What does a HFT data feed actually look like?

Discussion in 'Data Sets and Feeds' started by Rocko Bonaparte, Apr 23, 2010.

  1. Out of my own curiosity, I wonder what kind of data an HFT system would have to work with at all. Are we talking something at a level even more basic than level 2 data?
  2. thstart


    colo on the exchange=low latency - the only difference. you get a "license" to look milliseconds before "others" what big orders there are, with "flash" orders you scan the market for available shares - when there is a big institutional order you have milliseconds to get the (possibly all) shares, create artificial demand, then offer these shares to the institution at higher price. of course this is very rough description but basically it is that. SEC posted a request for comments to change that.
  3. So the feed by itself would look like a level 2 feed perhaps shifted in time?

    I'll preface all this with the understand that if I was a supercomputer or whatever and I touched that data feed at a given time the outcome would wind up really different, but I was generally interested in what I would have seen--as a supercomputer.

    It would be a fun little technical project to take a recorded level 2 feed and have a program that can replay that over a socket. Or do they already have stuff like that out there?
  4. Openecry has a market replay feature that records and replays level II data.

  5. thstart


    all good feeds have a time stamp. the problem is "when" you get the data in "real time". If the time stamp is supposed to come from the exchange, the latency will cause you actually to get the data later. now if you do a market replay later you will see what happens if there is not delay (you are "in" the market (computer)). if that is so I am asking what is the reason to research tick by tick data in a market replay, etc. when in reality you will have a delay because of latency. minute by minute will be close to reality milliseconds being negligible to minutes. much less data to process.

    one possibility to play with tick data is to factor in the latency and "anticipate" what will happen. this is a possibility but the market fragmentation makes it more vague. that is one more uncertainty factor to deal with. and when time comes to "act" you order will come with delay to the market. so the milliseconds game makes no sense.

    Now I see the things not changed much from 100 years ago. The bucket shops actually "delayed" the data to customers with several minutes and knowing the actual situation acted accordingly. Now this delay is hundreds of milliseconds or so but the principle is the same.
  6. thstart


    ...something with the editor happened...
  7. thstart


    here you go:

    Underlying Problem: Undisplayed Trading Centers Compromising the NBBO through Sub-Penny Trading

    An abusive strategy that has been occurring with increased frequency is a practice called “subpennying”. It is the practice of a market participant stepping in front of a displayed limit order by a fraction of a cent.
    The explicit purpose of this strategy is to preempt the NBBO.

    Bottom line - don't trade thinly traded stocks. trade more liquid stocks like DJ30.

    "This problem does not exist in the most actively traded issues. This could however become a serious problem, if the public exchanges are allowed to quote in sub-pennies as well."

    "Sub-Pennies for Everyone – Not the Answer
    The public exchanges have recently disclosed their interest to quote in sub-pennies. They need a level playing field to compete with the undisplayed market centers (broker-dealer internalization, and dark pools), so they are proposing a move to displayed quotes in 1/10th of a penny increments."

    "But now an even bigger problem is created. All the algorithmic systems can now step in front of your order by as little as 1/10th of a penny in the displayed market centers as well. Currently displayed algorithmic systems can only step in front of your order by a full cent. This increases the number of prices that a displayed algorithmic system can step in front of the NBBO by a product of ten."
  8. So the HFT guys are basically using flash (fill or kill) orders to look to see if there is hidden liquidity out there, then jumping in front of it? But much faster than anyone not colocated could do?

    One solution would be to level the playing field for quotes. Colo all you want, but we are going to delay your quote feed so that you get them at the same time as retail, measured by the average latency of Reuters, Esignal, DTN, etc.
  9. thstart


    So, what is the active trader to do?

    "Clearly, any strategies that have an edge based on speed are out the window with the increase in high frequency trading. While the active trader used to front run the order of the institutional desk that was inefficient in execution, now even the small trader’s order is front run by the computer algorithm. Every human trader is now the inefficiency with their slower execution. Entering and exiting stocks will also be tougher. Any active trader is quite used to seeing his order front run immediately as he shows his bid or offer making it more difficult for him to get a fill. There is a high likelihood that active traders must become used to paying an added toll to HFTs for entering and exiting their positions.

    The trading business is forever changing, that we know for sure. Level II strategies based on speed of execution are certainly on the decline. Active human traders must therefore become more sophisticated. First, minimize the impact of HFT by trading “in-play” stocks that have large volume from “real” players. Second, avoid non-volatile stocks trading below average volumes. Third, greater anticipation based on sound technical analysis is also needed. Most of us will need to fight hard for better prices and avoid the temptation to buy highs or short lows as algos are programmed to manipulate prices around these areas. Fourth, many of us will need to cut down our size and look for larger moves in stocks. Scalping very small moves is not nearly as profitable when a predatory algo scalps 3 cents from you on your buy and another 3 cents on your sell, just as a hypothetical. Also, levels in stocks are not as clear-cut because algos are programmed to push stocks through the level to shake out weak holders. But, if you can begin trading for dollar moves on less size, you’re less likely to notice the 6 cents you paid as a toll and you’ll be able to give the stock a little extra room around levels.

    In order to successfully navigate through the choppiness that HFT has brought into equity markets, traders must spend an increasing amount of their after-hours time researching and learning levels. Spend more time analyzing charts on multiple time-frames. For traders who focus on very small timeframes, now might be the time to take a step back, decrease size, and look for setups and levels on higher timeframes. The higher the timeframe, the more powerful the setup and level and the harder it is for an algo to overtly cloud the area. Additionally, familiarity as to how particular stocks trade around levels helps provide the confidence necessary to follow-through on your ideas. Traders need to develop a universe of familiarity—a core group of “in-play” stocks and sectors—to follow each and every day.The more often you we trade a particular vehicle, the more familiar we become with how algos work in that particular stock.

    These are not fail-safe rules but HFT is a reality and it is here to stay. Active traders must adjust and come to find a new edge beyond speed of execution. Where there’s movement, there’s opportunity and the survivors in our business will become more sophisticated in order to continue trading profitably."

    DJ30 is an obvious choice.
  10. thstart


    I would not wait for regulators to do what is needed. Rather I would adapt. One route to minimize the impact of HFT is to adapt the technical analysis to 21st Century. Software with 1,000 "indicators" on stock by stock basis with inflexible time frame management don't works.

    That means - new, faster analysis tools, much more sophisticated. That is what I am working on.
    #10     Apr 23, 2010