curve fitting delema

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
danZman
Roundtable Knight
Roundtable Knight
Posts: 117
Joined: Thu Mar 10, 2005 10:07 am
Location: Newport Beach,CA

Post by danZman »

I've come into quite a curve fitting delema myself. I backtested an intraday Forex system for 9 years (with 5 minute bars). The CAGR was 741%, MAR 21.97 over 9700 trades. It's a very simple system that buys the dip...only 4 rules...two of which are moving averages (short MA is greater than Long MA, and close is > Long MA). The other two rules are just as simple.

In forward testing for one year, CAGR dropped to 65%, MAR 2.04. That's a HUGE difference and I considered throwing it away. However, I was thinking of using an *adaptive* algorithm to contantly optimize the parameters.

So say every Friday (In the After Instrument Day), I could loop over a few months of data and see which parameters worked the best. My *theory* is that markets change gradually, and what is working now will continue to work into the near future. By adapting to changing market conditions, I can make a better, robust trading system. A big benefit would be that most of the testing would be walk-forward in nature.

It's been my general belief that I should keep my portfolio the same as Richard Dennis did...I should not optimize each intrument seperately. But is there any reason why I should do this when there are so many trades for each instrument? I mean...the reason why we don't optimize for each instrument with Turtle was because there's simply not enough trades for statisical significance right?

For one thing, if I optimize for each instrument separtely, I could phase out an instrument that is not contributing to the portfolio (this would be automatic, because the best combination of the parameters might be generating ZERO trades). In fact, the whole system could phase itself out until some combination of profitable parameters is found.

Any thoughts?

I will be testing this idea ASAP, but would like to hear some input.

D
Attachments
forexintraday.gif
forexintraday.gif (19.6 KiB) Viewed 3191 times
sluggo
Roundtable Knight
Roundtable Knight
Posts: 2987
Joined: Fri Jun 11, 2004 2:50 pm

Post by sluggo »

Dan, get LeapFrog to review your system. He's the resident expert on short term systems, and yours seems to be a very short term system, holding trades less than two bars. He can suggest ways to test whether you've accidentally underestimated slippage or the effects of within-bar gaps.
danZman
Roundtable Knight
Roundtable Knight
Posts: 117
Joined: Thu Mar 10, 2005 10:07 am
Location: Newport Beach,CA

Post by danZman »

sluggo wrote:Dan, get LeapFrog to review your system. He's the resident expert on short term systems, and yours seems to be a very short term system, holding trades less than two bars. He can suggest ways to test whether you've accidentally underestimated slippage or the effects of within-bar gaps.
I think there's a problem with the trade duration block. It actually holds for 45-60 bars on average. I noticed a problem with this block when it comes to intraday data. I included 50% slippage in the testing.
Post Reply