needed: an expert automated trader who can advise on data feeds

Discussion in 'Data Sets and Feeds' started by perkindiafrawl, Sep 5, 2007.

  1. Hi,

    I have many detailed questions about data feeds for use in an automated trading system (ATS).

    I'm implementing an equity ATS with RedSky (now ITG). My box is located at the brokerage. I have the option of subscribing to various data feeds, one of which is the NBBO, and the rest are from individual exchanges/ECNs. I do not understand how I should be using the NBBO feed in my ATS, or how the exhcanges use the NBBO themselves. Is it wise to listen only to the NBBO? What about only the exchanges? What about a combination? ...and how are the exchanges bound by the NBBO?!

    A second major source of frustration is bad data. I have logged clearly eroneous data from individual exchanges. Furthermore, I detect crossed quotes on even the NBBO. When trying to combine data feeds from many exchanges + the NBBO, a single bad feed can spell disaster. It seems very challenging to determine the criteria by which to judge if incoming data is "bad" or what to do if it is!

    I would like to talk to an elite trader... someone who's been down this road and can speak from experience... someone who knows all the dirty interworkings of the NBBO and crossed quotes! Someone familiar with trading on ARCA, BATS, and Nasdaq via an ATS.

    It's asking too much for replies back in this level of detail on this forum.

    Would someone be interested in a brief phone call (half an hour or so) to chat about these issues? I will gladly paypal you for your time, as I expect some clear adivce on these issues could save me from a lot of mistakes.


  2. I would first start out by subscribing to all exchange/ECN feeds. I suppose you have DMA to the various ECNs thru FIX? I never listen to the NBBO; I construct my own centralized feed structure based on the multiple feeds. The issue of data filtering arises in this process as you have to deal with bad happens all the time. This is actually the trick to many system is how you handle bad data in your data sets....I have seen Kalman filtering used along with mult-stage feed forward neural nets, this is where you become the artist. The few items listed above should be a great start for you on your project.

    You can usually detect bad prints by trying to edge into the suspect print on a dark pool if the stock typically liquid. Some of the illiquid stocks (typically preferreds) tend to have very volatile price action so the art of determining good from bad becomes more difficult here.
  3. hi perkindiafrawl,
    you have good questions and not trivial ones but you have to provide more information about what you really need, for example do you need level2 quotes or only level1 and do you care for volumes on quotes or only prices and will you be trading NYSE or NASDAQ and liquid or illiquid stocks.

    Providing that information might make it easier for you to get answers to your questions.
  4. Hi,

    Here's a little more info:

    I'll need level1 quotes of the current best-bid/offer on the various exchanges/ECNs as well as the best-bid/offer quantity. I do not need depth quotes (LII).

    I'll be trading AMEX or NYSE listed stocks with total daily volumes typically over 600,000.

    I'm interested in trading these stocks on BATS, ARCA, INET, AMEX, and NSDQ. It is exactly these feeds (perhaps plus the NBBO) that I'll be listening to.



  5. Why don't you talk to feed professionals? Indeed they will try to sell you their services in some way but they also should know answers to tricky technical questions by heart. QuantHouse FeedOS, is focused on low latency market data delivery for automated trading.

  6. Hi perkindiafrawl,
    The easiest way is to just listen to NBBO and ignore all the others (works just fine if you are trading liquid stocks with tight spread) but you have to notice that the sizes on the NBBO is not the aggregate of sizes of all exchanges at the best price, but if you require to have the exact size of level 1 then you will have to do everything yourself which is reading all the exchanges and ECNs and handle them individually and aggregate them.

    you will not find a quick easy answer to what you are looking for as there are many issues you will need to handle like crossing quotes and bad prints.
  7. rosy2


  8. RedDuke


    Hi Perkindiafrawl,

    Any reason why you need ATS on equities and not futures? If you get interested in futures, I can describe several options with clean datafeeds to you.

  9. txuk


    RedDuke, please post the feeds that you recommend. I've looked at numerous futures datafeeds over the years to feed some customized tools I've built, but each has had limitations leaving me looking for something better. Thanks
  10. Thanks for the suggestions, guys. I'll check out those services.

    I don't know anything about futures... maybe some day. :)

    I have been combining BATS, ARCA, INET, NAST, and AMEX to get my own BBO... But I am concerned because it does not agree really well with the NBBO. And when I backtest on the combined data, I get very different results.

    Does anyone know how exactly the NBBO is calculated? What exchanges does it search??

    When I detect a crossed quote on an individual exchange (e.g. ARCA), I throw it out... it doesn't make logical sense how a single exchange could be crossed. But I've been keeping crossed-quotes that show up on the NBBO since it spans exchanges. Do you guys think this is the way to go?


    #10     Sep 11, 2007