Bid/ask historical datas

Discussion in 'Data Sets and Feeds' started by TraDaToR, Apr 14, 2007.

  1. TraDaToR



    I'm working on some new strategies and it would be suitable for this kind of strategy to use historical bid/ask datas instead of OHLC.

    Where can I find some of these datas? I am looking either on continuous bid/ ask datas or end of bar bid/ask datas.

    Otherwise, is there a mean to extract TS bid / ask end of bar snapshot datas and put it in a txt file?

  2. squeeze


  3. TraDaToR


    Thank you very much. Do you prefer one of the above provider?
  4. That is only tick data, which isn't enough to reconstruct a limit order book.

    I have yet to find any data vendor (including the exchanges) that provides enough historical trade data to reconstruct the full book in continuous time. This is unfortunate because you really need the complete depth to accurately backtest anything (tick data is useless as you can't calculate price impact, spread, crossing...)

    I recently started developing an order book reconstruction system (sort of an exchange simulator), but this is an extremely complex task and I'm only working with Nasdaq stocks at the moment. Eventually I'd like to open it up to other traders who realize how useless TS and every other backtesting program is.
  5. OpenTick claims to have historical book data. I haven't looked at it yet so no idea of how good it is. Free.
  6. squeeze


    Back testing has inherent limitations.
    If a strategy is traded in a size that significantly impacts the market then that inherently compromises the validity of back test results.

    You are also into the law of dimishing returns. Using level1 instead of interval data gives greater accuracy but the the cost of vastly increased data size and slower calculations runs. Using level2 is another big hike in data size and processing cost but with further diminishing real benefit.

    However, strongly agree that commercial back testing software leaves a lot of room for improvement and would be interested in seeing something better. I think the problem is that handling the size of data for high res back testing requires coding in a real language such as C++/C# to get the perfromance. This makes it difficult to use and pretty much limits use to quants/programmers and mainly to institutions that can afford this sort of set-up.
  7. TraDaToR


    rich idea. This must be extremely hard to do.