Optimisation results differ from single tests on them - page 2
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Continuation of the quote glitch epic.
Downloaded again the latest build 228 uploaded by Alpari. Installed it in a separate folder. I have not opened online charts. I have downloaded a history for USDCHF in "Quotes Archive" with a number of bars in the history by default - I do not need much, even the last 2-3 months will be enough. I pressed the "Download" button two times. The first time it was loading something from servers (I don't know which servers). After the second click it suggested recalculating all the timeframes - I agreed. After the recalculation I enabled the left proxy in the settings (so MT4 could not find the Internet), logged out and re-entered the program. There was no connection to servers anymore. I chose the required period, parameters and ran a single test. I suspiciously received a test chart showing a small amount of deals. I looked through the reports and logs: only one week out of the two weeks of the testing period I specified has been processed. It turned out that there is a BIG HOLE of two weeks in the quotes for this pair, which "ate" one week of my period.
The attached screenshot shows that:
- MT4 is offline, no connection to servers;
- testing period from 2010.10.25 to 2010.11.23 was chosen (I entered the end of period 2010.11.23 with some reserve, it was more convenient for me);
- testing was really carried out from 2010.11.01 00:00 till 2010.11.05 22:00, i.e. a whole week was missed at the beginning;
- in the "Quote Archive" there is a gap in hourly quotes between 2010.10.15 and 2010.11.01 - more than two weeks of quotes are missing;
- StrategyTester Report" writes that the modelling quality is 90% (maximum possible) and there are no discrepancies - everything is fine;
The only thing from where we can understand that there is a gap in StrategyTester Report is a discrepancy between the dates of the beginning of the testing period that I specified and the date of the actual testing period. But if the gap is within a period being tested, the periods will coincide and the user will have an illusion that the testing/optimization was performed correctly. And then they will lose money due to incorrectly selected strategy parameters.
There is the same hole in minute and other quotes in the "Archive". Although quotes were downloaded correctly and no errors were written. In the next folder on the same computer there is another copy of MT4. The quotes for this pair are present there for the whole October without errors, but they were downloaded several days ago. I have plenty of free space on my disk. I have a wide enough Internet channel, 4 megabits, stable and almost free. The connection was not disconnected at that moment for sure. I have two computers running ICQ, internet radio and a couple of other MT4s online via the same internet connection and nothing was interrupted.
There is a gross error of working with quote archives in MetaTrader4. Has no one encountered it?
I wonder why MT developers are silent? If there is no answer, how can I contact them other than through this forum? Maybe there is a bug tracker or direct access to support?
Immediately after writing the previous post, tried a few more times to load the quotes in the "Archive". Nothing was downloaded. Neither after several clicks on Load, nor after several closing/opening of MT4. The hole in quotes has remained.
I manually cleaned the folders \history\Alpari-Demo and \history\downloads. Now all the quotes that were not in the "hole" are loaded the first time without any problems. So it doesn't seem to be about Alpari's servers.
Use the tester as a means to find errors in your algorithm, the correctness of the Expert Advisor, but not as a tool for optimization. For this purpose the "Visual Tester" of Hypurga is quite good (it is an indicator)
What kind of tester is this and where to get it? Yandex and Google do not know about it.
I wonder why the MT developers are silent. Who knows how else to contact them other than through this forum? Is there a bug tracker or direct access to the helpdesk?
So there really is a problem. Personally, congratulations ;-)
Thanks, sever30, but I need to optimise almost half a dozen pairs quickly over the weekend. It will be a very long process on the equity tester. Although MT4's tester is slow, it would be easier and faster to test on it.
Obviously, I am not the only one who has errors. The test results are different because of glitches in "Quote Archive". There's a very long discussion on https://www.mql5.com/ru/forum/102259, examples and tips. But there are no normal answers from developers there either.
The only option for checking the history for integrity, which I have found so far, is the "History data analysis for holes and gaps " script https://www.mql5.com/ru/code/7093, which is a development of the "history data analysis" script from Bagadul https://www.mql5.com/ru/code/8039. It at least to some extent allows you to have confidence in the integrity of the history.
But in my opinion, this is a huge flaw in MT4 (and seemingly in MT5 too). In three years of existence "Archive" and not bringing it in order and leaving such glitches in it is a complete irresponsibility on the part of the developers. :-(
So how did the topic end? Time goes on and the story is the same: the results of optimisation runs and simple tests are different... sometimes so different that it's a crying shame. At the same time if you run a single test once, twice, three times the result is the same. But if you mix in the results of optimization, the result is different... This is very silly.
1) Is the spread fixed? - yes
2) Is the quote archive good quality, without holes? - I checked it manually, no gaps
3) Did you check the algorithm of the Expert Advisor? - Yes, of course I checked it. On a single test, the result is the same, no matter how many times you run it.
4) With other brokers the same story repeats? - it is the same, not the brokers!
5) Did you choose a smaller period or a bigger one? - yes
6) Did you try to test the explicit control of bars? - well, i tried... just not for my EA
Well, if you tried everything, then why don't you just shoot?
The results of optimisation runs and simple tests are different... Sometimes they are so different that it makes you cry.
in your case, the subsequent run is essentially a forward run, if it doesn't leak, that's a good thing.
Try to swap actions, run, then optimise.