I've noticed many bad ticks (ticks which spike far outside of the bid ask spread, don't trigger limit order fills) do not ever get removed from their historical data. Specifically on Nov 19th at 14:32pm, SPY spikes down to a low of 120.00 on their data, though I can guarantee that did not happen on the market. I've been holding out hope they they buy data from somewhere and scrub their historical data, but its been 3 weeks since that bad tick and it still hasn't been fixed. The negative of all of this is bad ticks add an unbelievable amount of unrealistic performance to backtested systems in TS, making their historical data somewhat useless for designing and backtesting systems. This surprises me because TradeStations reputation and the fact that the software is mainly designed for creating backtested strategies. Maybe they leave the bad ticks in there because it makes systems designed in TS look more favorable, possibly increasing or maintaining the user base of their software. Does anyone any have insights into this problem? I may start shopping around for my own historical data provider in the new year and write a backtesting suite myself. I'd love to find somewhere cheap like OpenTick was, if anyone has any suggestions.
Unfortunately Ctrl-R does not fix most bad ticks. Such as the one from Nov 19th I listed above, which is consistent on fresh boots. And should still be visible to you if you look.
Instead of bitching here, where nothing is guaranteed to get done about it, how about posting on the TS support forum under the Market Data topic and see what their data integrity team tells you about it.
Does TS support tick filters? Maybe you just need to add one. With a tick filter ticks would remain in the database but would no longer appear on charts...
One school of thought is that it's better to use unclean / unfiltered backdata for testing, because in real life you will face errors in streaming data. So your backtests should include functionality to respond to apparent bad ticks.
How would you filter the data to only remove bad ticks and not good ticks that look like bad ticks? Seems like that would cause as much divergence from true price action as the bad ticks did in the first place. My interest in backtest data is to get it as close to live performance as possible. While bad ticks do appear in Trade Station during live trading, they do not really affect or trigger any limit orders running live, meaning that the closest simulated performance to my live performance would be obtained from running perfectly clean data. Was already in contact with their data integrity team. Their opinion is basically 'it is what it is'. Saying that the bad ticks get sent to them from the exchanges and there is nothing they can do about it except for some data filtering they apply which still allows 5-10 really bad ticks per day on SPY, and most do not get removed from historical data. So I am not so much 'bitching here', but I am looking for insight into the issue beyond what TS / TS Forums can give me, and getting ready to discontinue the use of TS. So I am looking for a good alternative data provider, Elite Trader seems like a good first place to ask. Thanks for your response.
Many users <i>don't want</i> marketdata vendors to manipulate the exchange-supplied data, preferring instead to see the data exactly as declared correct by the original exchange. Such users would prefer to make their own choices about filtering / cleaning., since there's no one correct way, and would prefer not to lose access to the original, as-supplied, dataset.
I have a 'clean' L2 feed coming in beside TS, unfortunately it does not provide historical data. The bad ticks I see coming through TS do not appear on the L2 feed. So I somehow doubt TS is constructing their charts from the same data I am getting. It would be very easy to filter live data coming in for these bad ticks myself based on the bid ask prices shown by various exchanges, the issue is that it wouldn't help with my problem with inaccurate back test data, since I wouldn't be able to filter the historical data since I don't have a historical record of bid asks.
Olsen's book http://amzn.to/fbpawG discusses in some detail how to clean data in the absence of a corroborating source.