Can you "Curve-Fit" six years of data?

Curve-fitting is a nasty world in the EA world....

But what's an EA programmer supposed to do? All we can do is test our strategies against historical data, and hope the "personalities" of the pairs don't change much.

So what makes a good EA? Firebird by Wackena is great in sideways, and MA systems do well in trends.

What does an EA need to stay robust in both markets?

My philosophy is to break even most of the time, and take small profits other times. Basically you add strategies that are profitable in both sideways and trending markets, and a lot of cancellation occurs. If we're breaking even, we aren't drawing down.

So I have an EA that made 11,000 pips in 6 years. That's a lot of data to curve-fit, but anything is possible in forex.

I guess all we can say is "Good Luck," you know?

quick reply, on blackberry.

optimize against 5yrs of data. then check performance on remaining yr.

I agree, optimization and backtesting should only be used long term to make sure that the idea/strategy behind the execution is solid.

It seems that curve-fitting has to do with exploiting irregularites in the data, which do not fit with the overall (smoothed) performance curve, so for me it is somewhat of a misnomer.
For instance, if you represent the profits as a curve it self, if you check around the optimized parameter you should see a (smoothed) curve peaking at the optimized point, and points around it corresponding to less and less profit, i.e. more or less an inverse U curve.

Btw, there is an excellent article on backtesting here somewhere on the forum.

yes you can curve fit 6 years of data, EASILY!

your best defense against curvefitting is to have a sytem for developing your systems.

Yeah, thx merlin, that was the article I was talking about.

http://www.meyersanalytics.com/publications/tricked.pdf, interesting stuff.

Suppose your trades are based on the TA/price action of D1.... you will have 24 times less data when compared to a system that trades H1 bars.