NinjaTrader_Dierk
ET Sponsor
Interesting, I'm always looking for methods to messure ...
How did you calculate trendiness and volatility ? What did you use ?
Lessons Learned From Profitable System Development
- Thread starter TriPack
- Start date
Interesting, I'm always looking for methods to messure ...
How did you calculate trendiness and volatility ? What did you use ?
NinjaTrader_Dierk
ET Sponsor
Hmmm, I would state: Only take roll-forward results, never take in-sample for "real".
Now..that's really food for thought this weekend
I also took a similar approach to use different breakout models to capture different market behaviour, but I was only able to discern three categories of breakout, volatilty breakout, opening range breakout and reversal breakout. Base on these concepts , I then create variation models using different measures of volatility or reference points.
I hope we are talking about the same thing when I use these terms
). I have looked at the performance summary of prod5 you posted on another post and the it really kick my system 's ass . Some of your other sub-catorgories (mid day continuation breakouts, time of day based breakouts, trend day breakouts)
seem to be pattern-based( definition of a trend-day?) , from my relatively little experience, these system seems to trade less (less overall profit) because the patterns act as stict filters and they also "seem" ( I am being very subjective here) to be less robust due to the increase in the number of parameters to discern the patterns.
I had some success in developing such system( their profit factor always beats the plain-jane breakouts) but they have never made into my arsenal because I always thought I was curve-fitting. But you have really made me think again. Are such system less robust to a extent than other breakouts systems?
My other question is how do you allocate capital to these system.
Because the trading frequency/overall profits of the models differ, do u give different weights to system to balance it out or do u risk the same 0.5 percentage( I remember you said this somewhere) on all models?
As another complication, do u use a seperate equity curve for each model or lump them together for position sizing purposes. So that if a system does better, the profits are invested back into the system?
I think I can share some experience here on systems designed for index future ...
1. Different stage of a trader (beginner, experience, pro) requires different types of trading system. One size does not fit all
2. Beginners with less capital tend to need systems that trade more often, smaller targets and most important of all, very tight stop. Percentage winning is not that important. Key is to grow the account slowly and safely.
3, Most beginners like to widen their stops to increase the P/L and/or the winning percent. Due to curve fitting, they will likely experience the worst drawdown of the systems they've created when they go real-time. Use monte carlo simulation to check out the worst possible drawdown, that is usually bigger than the one reported from your backtesting program.
4. Worst case analysis - what if you do not get fill because your trades depend on limit prices, check out the worst case scenerio using your backtesting software - if it does not support that, do it by hand.
5. After successfully trade a low risk system real-time for a while, a beginner will acquire the basic skill set in terms of execution, reality check of a system comparing to real life condition, and some basic ability to analysis a system. Most important of all, ability to stick with the system rules
6. When you are ready for higher risk model, you can then design your new higher risk taking systems based on discoveries you've gathered trading the first few low risk ones.
To summarize it all, it is like learning to ride a bicycle. Most people cannot learn by jumping onto a real bike and start riding a steep down hill slope. It is easier to take it one step at a time.
1. Different stage of a trader (beginner, experience, pro) requires different types of trading system. One size does not fit all
2. Beginners with less capital tend to need systems that trade more often, smaller targets and most important of all, very tight stop. Percentage winning is not that important. Key is to grow the account slowly and safely.
3, Most beginners like to widen their stops to increase the P/L and/or the winning percent. Due to curve fitting, they will likely experience the worst drawdown of the systems they've created when they go real-time. Use monte carlo simulation to check out the worst possible drawdown, that is usually bigger than the one reported from your backtesting program.
4. Worst case analysis - what if you do not get fill because your trades depend on limit prices, check out the worst case scenerio using your backtesting software - if it does not support that, do it by hand.
5. After successfully trade a low risk system real-time for a while, a beginner will acquire the basic skill set in terms of execution, reality check of a system comparing to real life condition, and some basic ability to analysis a system. Most important of all, ability to stick with the system rules
6. When you are ready for higher risk model, you can then design your new higher risk taking systems based on discoveries you've gathered trading the first few low risk ones.
To summarize it all, it is like learning to ride a bicycle. Most people cannot learn by jumping onto a real bike and start riding a steep down hill slope. It is easier to take it one step at a time.
I wrote the programs myself. The idea for trendiness came from a article in TASC. Here's the link
http://store.traders.com/-v10-c01-futures-pdf.html
The volatility idea came from measuring a rolling mean and standard deviation. I'd have to go through an old notebook to find out the formula. It's simply measuring the dispersion around the mean for any given period.
I use the trend rating to rank the futures markets for a given period. It helps with researching trading methods. Here's the ranking for a 30 day holding period for this year.
NinjaTrader_Dierk
ET Sponsor
Thanx for sharing ...
Dierk
SO, SO TRUE, Smitty. Having spent several years woodshedding with the wonderful TechnifilterPlus (EOD only, darnit) has been such an eye opener. I am amazed at all the conventional TA wisdom that is still so highly touted that is just worthless. So many newbies get screwed up by believing this stuff. Testing is the great truthteller.
g