Optimization what it is.

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
ratio
Roundtable Knight
Roundtable Knight
Posts: 338
Joined: Sun Jan 15, 2006 11:07 pm
Location: Montreal, Canada

Optimization what it is.

Post by ratio » Sun Jan 21, 2007 8:25 pm

I am building a system, it is the same signal for all instrument that will be traded. But I would like to introduce parameter on a per instrument, ex: Average Length or Channel Length or size of bets on per instrument basis.

I have been able to do that fairly easily with Tblox by reading the parameters from a File for each instrument. However, by doing this am I going into curve fitting? The fact that I use a simple set of signal with 3 to 4 parameters for all system make me believe that I am not.

I'm asking this question because, in other aspect of life dont we adjust our instrument base on what we are going to do.

Here is my thinking, I own 3 motorcycles. (here they are)

they are basically all the same, 2 wheel, a motor, a frame, lights, etc.

They area my trading system with the same trading model, They can all go on the same road, asphalt, dirt, off road. And most of the time, as in the market we do not know what is ahead (how is going to be the road).

However they are optimize for specific activity, the small Suzuki if more for playing with the kids, even do this thing can run up to 50 Miles an hour. The honda, for off road and dirt road and the BWM R1200GS for around the world tour.

The bike are all going to behave really well in any environment, the "trading model" is base on sound logics. BUt they where optimize for specific condition. And if I want I can even optimize their parameter as I am going. (Higher altitude, etc.)

We can find numerous example of thing that we optimize based on our past experience.

So the question is. Is it overfitting to optimize parameters of a simple trading system, to every market or market segment.

What are your experience, where should we stop.

Are the fund use specific parameter to the market they trade, or they trade the same system with the same parameter for all market.

Denis
Attachments
motor.PNG
motor.PNG (726.97 KiB) Viewed 5679 times

RedRock
Roundtable Knight
Roundtable Knight
Posts: 939
Joined: Fri Jan 30, 2004 3:54 pm
Location: Chicago

Post by RedRock » Sun Jan 21, 2007 9:59 pm

Your suzuki will always run like a suzuki. Corn, will not always trade like "corn".

BARLI
Roundtable Knight
Roundtable Knight
Posts: 650
Joined: Sat Jan 17, 2004 6:01 pm
Location: USA

Post by BARLI » Mon Jan 22, 2007 3:56 pm

its also been one of the questions I've kept asking myself, I've noticed that some markets tend to behave in some certain way, for example take Japanese Yen, or Euro Dollar. these are trending markets. Why? Because theres too much money being poored into it and its impossible to manipulate them. Cotton from the other side is "chaotic" market where trend follower gets whipsawed and so on

ratio
Roundtable Knight
Roundtable Knight
Posts: 338
Joined: Sun Jan 15, 2006 11:07 pm
Location: Montreal, Canada

Post by ratio » Mon Jan 22, 2007 6:01 pm

Maybe Corn will not trade always like Corn, but it seem to me that market do not trend all in the same manner.

Some will tend to have shorter Burst, like the Grain and soft. However when I look at Currency and Interest rate they tend to have much longuer and smoother trend.

And Back testing also make me believe that. BEcause I will get much different parameter from market to market. ex: Channel Breakout lenght of 50 rather than 15 days.

So this is the point of my question.

If I used the exact same logic in my system for all the market, but adjust the parameter for each individually, I am surely curve fitting the system to this market.

But should'nt I trade with what I know work best in the past on a per market. This is obviously given that I have a well defined and intelligent logic in the system.


Anybody is doing that or nobody is left to answer the questions because they all want bancrupt trying to do just that.

Denis

RedRock
Roundtable Knight
Roundtable Knight
Posts: 939
Joined: Fri Jan 30, 2004 3:54 pm
Location: Chicago

Post by RedRock » Mon Jan 22, 2007 9:45 pm

The only way to know the answer is to do the tests. Log in a spreadsheet periods of time with optimized parameters. Then test out of sample on the portfolio and compare results vs an initial portfolio wide optimization set. Walk that forward. And back... You should be able to quantify which method yields your most cozy set. Which method actually yields out of sample results most like your development period. Ive seen results that suggest the more diverse your data in development, the more robust going fwd. vs finely tuned to each market sets. You may come to another conclusion.

Corn was a grain until ethanol / E85. Now its an alternative fuel source. Will it trade the same as always? perhaps. Perhaps not.

The Sp before and after apx 1996. Are they the same? Does the best data set pre 96 work best post 96?

tony5147
Contributing Member
Contributing Member
Posts: 8
Joined: Sun Jan 21, 2007 11:17 am

Hi

Post by tony5147 » Tue Jan 23, 2007 4:00 am

Hi Traders :

I've been trading stock options for about 2 years. I truly
understand that using mechanical system parameters to track
the trend of the stock could be rather risky since certain stocks
are volatile.

That's why I use options as a leverage tool to take advantage
of such sudden price movement (eg. gapping) in stock for
short-term trades.

painless
Roundtable Fellow
Roundtable Fellow
Posts: 59
Joined: Tue Nov 14, 2006 12:52 pm

Post by painless » Wed Jan 24, 2007 10:41 am

I think you are on the right track. Most of my testing has been on indices, currencies and bonds. I have found the markets in each of these groups behave similarly but across groups is something else. I use two parameters and I do not expect to have one magic set of numbers for all markets.
By definition if your system has even one optimised parameter it is curve fitted to that parameter, isn't it? The deviation from optimum is or how much performance degrades as you move away from optimum, has to be considered. As does the number of parameters. Curve fitting needs to be seen in this context.
I have spent a lot of my time getting rid of parameters and now I have two. From here I am now adding a stop and re-enty then finally comes position sizing/accumulating.
I think it is important to choose something that suits your trading psyche. I have little time for trading and I work at a job so I am happy to get 3 trades a year. That would not suit many others. Many traders use stops based on their trading capital. I can not do that since it does not match with my system. I approached my development from the tool end first. This means that if I take away my stop there is hardly any change to my result. I would not be able to sleep at nght if my stop was a necessity. I think there is some merit in looking at your testing in this light or at least quantifying the value of each parameter, stop, rule etc. This information gives you a deeper insight as to what is happening when you optimise and can allow you to remove superfluous parameters.
Just my thoughts.
Blair

ratio
Roundtable Knight
Roundtable Knight
Posts: 338
Joined: Sun Jan 15, 2006 11:07 pm
Location: Montreal, Canada

Post by ratio » Wed Jan 24, 2007 12:49 pm

Funny how thing happen.

On testing my system, I decided to change the portfolio I was using. So rather than modifying the parameter, I modify the instrument and kept the same set of parameter.

So I took, the Trending Portfolio in TB and test against.

From 1982 to 2007 I got a 45% CAGR with a MAR of 1.0 and a Sharpe of 1.05

So i guess in this case is not fitting the parameter, but rather fitting the instrument to the system....

Denis

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

Post by danZman » Wed Jan 24, 2007 4:52 pm


Post Reply