Completely different results from same expert - page 2

New comment
 

If you won't perform your tests in a consistent way why do you expect to get consistent results ?

I'm not knowledgeable about Oanda . . but Alpari Demo accounts have their best spreads, their smallest deposit live account has spreads which are much, much worse.

Reconnecting to your Broker will more than likely overwrite some of your data . . . which may well lead to data mismatches.

 

Your experience with Alpari may be misleading you: Oanda's live and demo clients appear to be very similar in most ways. Certainly there is no systematic difference in the prices or spreads being offered (it is possible to find occasional 0.1 pip differences between the demo and live charts of bids and asks, but I am assured by Oanda that this is because the charts don't always include every tick). But more crucially to the topic of this thread, the large majority of the trades occur months earlier in the period covered by downloaded data. If you glance at the graphs of 7 months of backtesting in the first post in this discussion, I think you will perceive a difference from the very start.

So the difference is still in need of an explanation - I now believe something is not working right, and this has the potential to be highly misleading to users.

 
My experience with Alpari is not misleading me, it was an example of what can happen sometimes with Brokers, i.e. their Demo and Live platforms can be very different . . .
 
What about commision calculated in demo and live environment?
 

@RaptorUK, I am sure you are right in general, but my main point is that in the backtest over 7 months or so with the history data having been constructed from the same 1 min data in both cases, the majority of the data being used by the backtests should surely not have come from the broker at all. Hence most of the difference is still in need of an explanation. The server connected to may well influence the results, but the scale of the effect is bizarre and the mechanism unknown. The reason I am not happy to leave it here is that it is certain now that metatrader backtests can be spectacularly misleading sometimes and I want to find out how and when this happens. What sort of effect can increase the profitability of trades by several tens of pips each over a run of around 200 trades? It would certainly not be justified at this stage to come to the conclusion there is no problem when connected to a live server.

Following zzueg's input, I intend to try creating a new installation of metatrader, disconnected from the server, deleting all history data, reconstructing the data from scratch from downloaded data and doing backtest then, which should take the server out of the picture entirely. The only practical problem with this approach for regular development is that the current month will not be used in the backtesting.

 
@zzuegg, there is no commission - just a spread which is meant to be identical in both environments (I understand it's two forks of the same data feed - the only difference is in the execution, and usually no more than a fraction of a pip there in my experience).
 
zzuegg:
What about commision calculated in demo and live environment?
I don't think Oanada charge commission . . .
 
Elroch:

Following zzueg's input, I intend to try creating a new installation of metatrader, disconnected from the server, deleting all history data, reconstructing the data from scratch from downloaded data and doing backtest then, which should take the server out of the picture entirely. The only practical problem with this approach for regular development is that the current month will not be used in the backtesting.

Which is what I suggested initially . . . except for the new installation of MT4 . . . not sure that is needed.
 

Take note of this:

  • Lot sizes including initial size and increment step, commissions and swaps should be taken from the active account settings
Before testing, it is necessary to make sure that there is at least one activated account in the list in "Navigator" window of the terminal.

from here: https://www.mql5.com/en/articles/1512

 
I'll make sure of that, RaptorUK.
123
New comment