Page 1 of 1
Have the Optimal Parameters Changed?
Posted: Sun Apr 20, 2003 4:12 pm
MODERATOR'S NOTE: Topic moved from Original Turtle System forum.
How have the parameters changed over the years and how might one adjust them? I use walk forward optimization in a day-trading S&P system. It trades enough, about 150-175 times per year, so that optimizing each month on the previous year keeps it tuned to market. This is fairly simple with a single market system but how would one adjust the parameters of a multi-market system and not have the results distorted by one or two markets or one volatile event that shakes all markets? Should the parameters be adjusted with or with out position sizing? A big win in one market early on may distort the parameters for all other markets. Have all markets experienced the same type of trading changes over the last twenty years?
Posted: Wed Apr 23, 2003 5:50 pm
Yes, the optimal parameters have changed over the years in some fairly significant ways.
This is fairly simple with a single market system but how would one adjust the parameters of a multi-market system and not have the results distorted by one or two markets or one volatile event that shakes all markets?
I tend to optimize using all the data I have; all the markets, and all the years. This means that as time goes by the new years add data and this might change what the "optimal" values are.
I don't think it is a good idea to change in response to the recent past except that the recent past is also part of the larger "entire past". So the recent past gets a vote, but only as part of the overall history.
The results of an individual market are seen only in the context of the other markets so they won't affect the results any more than the other markets.
If you test self-optimizing systems make sure that you are measuring the changing algorithm itself as part of your testing.
Should the parameters be adjusted with or with out position sizing?
Well, I wouldn't even consider testing any part of the system in isolation when trying to determine optimal values. You have to test whole systems, not individual parts. This is especially important when considering money management. If you aren't testing parameters in the context of the whole system (money managment, exit stops, position sizing, etc.) how do you know they are the best in that context?
If you are looking for risk-adjusted returns you have to have all the risk mitigating portions of your system in place to get an accurate assessment.
For example: A certain value for X with an X-day breakout entry might have better profits but result in larger drawdowns and lower risk-adjusted returns when combined with a certain money management method, how do you know this if you don't have the money managment connected?
Have all markets experienced the same type of trading changes over the last twenty years?
Since I never look at single markets, I can't tell you whether they have changed.
Posted: Fri May 09, 2003 7:52 am
Interesting subject. How do others change their system's or system values or parameters over time?
How do you know when you should make changes?
Posted: Sun May 11, 2003 3:30 pm
Any thoughts on using different parameters for different markets? I know this screams of curve fitting but it may be more profitable to acknowledge that for various reasons each market has its own personality, like people, and its characteristics or habits become predictable and reliable over time. When one has a valid system or premise with which to trade, like the Turtle system, can we then safely tweak the parameters for each market? After twenty years we see that some markets are trendy and some are not. â€œOf course all animals are equal; some are just more equal than others.â€
Posted: Sun May 11, 2003 6:46 pm
Someone else said that trends can be classified as good bad or ugly. Good trends were those that broke out fairly cleanly and then retraced very little. Ugly trends have lots of 60-110% retracements in them. Bad is somewhere in the middle.
A good trender like JY will make money on most trendfollowing systems and an ugly trender like SP will loose money on all trendfollowing systems. BP was good once but has been ugly for quite a while now (at least as a crossrate with USD). So increasing the length for a breakout or the distance from the center line for aberration or trendchannel wont help them. To trade uglys you need to use a strategy involving buying retracements and you probably also need to exit on targets. This misses the big moves but makes money on the choppy periods where a trendfollower just throws money to his broker
A good diversification might be to trade this style as well as TF systems. Certainly a lot of contracts spend most of their time in the semi-trending, bad or ugly, modes. The only problem I perceive is that the systems I have looked at that try to do this have much poorer outputs than the trendfollowers. My solution was to go discretionary.
DRM over at viewtopic.php?p=1152#1152
pointed out that some commodities had a negative expectation using the turtle trader rule. This is my experience with a number of systems but the value of BP or US or CT or LC or C which might not provide much revenue over 20 years is that they are fairly independent of the "good" trenders and provide equity highs at times when they might be in drawdown thus improving your sharpe ratio or MAR.
Interestingly this is system dependent --- Mark Johnson tested a bunch of contracts for correlation and then ran a test on them with the system 13. He concluded amongst other things that Cotton and FV where uncorrelated but their system open trade equity results were highly correlated. I ran aberration on them yesterday and the results were much less correlated (didnt have time to generate the equity curves for excel testing but eyeballing the TS charts showed much less correlation than marks charts at http://traderclub.com/discus/messages/1 ... 20030414pm
. More support for using multiple systems to improve overall results.
On the original topic I guess that the system parameters could be changed to reflect the move from a good trending towards a bad trending contract.
Posted: Mon May 12, 2003 12:43 am
Lately I've been playing around with the following idea: At each trading point, I have my trading system automatically calculate the profitability of each market over the last N-years and then only take trades in markets with a high enough % of winners. I am not quite done with the research yet, but initial indications show that taking trades in markets where the last three years have a high percent of winners (the magnitude of profitability seems less important), does provide a substantial improvement in profitability. The nice bonus is that I dont have to worry about selecting markets for my system. I simply throw all available markets at it, and let it pick the ones that have a track record of working with my system.
PS What may work with my system, may not work with yours. Make sure you run these tests yourself.