Big Bars are hard

Discussion in 'Automated Trading' started by EliteTraderNYC, Nov 15, 2012.

  1. Hey guys, how do you deal with trades with big bars with super wide volatility? Ignore the trade? Take the median position, etc?
  2. One trade is a singular event, I do not understand what you mean by a trade with a "big bar". How can one event have a big bar? A trade is a single transaction.

    Do you mean how to determine a fill in a backtest?
    Or maybe you mean if there is a large price shift within the time frame represented by a single chart bar which messes up your indicators?
    In this case I don't see why you should treat it in any other way than all the other bars. The OHLC is still OHLC...
  3. Yes, a backtest fill, something like that. lets say a stock has a range right around $17, but a big bar spikes that upto $20 but then it just goes back to $17 range again. Does anyone try to normalize this info or remove outliers? How?
  4. What time frame are you currently using?

    If you`re using 1-minute bars, you could consider 30-second bars.

    Of course, there`s also volume and tick charts.
  5. I don't use bar-based systems for this reason, and instead construct a sense of value using bid/ask + trades instead.

    There's only one thing I like bars for, and that's for just getting some sort of cleansing of time-series for relative-value trading systems, but even then, they are sort of sloppy. The more illiquid the stock, the sloppier they become.

    But if you insist on using bars, the question you have to ask is why that bar is there. If there was a strong, high-volume event vs. a gap, etc whatever you have to figure out why it happened. You shouldn't shove it under the rug as an outlier.

    I would go so far as to purchase the slice of data for bids/asks+ticks during the suspicious bar in your system.
  6. Hmm, if by range you mean price range (price is usually around 17) and not delta price range (price is around 1700 and ranges +/- 17), then such outliers are probably due to unclean data. In your example that would be a 17.6% price change (17 to 20) for just a short time, and that is probably not something that really happened.
    For example with real time IB feeds you can get a High or a Low that is even 50% or 100% above/below the price, however those are usually mis-ticks from option pricing data. To solve it, get some clean data.

    If you want to clean it yourself, figure out what works for you. Perhaps a tick that occurs more than 2 std dev above is a good criteria to delete it and give the bar sensible values. To do that, it depends on what software you are using.

    Sorry if I misunderstood, I hope that's what you meant but you might wanna be clearer.
  7. I may just remove these types of trades from my trading for now and reconsider them when I can.
  8. this is an issue every trader including myself deals w/ b/c there is no way to tell at the time whether the spike will stop at 20 and not go to 25 or 30.