Question about EA Optimization process

Back to topics list To post a new topic, please log in or register
avatar
19
eddieTrader 2015.09.22 14:44
 

Hi,

 

Quite recently I finished a "previous" version of an EA.

 I say "previous" because my idea was to use the Optimization process of MT4 to optimize the parameters.

However, now, thinking about it I am wondering if this will be usefull or a waste of time.

Because for Optimization I would use past data available on MT4, of course,  but......my concern is that....past data is past data...I mean...An optimized EA which performs quite OK for past data will not have to necessarily perform OK for the data to come...Am I right??

Since past data is not linked or correlated to future data...Why to use optimization?

Would appreciate any coments on that.

Thanks. 

avatar
1316
jjc 2015.09.22 15:30 #
 
eddieTrader:

Since past data is not linked or correlated to future data...Why to use optimization?

Would appreciate any coments on that.

I'll give you an answer which you won't often see on this forum, ultimately deriving from the only really successful algo trader who I know (> $100,000,000 FUM). The "you shoulds" in the following are statements of his very strongly held opinions...

Optimization tells you the best-performing values of parameters in the past - on the assumption that those values will continue to hold good in future - but its purpose is both narrower and broader than that.

Firstly, you should be able to estimate what the optimal value of a parameter is likely to be (in the same sort of way that any maths or science teacher will encourage you to pre-estimate the results of a calculation), and you should become very suspicious if the actual result of the optimization turns out to be significantly different. If you are unable to estimate what the optimal value might be, then you don't actually have a "strategy"; you're just randomly curve-fitting.

Secondly, you should see one peak, occasionally two, in the results of optimizing each variable, with a gradual, largely linear decline either side of the peak. If you see anything else - in particular, wild alternation of peaks and troughs - then your strategy is just random junk, and you should bin it and start again.

Thirdly, you should optimize on no more than three different variables.

The fundamental rule is that historic data is full of coincidences - in much the same way that nutjobs can find patterns in the layout of buildings in Manhattan - and indiscriminate mass use of optimization will simply pick out these chance patterns from the random noise of historic data.

Back to topics list  

To add comments, please log in or register