Backtesting opinion - page 3

 

RaptorUK:
How do you know that your curve fitted for the last 8 months is going to "fit" the next month, 2 months, 6 months ?  or do you repeat this daily and have a moving daily fitted curve ?  how do you know it will fit for the next day ?  

 

Well, obviously you don't know. But, on the other hand, what chances are there that something that worked for 8 months will stop working immediately after you go live?   

 
altr:

Well, obviously you don't know. But, on the other hand, what chances are there that something that worked for 8 months will stop working immediately after you go live?   
How did you test it ?  how many pairs ?  how consistent was it ?  did it have a logical basis for working ?  or was it just a few Technical Indicators thrown together because Fred said that it will work ?
 
Well, I did that for my expert that participate the atc2012 and it survive without intervention for the next 3 months making a profit. Im sort of have some confidence on this routine of back testing. 
 
RaptorUK:
How did you test it ?  how many pairs ?  how consistent was it ?  did it have a logical basis for working ?  or was it just a few Technical Indicators thrown together because Fred said that it will work ?

Maybe it's me, but thus far I have not been able to establish any good relation among these. When testing out of sample there are average strategies that perform very well, but there are also good ones that fail. I also look at 8 month - 1 year history and the worst case scenario would be break-even. Markets change but they don't change overnight (except if there's major policy change by some central bank). 

BTW you also have to look at number of periods/issues. 10 yrs on daily chart has equal number of periods as 1 yr on lower time frame 

 
doshur:
Well, I did that for my expert that participate the atc2012 and it survive without intervention for the next 3 months making a profit. Im sort of have some confidence on this routine of back testing. 

Did you switch brokers?

Some broker allow trading before others thus the math begins to sway causing all indicators to lose what was at the time of testing true value.

I begun to only use daily data, only trading smaller time frames beginning on Tuesday sometimes Monday depending on multiple brokers numbers.. The results better! Also because of it i have had to go back manually to make sure the results are the same for different brokers on smaller time frames.

Sunday is the worst by Tuesday the numbers begin to synch back up some what.

Documentation on MQL5: Standard Constants, Enumerations and Structures / Environment State / Symbol Properties
Documentation on MQL5: Standard Constants, Enumerations and Structures / Environment State / Symbol Properties
  • www.mql5.com
Standard Constants, Enumerations and Structures / Environment State / Symbol Properties - Documentation on MQL5
 
q.import:

Did you switch brokers?

Some broker allow trading before others thus the math begins to sway causing all indicators to lose what was at the time of testing true value.

I begun to only use daily data, only trading smaller time frames beginning on Tuesday sometimes Monday depending on multiple brokers numbers.. The results better! Also because of it i have had to go back manually to make sure the results are the same for different brokers on smaller time frames.

Sunday is the worst by Tuesday the numbers begin to synch back up some what.

You mean the values of those indicators will begin to sync around Tuesday because different broker have different opening hours? 
 
doshur:

In your opinion, if a EA is profitable in the 10 years backtest,

1. would it be profitable when it goes live? If not why? What are the reasons?

2. the EA is not optimize to the recent market changes and if it is profitable with 10 years data, isn't it very robust? 

doshur, in my opinion the answer depends too much in the EA algorithms, since the best backtesting is just analyzing history. 

A breakthrough of MT5 was forward testing, instead of just in the sample backtest, and even with these tools we are just talking about past. But I consider it a great tool to filter overfitting setups. Even better if you can reply this in the EA algorithms.

So my answer is yes, if you have smart and very adaptive algorithms with efficient overfitting filters, a good 10 years backtesting can be profitable when it goes live.

But the paradox is that I do not believe that technology that exists today can create and put all these algorithms in just one EA, even with cloud testing.

Distributed Computing in the MQL5 Cloud Network
Distributed Computing in the MQL5 Cloud Network
  • cloud.mql5.com
Connect to the MQL5 Cloud Network (Cloud Computing) and earn extra income around the clock — there is much work for you computer!
 
figurelli:

doshur, in my opinion the answer depends too much in the EA algorithms, since the best backtesting is just analyzing history. 

A breakthrough of MT5 was forward testing, instead of just in the sample backtest, and even with these tools we are just talking about past. But I consider it a great tool to  filter overfitting setups. Even better if you can reply this in the EA algorithms.

So my answer is yes, if you have smart and very adaptive algorithms with efficient overfitting filters, a good 10 years backtesting can be profitable when it goes live.

But the paradox is that I do not believe that technology that exists today can create and put all these algorithms in just one EA, even with cloud testing.

My EA actually is a one pattern finder. I think I just need to "curve fit" to suit the current market changes.

So I do agree on the statement > in my opinion the answer depends too much in the EA algorithms

 
doshur:

My EA actually is a one pattern finder. I think I just need to "curve fit" to suit the current market changes.

So I do agree on the statement > in my opinion the answer depends too much in the EA algorithms

And how do you predict when the current market condition will change ?
 
RaptorUK:
And how do you predict when the current market condition will change ?
Curve fit current period. Predict future change is not drastic but slowly change