ActiveTick vs. IQFeed vs. IB (TOFTT)

Discussion in 'Data Sets and Feeds' started by SeattleAlgo, Sep 6, 2012.

  1. Totally agree. The "serious trading" was meant to be something along the HFT lines, but then it turns out that neither of these feeds would be even remotely good for that.

    I originallly promised a report on latencies, and, unfortunately, I am still struggling with getting a working timekeeping card (had to RMA the first one I got). Yet during those few days when I had it working (with a 8 ms / hour drift), the latencies I observed were the following:

    During the first 2 hours after market open:

    IB (trades only): Range: -480 .. 1000 ms, Average: ~550 ms, Std: ~300 ms.

    AT trades: Range: 60-10,000 ms, Average: ~450 ms, Std: ~550 ms

    AT quotes: Range: 35-10,000 ms, Average: ~300 ms, Std: 600 ms

    (My local processing delay is less than 2 ms per tick).

    These numbers are for a dozen or so of mid-liquid (~$50M daily volume) stocks from both NASDAQ and NYSE (those just happened to match my trading filter for a particular day).

    Notice the negative number on IB timestamps. Makes one wonder what the IB timestamps really are and at what point they get stamped.

    The average is very close between IB and AT, IB being a 100 ms slower than AT trades, the latter being another 150 ms slower commpared to AT quotes. Still, AT has a wider deviation, though AT timestamps are probably more "real" than those of IB. But for both of them one can forget about any sub-100 ms trading or even sub-500 ms trading. Starting with 300-500 ms quote latency and addding another 100-200 ms of order handling brings it to a 1 second timeframe - must be an eternity in the world of HFT.
    #41     Oct 30, 2012
  2. Of course it is, but how many independent traders are willing to go up against the CO-LO guys from the serious HFT firms ? They've got such an advantage ! Plus, It's getting crowded too.
    So why not take a different approach with pairs trading or stat arbitrage or options arb ? 100 or 200 ms doesn't matter with those strats.
    #42     Oct 30, 2012
  3. vicirek


    IB does not return time stamp for "tick" price sampled data. Real time bar returns time stamp of the start of real time bar period. Since they create bars internally I would assume it is their server time. I do not think IB is very concerned abut ms time accuracy because their feed is not geared for HFT and has to satisfy variety of customers and most importantly keep their quote board active in busy markets. I am not quite sure if there is another timestamp they would provide.

    Thank you for sharing your results. It is always very informative.
    #43     Oct 30, 2012
  4. Where is your server geographically located? It takes about 100ms to cross the continent. I was able to receive QuantQuote TickView data with about 75ms latency (based on exchange timestamps) out of Boston. But that computer was in a datacenter which might have had a direct line to NY/NJ. I think the bulk of this latency could be due to geography/routing. You should be able to nail it down by running a ping.
    #44     Oct 30, 2012
  5. sle


    The B-firm. Pretty much because they got everything and have really good support.
    #45     Oct 31, 2012
  6. B-firm = Interactive Brokers??
    #46     Oct 31, 2012
  7. hft_boy


    Being in Seattle probably has something to do with it. From NJ, AT has about 150 ms latency (mean ~150, if I recall from my tests median is around 120, and sd is around 50 ms). IB order handling in my experience is about 30 ms (after packet latency). Note though that AT servers for whatever reason are in St. Louis, which you can confirm easily using a traceroute. So you're not going to get very good latency with them in any case.
    #47     Oct 31, 2012
  8. sf631


    Thanks SeattleAlgo. TOFTT is appreciated
    #48     Nov 1, 2012
  9. Possibly, though my ping time to AT server is 65 ms (and that's the round-trip time). My ping to IB is 90 ms. So it's not that bad. The stability of one's clock is key when trying to measure all these latencies - as I already mentioned my built-in RTC drifts by about -150 ms per 15 minutes. If I just use it for measurements, the numbers will come out way better. So it is important to pay attention to how the true time is obtained. (For the numbers above I was using my own NTP sync against a collection of geo-nearest Stratum 1 servers, the sync tends to be accurate and repeatable within ~3 ms), did it once before market open and then kept time with a free-running MasterClock OSC card (the latter still had a drift of ~8 ms per hour, which was however 2 orders of magnitude better than my built-in RTC).
    #49     Nov 2, 2012
  10. gmst


    B-firm = bloomberg
    #50     Nov 2, 2012