Is it super-gral or Dont trust optmize results for the past

I'll show one example, which still gets me curious about this EA stuff.

Here

[url]http://rapidshare.com/files/352334118/optimize_EA.zip[/url] 

lies an indicator, lib and EA

password is

thanksToysRus.pdf

You can easily optmize this EA to become profitable

by modifying only these parameters:

ATRPeriod (3 to 9)

Factor (2 to 6)

Trade1VolumeFactor (0 to 1)

Trade2VolumeFactor (0 to 1)

Trade3VolumeFactor (0 to 1)

Practically any pair, any timeframe, any period

(I've found H1 and GBPJPY to be the most sucessfull).

For example, on $10K account this set of parameters

ATRPeriod=3

Factor=5

Trade1VolumeFactor=1

Trade2VolumeFactor=1

Trade3VolumeFactor=1

would make a wooping $9K just from

September till December 2009

(Don't try to trade it in real account, because

it could enter up to 3 trades per 1 signal

with "huge" (in scalping terms)

target-profits and stop-losses of around 100 pips).

Fine, is it Holly Graail?

Not really, just because set of best

performing parameters is different every time.

So, even if you know set of of best (or at least

profitable) parameters for the previous month,

you will be dissapointed next month (or even week)

with these parameters - they lead to complete lose,

while another (unknown yet to you) set of parameters

could exist - but useless, because unknown and not stable,

and not repeating.

I've spent a fortune, trying to understand the reason of the problem

and still don't know why,

but I do know much about EA optimization now.

For any past period you could optmize nearly

every EA (even $99 rubbish one) to become

super-graal. The problem is, that these

parameters won't work (usually) in a future

and nobody will tell you.

To find stable set of repeatevely profitable

parameters - this is the case, which I

cannot fulfil yet for any of EA from my big collection.

So, results of the backtest doesn't mean much.

They only make sense, if set of parameters would

continue to work again and again,

and simple optmization couldn't help much here.

Maybe another method exist, but I'm not aware about it.

Could someone give an advice?

Re: Is it super-gral or Dont trust optmize results for the past

Normally how long parameters continue to work well depends on the period of optimization. If lets say you optimized for all of 2009, the system should continue to work profitably for at least the first couple of months of 2010.

The problem isn't so much with optimization as with the fact that the markets and people's behaviors are changing over time. Looking at historical data, I've found trades that worked well, lets say in most of 2009, but the same ideas don't work as well in 2008. The average profit targets vary, or sometimes the idea almost seems to have no application.

Also, apart from some super-general trading setups on higher timeframes, there is variation in how well different trade setups will work across different pairs.

Personally, I'd look at the equity curve of the EA, whether its fairly straight, and whether its consistently profitable in the last several months.

For something like that EA, you probably don't want to change the ATRPeriod... one additional factor I suppose, is to make sure that there are a fair number of trades that form your equity curve on the backtest. If you've filtered down to it having very few trades on the backtest, then the backtest isn't going to be very useful.

Re: Is it super-gral or Dont trust optmize results for the past

You also must be careful. Sometimes an EA will show a profitable result of 200% for 3 months as an example, however if you had entered the trading on the 11th day or the 25th day or 39th day etc. it could possibly have been a total loser. Even if it is alleged to make money in the long run....would anyone stick with an EA if it hit a HUGE drawdown???

Re: Is it super-gral or Dont trust optmize results for the past

Yes, thanks for the idea to keep ATR parameter unchanged.

However, it's not much difference between ATR being 5 or 6.

With the results - interesting thing about this EA is,

that it shows profit nearly at all timeframes and during

any testing period, but - with different set of parameters

every time. And even if the curve is going up,

one cannot use it - because profitable set of parameters

is know for yesterday only, but tomorrow it won't work,

or would work accidentaly only.

Re: Is it super-gral or Dont trust optmize results for the past

noone22,

Maybe I'm missing something but I downloaded the EA and I don't see any parameters for ATR or factor or anything like that...... are you sure you uploaded the correct EA?

Re: Is it super-gral or Dont trust optmize results for the past

Maybe I'm missing something but I downloaded the EA and I don't see any parameters for ATR or factor or anything like that...... are you sure you uploaded the correct EA?

I'm 110% sure.

Just went to rapida, download my file,

and install this EA as it is.

Once you drag this EA to charts,

you'll be able to modify its input as follows

http://i39.tinypic.com/sc9jer.jpg

and in strategy tester you could do exactly the same

again - see pic

http://i42.tinypic.com/149z5w8.jpg

Interesting to know, which inputs you're able to see,

how and where are you doing this and which

EA are you actually looking for.

Name of my EA is

BAT_EA.ex4

N.B: I'm not teaching kindergarden:

where to put EA, where to put indi, where to put lib,

how to put indi or EA to the charts - site

[url]http://www.mql4.com[/url] 

will be able to help you on the matter.

In fact, my question and problem are more general

and don't depend from this particular EA

(it was given as a funny example as

"every timeframe and pair profitable EA")

I could give you tens and hundreds of others EAs

with exactly the same questions and concerns.