I've always used TickStory data to run tick-by-tick backtests in MT4 till today when I launched the tester strategy with the control points model and the outcome surprised me.
When I launch a tick-by-tick BT I use the FXT file packaged by TickStory during the export process (e.g. EURUSD5_0.fxt for EURUSD on M5) and it's expected that the only warning message produced by the TestGenerator in the log is that the FXT file already exists and that the modelling quality factor in the Strategy Tester Report is 99,9% (of course, it was embedded in the FXT file by the TickStory export process!).
The first time I launched the BT with the control points model the story was different: the Test Generator had to generate the FXT file (e.g. EURUSD5_1.fxt for EURUSD on M5) and the surprise was that it reported several unmatched data errors, not only for volume limits exceeded but also for highs/lows mismatches. The surprise comes from the fact that I was expecting that the HST files (and the FXT file of course) generated by TickStory during the export process had data consistent across all the timeframes. And data consistent across all the TF would not produce unmatched data errors.
Is my expectation wrong or is there a problem with my data set?
If my expectation is wrong, what does 99,9% data quality really mean? To me that was the measure of the consistency of the data across the different TF (including tick data), or is it just a flag saying that I'm using real tick data, i.e. not artificially generated by the Test Generator?
To support my observation, I attached a zip file with the logs and the reports of 2 BT on the same set of data, one with the tick-by-tick model, one with the control points model (first run, subsequent runs reuse the same FXT file and don't display the unmatched data errors).
Are Dukascopy data consistent across TF?
Are Dukascopy data consistent across TF?
- Attachments
-
- Unmatched data errors example.zip
- logs and reports of 2 BT on same data with tick-by-tick and control points model
- (16.24 KiB) Downloaded 684 times
Re: Are Dukascopy data consistent across TF?
Hi Puglios,
As you're aware, Tickstory is designed primarily for use under the 'Tick based' modelling option. Given that MT4 attempts to produce its own data (FXT) for the other modelling methods, it is possible to get data mismatch issues. Volume data, for example, is currently generated from fractional bid/ask volumes which will not necessarily match on every time-frame due to rounding, etc.
FYI, the modelling quality of 99% is just an arbitrary number which is embedded within the data file.
Hope this helps.
As you're aware, Tickstory is designed primarily for use under the 'Tick based' modelling option. Given that MT4 attempts to produce its own data (FXT) for the other modelling methods, it is possible to get data mismatch issues. Volume data, for example, is currently generated from fractional bid/ask volumes which will not necessarily match on every time-frame due to rounding, etc.
FYI, the modelling quality of 99% is just an arbitrary number which is embedded within the data file.
Hope this helps.
Re: Are Dukascopy data consistent across TF?
Thank you TickStory support team, your answers confirm my guesses: the 99% is an arbitrary number and the HST files downloaded from Dukascopy are not 100% perfect. In particular not only the files have data mismatches between M1 and higher TF (and not just for volumes) but recently I found also some data gaps (i.e. prices data missing for few short periods of time). These gaps can be easily spotted on MT4 with this simple MQL code http://www.mql5.com/en/code/11643.
Anyway I don't want to sound over-critic: these errors are just little glitches, especially in comparison with what MetaQuotes offers by default. Having all these data and also tick data available for free, is simply a superb service.
Thank you TickStory.
Anyway I don't want to sound over-critic: these errors are just little glitches, especially in comparison with what MetaQuotes offers by default. Having all these data and also tick data available for free, is simply a superb service.
Thank you TickStory.
Re: Are Dukascopy data consistent across TF?
Hi Puglios,
Thanks for you update. Just to clarify, Tickstory utilises the Dukascopy tick data to form its own bars/candles, so we certainly would expect that the tick data corresponds to the historical bars in all time-frames. If you have a specific instance where you can see a range of tick data that does not form the correct candles, then we would look into this further - just provide us with the symbol and specific candle that is not matching. Given that the errors for MT4 'mismatch errors' are not very well documented, we would prefer to look through the raw data and see if the various corresponding bars in all time-frames match up.
Kind regards.
Thanks for you update. Just to clarify, Tickstory utilises the Dukascopy tick data to form its own bars/candles, so we certainly would expect that the tick data corresponds to the historical bars in all time-frames. If you have a specific instance where you can see a range of tick data that does not form the correct candles, then we would look into this further - just provide us with the symbol and specific candle that is not matching. Given that the errors for MT4 'mismatch errors' are not very well documented, we would prefer to look through the raw data and see if the various corresponding bars in all time-frames match up.
Kind regards.
Re: Are Dukascopy data consistent across TF?
Hi TickStory support team,tickstory wrote:Given that the errors for MT4 'mismatch errors' are not very well documented, we would prefer to look through the raw data and see if the various corresponding bars in all time-frames match up.
I've drilled down few of the mismatch errors reported and looked into the raw data myself: it seems that you are right. In all the samples data were consistent and MT4 was reporting a non-existent error. Although this sounds like a bug of the MT4 Test Generator, knowing that we can ignore the mismatch errors if we trust the source of the historical data is of some relief. Thanks TickStory for the light shed over the question.
Instead as far as it concerns the data gaps, these are quite easy to spot using the utility linked in my previous post. In the screenshot attached there is an example of a data gap on 24 Feb 2014 for EURUSD. Anyway they are quite rare indeed.
- data gap on 24 Feb 2014 for EURUSD
Re: Are Dukascopy data consistent across TF?
Hi Puglios,
Thanks for your update. We've some some searching on the MT4 forums, however it's not always clear when exactly these errors are generated. Since we are testing the limitations of MT4 by using tick data, it's not always easy to get an answer so we instead try to establish whether our system has produced any incorrect data.
Kind regards.
Thanks for your update. We've some some searching on the MT4 forums, however it's not always clear when exactly these errors are generated. Since we are testing the limitations of MT4 by using tick data, it's not always easy to get an answer so we instead try to establish whether our system has produced any incorrect data.
Kind regards.