Strategy Tester. Millions in five months?
And how is this to be understood? How acceptable (correct) are the results given by the tester (in the form of tens of millions of virtual dollars on an initial virtual deposit of 5 thousand in 5 months)?
You'll be the first in the championship!))
Thanks for the kind words :) But I wasn't even going to send these designs to the championship.
As for the subject. I'm afraid that such results are either simply misleading, or signal some kind of internal mistakes of the expert. But I can't understand these signals yet :/
Thanks for the kind words :) But I wasn't even going to send these designs to the championship.
As for the subject. I'm afraid that such results are either simply misleading, or signal some kind of internal mistakes of the expert. But I can't understand these signals yet :/
Bugs in an Expert Advisor are very easy to catch with such a profitability. Just ask yourself why these five months?
I would suggest to test the Expert Advisor on the same TF, but on a longer historical period (and in a different location).
Why not run this EA on a different instrument or say a year or two earlier?
And bugs in the EA are very easy to catch with returns like this. Just ask yourself why these five months?
Well, if the problem can be explained only by the choice of testing period, the answer is quite simple: Originally, the period was chosen at random during testing/optimization of the first variant of the Expert Advisor (in the third decade of May) and it was used as a "test" period for all subsequent variants. So that there would be a certain relative environment in which to compare the results of different variants. At first it was about 40000, then 120000, then more than a million, and now it's off the scale :) And cast some doubts.
As soon as the next run of the tester is over, I will definitely follow your advice. Thanks for the idea (if I understand it correctly) that the overestimated return allows to easily catch bugs in Expert Advisor by changing periods, timeframes and instruments.
The result of the last 1.5 years:
From the picture, you can see that the initial hundred million is the result of some activity in the spring of 2010.
Separately, I checked the previous five months (relative to the period originally selected):
This result is, of course, an order of magnitude lower than the 2010 result, but... 5.5 million in 5 months??? All the same question posed in the title of the thread.
What kind of bugs can my Expert Advisor have in such a situation?
P.S. Other periods and tools that managed to check, either also surprise, or can not calculate certain data (which fits into the general ideology of the Expert Advisor).
yamik:
Всё, просто, присылаешь мне советника и я говорю причину xD
Yeah, and the key to the flat where the money is. :)
Back on topic. I found another bug, which I still think is a 100% tester glitch.
As is easy to see, the top three lines have the same result: 8,315,991.23. But clicking on each line (running the test procedure) gives different results:
1) 1 216 967,60;
2) 6 877 260,31;
3) 8 310 991,23.
Based on these experimental data, I conclude that the results of the tester during rapid optimization are not yet trustworthy. Perhaps there are peculiarities of the genetic algorithm I'm not aware of. But from the viewpoint of a novice user the conclusion is exactly the same.
And you guarantee that there are no random events in the Expert Advisor code, which produce different signals on the same interval?
To check the signals, use the"Open chart" command in the tester report and look at the graphical display of the trades.
Do you guarantee that the Expert Advisor's code does not contain any randomness that produces different signals on the same interval?
I don't quite understand what kind of randomness we are talking about, so let me explain.
1. The Expert Advisor code has not been changed since compilation.
2. the Expert Advisor uses 9 parameters. Correspondingly, different combinations of parameters result in different signals on the same time interval and, therefore, different results. It can be seen in the figure that the third to last lines show different results.
In this case ( fast optimization) the top three rows gave the same results with different combinations of parameters. The values of all the context menu parameters were also the same. For the sake of interest, I tested each line separately three times. And each of the lines gave the same income graph and the same result three times*. But the results of the different lines differed significantly from each other, as I pointed out above.
I suppose that if each of the lines produced the same income graph and the same result three times, then the graphical representation of transactions for each of the lines would be the same (there is no way to check it now).
______
*Especially I couldn't understand for a long time why instead of the promised 8 million the first line would persistently show the same 1 million. :)
An investigation showed that the swaps in the tester were counted in error. Hence these results. Wait for the next build.
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
You agree to website policy and terms of use
And how is this to be understood? How acceptable (correct) are the results given by the tester (in the form of tens of millions of virtual dollars on an initial virtual deposit of 5 thousand in 5 months)?