Optimization has a Drawdown of 23% but "run single test" returns 87% DD

[Deleted]  

Hi everyone,

after having optimized my EA, the table of the results shows 23% as max drawdown.

However, when I run the optimization variant with right click "Run single test", the drawdown is computed to be 87%.


What could be the reason for this issue? Is there a way to fix this from my side?


Thank you very much!

Files:
Eleni Anna Branou  

What modelling have you chosen for your optimization?

Only every tick modelling will give you trustworthy results, but it will be very slow to complete.

[Deleted]  

Thank you. In both cases I am using OHCL.

Eleni Anna Branou  
hanneswaser #:

Thank you. In both cases I am using OHCL.

That's the reason then.

[Deleted]  

Uhm,

what could be the reason for a backtest using the same candles to compute the profit differently?

Fernando Carreiro  
hanneswaser #: Uhm, what could be the reason for a backtest using the same candles to compute the profit differently?

OHLC testing only works properly if your strategy opens AND closes at the opening at the bar and at no other place what-so-ever. That means, no S/L and no T/P. Only managed open and close at bar open.

M1 testing is better but still only works properly if your strategy opens AND closes at the opening at the M1 bar and at no other place what-so-ever. Again, that means, no S/L and no T/P. Only managed open and close at M1 bar open.

If that is not the case, then OHLC or M1 testing is only for a quick very rough estimate of results.
Fabio Cavalloni  
Fernando Carreiro #:

OHLC testing only works properly if your strategy opens AND closes at the opening at the bar and at no other place what-so-ever. That means, no S/L and no T/P. Only managed open and close at bar open.

M1 testing is better but still only works properly if your strategy opens AND closes at the opening at the M1 bar and at no other place what-so-ever. Again, that means, no S/L and no T/P. Only managed open and close at M1 bar open.

If that is not the case, then OHLC or M1 testing is only for a quick very rough estimate of results.
Not completely right...

SL and TP are managed correctly and in the same way by the tester in all modes, not only every tick.

Of course if a function for modifying SL or TP (like a trailing stop) are used and based on Ask/Bid prices or current bar information, they work correctly only in every tick mode.
[Deleted]  

Thank you very much for your comments.


I have a feeling that we are moving into the direction where we compare backtesting with OHCL and backtesting with tick data.

However, I use OHCL in BOTH backtests. The settings are exactly the same. The run within the optimizer however, returns quite different results.

From a systematic perspective I could think here of different reasons:


- The optimizer skips candles based on some undisclosed criteria

- the optimizer uses some random prices between the candles open and close

- the computation of the profit and DD sum is somehow vectorized over threads  and not reduced correctly for the total sum.


Or did I get something totally wrong?

[Deleted]  
Enrique Dangeroux #:
Do you use a beta build?

no. its 3320

Reason: