Download MetaTrader 5

EA Tester's result

To add comments, please log in or register
Write your articles using our editor. It is convenient!
Fa Cao
1141
Fa Cao 2013.12.24 10:07 


Above pictures are M15 and below pictures are H1:

this result is testing the EA for 5 years from 2009 to 2013.

but the EA for testing 2001 to 2007 is not good;

about 2008, it is very strange: when I test it on 2008's data, only traded 10 orders; when I test it on continuous data from 2008 to 2009 or 2010, still these 10 orders have been traded; I don;t know why? maybe the data before 2009 is wrong? yes?

In addition, shall I get some advice about this EA according to this result?

thanks.

Simon Gniadkowski
Moderator
18000
Simon Gniadkowski 2013.12.24 10:20  
vx0532:

In addition, shall I get some advice about this EA according to this result?

Is this with optimized parameters or not ?
Fa Cao
1141
Fa Cao 2013.12.24 11:55  
RaptorUK:
Is this with optimized parameters or not ?


optimized to select some parameters.
Simon Gniadkowski
Moderator
18000
Simon Gniadkowski 2013.12.24 11:58  
vx0532:

optimized to select some parameters.
Probably curve fitted . . . repeat the test without Optimization on 3 or 4 different Symbols.
Fa Cao
1141
Fa Cao 2013.12.24 13:12  
RaptorUK:
Probably curve fitted . . . repeat the test without Optimization on 3 or 4 different Symbols.


I have test it on other symbols, not too long, just several months; only one symbols' test result is nice.
Fa Cao
1141
Fa Cao 2013.12.24 23:49  
In this EA, orders are opened automately according EA's computing; about their close, two methods: one is that, for example, now this EA hold an long order, when EA gives signal which should open an short order, the long order should be closed and the short order should be opened(EA should hold one order mostly at the same time); the other one is that all opened orders should be closed by moving stop loss. when the results from the two methods are very different, how do you think? EA's signal to open orders is not good enough?
Mostly Misunderstood
21
Mostly Misunderstood 2013.12.25 20:49  
I don't trust the back testing. Run the same test 10 times and you won't get the same results 10. Too much emphasis can be placed on back testing. Even running on demo you can get different results to running on real. I have found you are better off using a small sum of money, £10, and open a real account with micro lots and run an EA this way. In the long haul it has not cost me money and I have learnt more quickly the modifications needed on the EA. Maybe the EA's I have are more suited to be tested this way. I have wasted too many hours back testing, of that I am sure.
Simon Gniadkowski
Moderator
18000
Simon Gniadkowski 2013.12.26 00:41  
properbearkhunt:
I don't trust the back testing. Run the same test 10 times and you won't get the same results 10.
Of course you will, unless you allow things to change like Spread, History data, etc. If all parameters are the same all runs will produce the same result.
Mostly Misunderstood
21
Mostly Misunderstood 2013.12.26 13:11  
RaptorUK:
Of course you will, unless you allow things to change like Spread, History data, etc. If all parameters are the same all runs will produce the same result.


I understand what you are saying about the parameters. I run a test on EUR/USD using a simple 3 EMA crossover. I get 2 different results, with a notable discrepancy between the two. I go further and do the same with another broker to find discrepancies. You have in the past recommended dukascopy as a reliable source of data. Perhaps that may produce consistent results.
Simon Gniadkowski
Moderator
18000
Simon Gniadkowski 2013.12.26 13:34  
properbearkhunt:

I understand what you are saying about the parameters. I run a test on EUR/USD using a simple 3 EMA crossover. I get 2 different results, with a notable discrepancy between the two. I go further and do the same with another broker to find discrepancies. You have in the past recommended dukascopy as a reliable source of data. Perhaps that may produce consistent results.
If you test with different data you will get different results . . . isn't that blindingly obvious ? if the spread changes between 2 run you will get different results . . . if you want the same result keep ALL the data the same . . . it is that simple.
Mostly Misunderstood
21
Mostly Misunderstood 2013.12.26 14:18  
RaptorUK:
If you test with different data you will get different results . . . isn't that blindingly obvious ? if the spread changes between 2 run you will get different results . . . if you want the same result keep ALL the data the same . . . it is that simple.


Yes, I do understand that broker A will have different results to broker B, as their prices, spreads will always be different. But, as I tried to explain, results are not consistent for each broker. I was merely trying to highlight that some brokers data for back testing is not suitable to deliver accurate testing results. As a point of order... where did I say "If you test with different data you will get different results". Please don't be so quick to jump on me in your usual over zealous fashion. You possess no awareness of how you present yourself to others. Yes, you are an expert, but that is not an excuse. Google Johari's window.

1234
To add comments, please log in or register