades are having an adverse movement against them and thus are left with the 'bad' fills that the original poster you took issue with was saying. Page 14, "Second, HFTs seem to provide liquidity by putting resting orders in the direction of the anticipated the price move." This is where "order queue management" comes into play. They attempt to be first in line as the price moves so they can get fills from natural buyers/sells which are the 'good' fills (or better put from non-informed traders). And your statements fly in the face of all the evidence of the race for speed which is all about being able to get first in line when the price moves, taking liquidity prior to a price move, and cancelling prior to an adverse price move. And the paper is using data from the exchange which indicates which trades are from HFT so we can not be talking about different things because they are capturing it all in the paper. Unless you're talking about some super small players who have no impact on the market and just are not worth discussing.
"the race for speed which is all about being able to get first in line when the price moves, taking liquidity prior to a price move, and cancelling prior to an adverse price move. " It now appears to me that other than access to white papers and other academic references, you have no applied experience in the field of FIFO order matching in the regulated futures markets. You seem to be of the opinion that an ATS knows when there is going to be a price move, and so the ATS takes liquidity before such a price move. Likewise, you believe that an ATS knows when there is going to be an adverse price move, so it goes ahead and cancels orders beforehand. These markets in the FIFO futures are turning on size from single orders - and you can't race that. They don't peck away at it - it goes with a 2,000 lot. Yes, alot of the volume is ATS driven, but you don't seem to recognize the fact that it is largely driven by basket arbitrage and relative value spread trading - it doesn't necessarily rely upon liquidity in one market name.
Did you not just read what I had in my last post or the research paper? HFT's on average are informed traders on a short time horizon so do know on average when the price is going to move and thus are able to cancel/take liquidity/place orders based on that. That is how they consistently make money. It's not just opinion, it's fact based on exchanged provided data as shown in that research paper which the CFTC had a hand in. And I'm not going to bring in my personal experience in any meaningful way (I probably trade more size/volume in a day in futures/equities than you will in a lifetime, yes it's all automated) cause it's not worth losing my salary/bonus over. So using research papers with more than relevant information will have to do "Tradebot, one of the biggest high-frequency traders around, had not had a losing day in four years, he said." above is from http://www.nytimes.com/2010/05/17/business/17trade.html How ya think they do that? eh? These certainly are not those hacks you were talking about earlier that lost money in 2007/2008. 2008 was an amazing year for HFT. Was damn near impossible to lose money as long as you had half a brain.
Here is a screen shot of the VWAP function on bloomberg from 12/3 for ESZ0 The largest trade size is 1095 contracts. The 10th largest is only 563 contracts (and some of those are after/before regular market hours). The average trade size is 4 contracts. That is more indicative that the markets do not turn on size from single orders, how you can draw that conclusion is beyond me given that the CME does not provide an order level market data feed like nasdaq and the other equities exchanges do. I can whip out tick data screen shots from bloomberg if you still think it turns on size and show ya otherwise. And if I'm bored enough I'll whip out the millisecond time stamped tick data I got to show you that it doesn't turn on size (not even the CME provides kind of data http://www.cmegroup.com/market-data/files/FAQ_TS.pdf they provide timestamped to the second
It depends on what you mean by <i>order</i>. Taking 1000 from the offer side will typically show up as an assortment of small fills, because it's transacted against a scattering of resting offers of various sizes, mostly small. Also, of course, buyside traders will themselves shred a 1000 lot into tiny pieces, so the size of individual prints doesn't say much about who's moving size around.
Well note that I said it is more indicative not that it is the case for sure because there is no knowing given the information available since we can not see the orders. I contrast that with the nasdaq itch feed that actually has displayed orders in which case we could see order sizes. No way of doing that on the CME. And as you were indicating now a days buy side and even sell side will definitely be running a POV/VWAP/IS algo to work their order instead of just slinging a 1000 lot limit order out there which is what bone thinks is happening constantly. If 1000 lot limit orders were extremely common I would expect a higher avg trade size. Look at the standard deviation which is just about 3 contracts.