toby crabel uses a factor or percentage of the 10 day avg true range in his trading and i have systems that use it as well so its not a bad thing at all to use. the one problem is that if you have an anomaly like a large day it can skew future readings until its out of its system and what i do is ignore any reading outside a 3rd quartile reading. also you said something that a more complicated method would be needed to make money dont think that way ok be open minded. there are some really simple stuff making money.
ORB methods would be a good addition to what the OP has. ----------------------------------------------------------------------- Everyone seems to be talking about switches... First thing to do is to create individual "systems" that are profitable on a stand-alone basis. Once you have 2 "significant methods", you combine / coordinate them accordingly to appreciate each system's performance. It seems like people are trying to develop a supplementary element based on top of what has been developed. Once you know the vulnerability of the system, you have to approach the 2nd system in a seperate matter. The whole idea is to have a system that is profitable while the other systems are lagging, not to develop a single "perfect" system. Creating switches ("biased filters", really) to make one system fit all market conditions (Creating switches for bull market and another for bear markets) can be a very strong indication of development process vulnerable to curve-fitting. (Meaning, the way you develop model has a major flaw.) I mean no offense to people using "switches" but it would be a good time to re-examine the way you develop models. ... Though, switches in itself is a tough subject. As much as the implementation criteria (decision basis of whether you put money on the system) is important, maintenance is equally important. On/Off switch of a system (or re-adjusting the risk profile for the portfolio of systems) comes down to both risk management in a quantitative and qualitative sense. Switch should be placed under a risk management perspective, not within the development phase.
hi gtg, thats a good way of looking at it. ive actually gone back to start a new system (no point in keeping the other one i think) and what ive done is looked at great trending periods and looked at the shares that did exactly what i want them to do. ive now set a loose set of rules so that these shares come up in my scan for a certain time period. so now i have some basic rules. at the moment im trying to get some of the uglier charts out of my scan (even though they made a good profit) once this is done i will start backtesting. im also thinking about using some form of a switch. even if thats just using half position sizes when the market is directionless.
Hi Bo_D, This system has too many rules and uses too many lagging indicators ie. uses moving averages too much. Try to make the system as simple as possible because the more rules it has, the greater the chance that it will be curve fit to the in-sample data. Obviously, the walk forward analysis will be the ultimate test of whether the system has been curve fit or not. And your data MUST be split up into in-sample and out-of-sample data so the system can be validated. We are not trying to find a system that would have been profitable at a certain time in the past. We are trying to design a system that has a high likelihood and give us a certain level of confidence that it will be profitable in the future. Best regards, Nizar.
thanks nizar, well everyone thinks the same thing so ive obviously got a big problem to fix. i will have to try and limit the amount of indicators i use in my new system, im also not even backtesting until my scans are picking up the stocks i want. this will hopefully prevent curve fitting. any suggestions on leading indicators that may work on a short term trend following system?
I only use a Price breakout and rate of change filter (and liquidity). The latter especially is a very robust parameter and while it works to increase profits and decrease drawdowns, it also makes your equity curve smoother.
I agree. You must simplify this and rely less on lagging indicators. Backtesting this type of system is meaningless with all of the variable. On the right track. Good luck and Happy Trading.
Way to many rules and parameters. With so many degrees of freedom you should be able to make something that is 99% profitable Keep it simple, make sure it works if many types of markets (different overseas markets, different timesframes) and is robust to changes in variables...
only six parameters though??? the rest are stops and what market im trading. my new system has more. but most are just the same indicator using a longer time frame and shorter time frame. so i buy medium term trending shares that are going up short term as well. i had a loose set of rules, but there were stocks that came up i didnt like. these rules have made my system take the type of trades i like the look of (and rejected the ugly). it hasnt been backtested so it hasnt been curve fit. ive simply looked at shares that are trending very nicely and tried to create a system that picks them up. there are lots of losing trades i will NOT try and get rid of because the charts looked good before they went down. there are also shares that made a good profit but had ugly charts. so they were cut out.
Greetings -- I'll add the two cents worth I regularly add. It appears that all the results are in-sample. That is, all of the data has been used to develop the system. (Correct me if I am wrong.) In-sample results have essentially no value in predicting the likely performance on out-of-sample data -- data that has not been used in development. I recommend reserving a year or so (depending on the typical holding time per trade and the number of closed trades per year) of the most recent data. Develop your system using, say, 1995 through 2005. Tweak it as much as you wish. Then one time -- only one time -- run the system over the 2006 and 2007 data. Those results will help you decide whether to trade it or not. Note that repeating the process of developing over the in-sample period, then testing over the out-of-sample period will contaminate the "out-of-sampleness" of the recent data and make it just another part of the in-sample data. Which means that you will need still another out-of-sample data set. Thanks for listening, Howard www.quantitativetradingsystems.com