why is curve fitting so bad?

Discussions about the testing and simulation of mechanical trading systems using historical data and other methods. Trading Blox Customers should post Trading Blox specific questions in the Customer Support forum.
Post Reply
TradingCoach
Roundtable Knight
Roundtable Knight
Posts: 176
Joined: Thu Apr 17, 2003 9:52 am
Location: Sacramento, CA
Contact:

why is curve fitting so bad?

Post by TradingCoach » Thu Jul 15, 2004 9:08 pm

I am trying to understand this - I do not mean to curve fit a system for a specific commodity or an index for example but having a system curve fitted for say,l commodities is acceptable in my opinion. What do you think?

verec
Roundtable Knight
Roundtable Knight
Posts: 162
Joined: Mon Jun 09, 2003 7:04 pm
Location: London, UK
Contact:

Post by verec » Thu Jul 15, 2004 9:52 pm

You may want to read what c.f. had to say on this.
But to make the discussion more productive, you might want to expand a bit on what you mean by
having a system curve fitted for say,l commodities is acceptable in my opinion
What kind of curve fitting? Parameter optimization? Market selection? And as a way to achieve what, better prediction ?

TradingCoach
Roundtable Knight
Roundtable Knight
Posts: 176
Joined: Thu Apr 17, 2003 9:52 am
Location: Sacramento, CA
Contact:

Hmm.

Post by TradingCoach » Sun Jul 18, 2004 2:35 pm

Good link. Thanks
I meant primarily parameter optimization and testing on a well selected group (basket) that one would normally trade from. I suppose it is ok to exlude Palladium if you do not ever want to trade that market, but trade gold and silver....

TC
Roundtable Knight
Roundtable Knight
Posts: 101
Joined: Thu Jul 08, 2004 2:12 pm
Location: Toronto, Canada

Post by TC » Sun Jul 18, 2004 11:50 pm

Curve fitting gets a generally bad press because it is so often associated with overfitting system parameters to past data to the extent that the system breaks down on out-of-sample data.

I do not have a problem conceptually with the idea of curve fitting, it seems reasonable to explore a range of parameter values to find a region that seems to be yield acceptable results.

However, I will never trade a system that has been optimised on a single market, even if that is the only market I plan to trade. If my system is not broadly profitable on a basket of futures markets or a random portfolio of 30+ equities I do not want to risk money on it.

Given the huge advantage equities have over futures (8000 stocks cf 80 futures markets) I have the luxury of leaving all of the equities I plan to trade out of the testing phase and keep them exclusively for out-of-sample testing.

For futures I optimise my parameters on a portfolio of every reasonably-liquid continuous contract I can find. Currently in TradeStation this is about 40 but hopefully soon, with data importing, I will increase this to 80.

blueberrycake
Roundtable Knight
Roundtable Knight
Posts: 125
Joined: Mon Apr 21, 2003 11:04 pm
Location: California

Post by blueberrycake » Mon Jul 19, 2004 1:23 am

TC wrote: Given the huge advantage equities have over futures (8000 stocks cf 80 futures markets)
While I do not want to start on a discussion of whether futures or equities are "better". I should point out that the 8000 stocks figure is extremely misleading. Not only are most of the 8000 highly illiquid small cap stocks, but the vast majority of them are also well correlated to one of the primary indexes. The number of liquid stocks that have very little correlation to each other is actually relatively small.

-bbc

Post Reply