Is the tester working correctly ? - page 4

 
Read my last sentence - I am not going to communicate with you.
 
nchnch:

Renat, good afternoon. Any comments? :)))

I think to make the conversation more constructive you should take Alpari MINUTES, then Erase all HST-files of interesting currency pair from the history, put there ONLY the history of downloaded minutes, run the terminal in offline mode, convert ALL OTHER timeframes from a minute one using a script. Then switch to daily and test on FULL ADEQUENT HISTORY.
If the picture you've attached repeats, then welcome back here with calculations, but this time with correct data.
But first do ALL that I wrote above in this sequence and without missing a single step.
Then I think Renat still soften and consider your situation. :) But there's a good chance that there won't even be a situation. :)
 

OK...thanks...I'll try...:))))

 
Simca:
I think that in order to move the conversation in a constructive direction you should take Alpari's MINUTES, then delete all HST-files of the currency pair of interest from the history, put there ONLY the history of the downloaded minutes, run the terminal in disconnected from the Internet, convert with a script ALL the remaining timeframes from the minute. Then switch to daily and test on FULL ADEQUENT HISTORY.
If the picture you've attached repeats, then welcome back here with calculations, but this time with correct data.
But first do ALL that I wrote above in this sequence and without missing a single step.
Then I think Renat will take it easy and look at your situation. But there's a good chance that there won't even be a situation. :)
A question for Renat. I have been doing everything this way for a long time. I am confused by one nuance: when testing on such a story the quality of simulation for 2006 remains at 90%. According to my thinking it should be 100%. And one more thing: I tried to test from 2004 - from the beginning of the minute history - the quality drops, I don't remember exactly, but about 87.5%. Again, according to my assumptions the quality should not change.
It's not about profitability or serviceability of the strategy (we can use any standard example, as long as it reaches the end of the year, or any Expert Advisor that doesn't open trades) - the reason is in the approach to price movement modeling. If the number of ticks is the same, then the quality of modeling must be maximal, at least not decreasing with increasing history.
Anyway, I have an assumption that this happens.
I can be wrong of course, but it seems to me that the reason is in the wrong approach to creation of history files. If you are interested and it's worth it I will continue the topic. Maybe you have other ideas and it should be like this ?

Good luck.
Regards, Vladislav.
 
Thanks, got it. Quality is assessed by some empirical formula. I thought it was by the coincidence of ticks in the timeseries. But one thing bothers me though - maybe you can tell me. Maybe it has no effect - your programmers can check it much faster. In general, the thing is this:
If you look at the file of the minute history, you can see that the minimum tick volume on the minutes is 1. And it appears when all four prices at a bar (open, high, low, close) are the same. From my point of view, it is not correct, because there can be two cases:
1. The close of the previous bar is equal to this price and therefore the tick movement that led to the price change to this level has already been taken into account in the tick volume of the previous bar, then the value of the tick volume of the current bar must be 0.
2. The close of the previous bar is not equal to this price - so the tick has come during the change of the bar, therefore it is not considered in the tick volume of the previous bar and must be considered in the current bar and then the value of the tick volume of the current bar must equal 1.

Then when forming bars of higher prices the operation of adding tick volumes will give more correct results.

Now the first case is clearly not considered.

Perhaps this will help improve the performance of the tester and there will be no need to evaluate the quality of the simulation using weighted coefficients?

Good luck.
Regards, Vladislav.
 
VladislavVG писал (а):

From my point of view, this is not correct because there could be two cases:
1. The close of the previous bar is equal to this price and therefore the tick movement that led to the price change to this level is already taken into account in the tick volume of the previous bar, then the value of the tick volume of the current bar must be 0.
2. The close of the previous bar is not equal to this price - so the tick has come during the change of the bar, therefore it is not considered in the tick volume of the previous bar and must be considered in the current bar and then the value of the tick volume of the current bar must equal 1.


If there is no tick, MT will not draw a bar. Everything is fine with the volume.
 
Integer:
VladislavVG:

From my point of view this is not correct as there could be two cases:
1. The close of the previous bar is equal to this price and therefore the tick movement that led to the price change to this level is already taken into account in the tick volume of the previous bar, then the tick volume of the current bar should be 0.
2. The close of the previous bar is not equal to this price - so the tick has come during the change of the bar, therefore it is not considered in the tick volume of the previous bar and must be considered in the current bar and then the value of the tick volume of the current bar must equal 1.


If there is no tick, MT will not draw a bar. Everything is fine with the volume.
I'm talking about the tester and for it the tick is there since it's worth 1.
And then what do you mean by normal ? Did you check?

Here's the problem in my opinion : with DIFFERENT inputs we get the same record for history. I agree that this may not be decisive, but again, in my opinion, the issue needs to be looked at more closely than just brushing it off - like "it's OK" ....
 

Since 1 stands, there must be a tick. MT does not fill in time holes, but misses bars. Once again, if there was no tick (i.e. price change) MT does not draw a new bar!

 
Yeah... I jumped the gun on that one.... Turns out there are bars that have an opening price equal to the closing price of the previous one. I don't know what system MT uses to draw bars. I think only the developers will be able to explain it. I would like very much to understand it.

If there has been a period of off-quota, then at the appearance of quotes the broker may give a tick with the same price. But this is only a guess.