Market data lag

Discussion in 'Data Sets and Feeds' started by topdog, Apr 27, 2022.

  1. topdog

    topdog

    Yes. I think it's mandatory that you have exchange generated timestamps. I still would calculate the deltas between timestamps and my h/w clock ticks. I think this will give me a better accuracy. The exact times aren't so important in this situation. It's more a question of lag.
     
    #21     Apr 28, 2022
  2. easymon1

    easymon1

    You'll need a chronosynclastic infundibulum to obtain extreme accuracy and precision - oh, and a large budget.

    https://en.wikipedia.org/wiki/Flash_Boys
    "...the faster a market participant's computer system can receive and act on data, the better their edge, and opportunity to profit, with even nanoseconds making a difference."

    The first chapter tells the story of a $300 million project from Spread Networks that was underway in mid-2009—the construction of an 827-mile (1,331 km) fiber-optic cable that cuts straight through mountains and rivers from Chicago to New Jersey—with the sole goal of reducing the transmission time for data from 17 to 13 milliseconds.[6] (The construction of the line was dramatized in the 2018 film The Hummingbird Project.)
     
    #22     Apr 28, 2022
  3. topdog

    topdog

    I think you don't have to. You just have to look a time delta between 2 timestamped actions and compare the time with your h/w clock ticks.

    For example: You will receive a bid change 12:22:20.150 and the next will come 12:22:20:160. Because the timestamps will come directly from the exchange you know the time spent between these actions is 10 ms. So you should receive the information within 10 ms if not then the excess time will be the lag.

    You of course have to optimize your local h/w for the speed so the lag you produce by your local h/w is more or less irrelevant. There is many ways to do this but it's a different topic.
     
    #23     Apr 28, 2022
    easymon1 likes this.
  4. NorgateData

    NorgateData Sponsor

    No that's not the correct definition of lag. That simply represents the time between quotes. For example, even if you're on the other side of the world, the time between quotes will be 10ms. You could be on the moon and the time between quotes will still be 10ms.

    Assuming your PC's clock is correct and synchronized to a known good nearby NTP server, the lag is calculated as:
    Your PC clock when you received it - exchange timestamp

    This will determine all of the lag elements between the exchange and your PC.

    Perhaps another way of thinking about "lag" is pretending you are on the moon and someone on earth sends you a radio signal. Assuming you have the greatest equipment available, even at the speed of light, it takes 1.3 seconds for that signal to reach you, so the lag is 1300ms. So a transmission sent at 12:20:20.150 will reach you at 12:20:21.450. A second transmission sent at 12:20:20.160 at 12:20:21.460.

    You should also monitor for jitter (basically the variation in lag) indicating that some intermediate point is congested, especially at busy times (open and close).
     
    Last edited: Apr 28, 2022
    #24     Apr 28, 2022
  5. rb7

    rb7

    You just need to compare the exchange time stamp on the trade or bid/ask message that you receive with the time at which you receive them.
     
    #25     Apr 28, 2022
  6. topdog

    topdog


    No.

    In my case I am not interested in the lag produced by the internet or connections. I am interested in how much overhead will the data provider do. How good or bad is my data provider.

    Then by knowing that the quotes must come in within the 10 ms is a crucial part of the validation. Also this is why I am colocated to minimizing the connection delays. My connection and processing overhead are minimized into a level where they are irrelevant.
     
    Last edited: Apr 28, 2022
    #26     Apr 28, 2022
  7. topdog

    topdog

    Yes this can be done, but as I said it's better to do by comparing delta times. It will give you more accuracy. No need to sync your local h/w clock because you are counting it's ticks only.
     
    #27     Apr 28, 2022
  8. NorgateData

    NorgateData Sponsor

    I respectfully disagree.

    You're examining jitter with your method. It's quite different to lag (latency).

    See my post above about the moon analogy.
     
    #28     Apr 28, 2022
    rb7 likes this.
  9. topdog

    topdog

    No again.

    Jitter is a variance of latencies.

    I am not calculating that.

    I am calculating how long does it take to receive quote OR trade prints from the exchange -> data provider overhead will change that. I of course accumulate calculations and then average the results to get the answer.

    After examining and testing things I now understand that my method works 100% right.
     
    #29     Apr 28, 2022
  10. topdog

    topdog

    Thank you very much for your answers. Was very helpful and helped me to find a solution to my question. Happy trading for all of you.
     
    #30     Apr 28, 2022