You can improve the performance of your systems by developing a normal operating framework for each of them. Do a sensitivity analysis on the parameters and determine the best/worst conditions under which they operate. Then screen for the conditions and suspend the system when the environment is adverse to your parameters. If you find your parameters only work in a narrow environment, build a second model with different parameters and screen for that as well. Then, split the allocation of funds between the models. If the returns correlation between the two is under .2, you will be well served.
You are using an out of date browser. It may not display this or other websites correctly.
You should upgrade or use an alternative browser.
You should upgrade or use an alternative browser.
How to choose the right system for the current market?
- Thread starter elit
- Start date
Quote from alittlebird:
You can improve the performance of your systems by developing a normal operating framework for each of them. Do a sensitivity analysis on the parameters and determine the best/worst conditions under which they operate. Then screen for the conditions and suspend the system when the environment is adverse to your parameters. If you find your parameters only work in a narrow environment, build a second model with different parameters and screen for that as well. Then, split the allocation of funds between the models. If the returns correlation between the two is under .2, you will be well served.
Isn't that really like the moving average in my first post? I.e. If markets are trending up then use the appropriate system...
Just a word of caution about monitoring your systems and comparing actual performance with backtesting results. The markets are very good at misleading people, and so keep changing patterns to hide their "hands". Sometimes, one needs to stop overanalyzing its sytems. Most of the times, there is nothing wrong about them, but just temporay anomalies in the market, which, eventually, goes back to previous patterns. Obviously, this is more art than science. What if your actual drawdawns are deeper of more often than backtested ones? Is it time to change, abandon the system ? For me, a solution is to keep doing research on how to improve the current systems and developing new ones, with new, fresh ideas, test them and eventually implement them. This is what I am doing, working on new ideas, and testing them. It is a lot of fun. And it keeps me away from monitoring too closely my existing systems. 2 for the moment.
Quote from elit:
Is no one switching between different systems and advocate that?
I would consider it but only in extreme cases. It's like saying "I'm 100% sure that this system will absolutely not make any money in this environment so I'm switching it off completely and I'm 100% sure this other system will make lots of money in this environment so i'm maximizing my exposure to it", but the market is not always that clear-cut and doesn't change in an instant, and you're often surprised, so why not make it a gradual process of adjustment of exposure...
Doesn't it make a lot more sense to allocate more money in a system that is currently in a drawdown, based on the assumption that it will come out of that drawdown?
I have heard of people using moving averages on the equity curves of individual strategies to determine where to allocate capital. This doesn't assume certain market conditions favor certain systems, it just knows that they are profitable.
itmediaco,
when calculating correlation coefficients, why do you calculate it off the logs? I have heard it is something to do with taking out the "drift" that occurs over time, but if you could elaborate, I would be grateful. Thanks.
itmediaco,
when calculating correlation coefficients, why do you calculate it off the logs? I have heard it is something to do with taking out the "drift" that occurs over time, but if you could elaborate, I would be grateful. Thanks.
you have to check, using backtesting, if your systems outperform markets during those specific periods. if they not-then you don't have any system.
good one must work in any markets with steady PL.
you can try to use MM combined with something like market breadth and sentiment to allocate certain amount of money to each system and trade them all @ same time. easy to say, not so easy to implement and backtest.
good one must work in any markets with steady PL.
you can try to use MM combined with something like market breadth and sentiment to allocate certain amount of money to each system and trade them all @ same time. easy to say, not so easy to implement and backtest.