Without the algorithm, there is no saying how well this system performs over time. EminTrader has said that he has tested over a 10-year timeframe, but it is his word against ours, without the proof. I really commend him on the documented results over the last several months, but proof is in tangible long-term results, or in documented backtesting. I've seen and backtested thousands of systems that have performed well over short durations (even as long as 1-2 years), but have failed over longer stretches. Proof is in the long-term viability in all different sorts of market conditions. I need to see 10 year results before I buy in, lock, stock, and barrel.
i totally agree with you. I am also impressed by eminitrade's results for the last several month which prompts me to do the above study. without his algorithm, it is hard to be convinced the system will stand duration of the time. I believe OP has done the test for longer time frame otherwise it is dangerous to commit real capital.
And I appreciate your testing here, Trend. Interesting characterization for sure. The use of ATR is something that intrigued me from a prior post in this thread as well, and your results are quite revealing. I'm currently in the hunt for some 1-minute data so that I can dust off my back-testing S/W, and do some due diligence in this arena as well.
Very nice research trend! That must have kept you busy for quite some time "you can see the three day ATR gives almost similar results as L5" The ATR almost generates double the profit....have you tried optimizing the parameters for the large timeframe? I am trading the system for real now, over only the last 3 days I have seen a 13% return using the ATR and the basic system rules (+the going short with the system). Of course the past days have been quite extraordinary....I really doubt if on the long run the systems will perform as good... Best comparison of this system is with: http://www.youtube.com/watch?v=Il5TBgD9kHI&feature=related I never expected such a simple system to perform so well...but it's really amazing:eek:
Results : HCBK : Buy at 11.915. Close at 12.16 So + 2.0561% MRVL : None RIMM : Buy at 66.047. Close at 67.03 So + 1.488% JNPR : Buy at 18.155. Close at 18.31 So + 0.85% SCHW : Buy at 17.049. Close at 17.418 So + 2.1638% 2009-04-22 : HCBK : L5 = 2,0992% MRVL: L5 = 3,301% RIMM : L5 = 2,493% JNPR : L5 = 2,2844% SCHW : L5 = 2,2117% Have good trades... Overall bilan : + 125,5613% FSLR: L5 = 2,8512% SKF : L5 = 4,4556%
I spent couple of hours working on the analysis. I did try to optimize the ATR for longer time frame. For example, if the previous day is down, the profit is better. Even though the optimization produced better results, i still have doubts over those optimization if they work for even longer time frame. Since emintrade mentioned his L5 was derived from human behavior, I tend to believe his L5 will work for more difficult time periods.
Hi, Trend2009 said : "so the results become somewhat interesting. the results indicate the time frame from 2008/12/22 to now is quite different from 2007/08/01 to the current. Since I do not have value of L5 before 2008/12/22, I am curious how L5 would work starting from 2007/08/01. " -> i can say you : very well. But you have reason : don't try me ! That's why i I post IN REAL my orders. Never forget that a good programmer can find crazy result. You arent' behind the scene to see the code !!!! I will soon send you the proof.... As i said in a previous post, i use PROREALTIME. And it is limited for testing at 100% my strategy. I can automate 95% but not 100% because they doesn't permit to have different time frames in the same code. And i need "daily"datas and "intraday" datas at the same time to be rigorous. So i needed to make manually a lot of test (many times !) and i find a 5% margin of error so my automated tests were really realists. So if you are interested i can give you these results ? I have APPL for example while 18 years... "This analysis is by no means to imply that L5 would fail as 3-day ATR. If L5 can perform as well as from 2007-08-01, it is truly amazing, and I believe emintrade has already done such kind of analysis." -> YES. Never forget for building a winner system in a long-term : think first at your PHILOSOPHY OF TRADING. When you have the good idea THEN backtested it if you can...If you use ATR-3, think : Why ? If it is because i have just good results than you really have good luck that your system be wrong in long-term. -> If 1000 people use the same trading system in the same stock you are sure your system will disappoint you !!! So tell me if you are interested with my backesting (true at 95% it means : if i find overall% gain of 200% in fact it must be about 195%).
"Since emintrade mentioned his L5 was derived from human behavior, I tend to believe his L5 will work for more difficult time periods." Good analysis ... My L5 are derived from stock market behavior = derived from human behavior. I don't try to predict stock market, long or short. That was my error while many years. "optimization " = optimization THE PAST...That's why it can be dangerous.
I can't comment on the behavioural analysis part, but I doubt it's possible to properly backtest using human derived parameters. Since you'll have to use some kind of human benchmark for human sentiment or human actions on price action, which cannot be automated easily. You state that optimizing doesn't work in the long run, you're right, that's because markets are constantly changing. But doesn't human behaviour in the stock market also change over time and more importantly: together with the market? To solve this problem I have been working quite hard to integrate recurrent neural nets in the strategy to learn from past human price behaviour. Results so far are very promising so far. The recurrent nets keep learning while the system is running, so less worries about old models.
laurentb, you said you were using neural net to find rules. I already have had that experience for two years. My experience is that, neural net will give your lots of winning strategies from backtest, but you have to analyze each strategy if it indicates some human behaviorial pyschology. otherwise, pure data mining will lead you nowhere. just my 2 cents. human behaviour in the stock market might also change over time, but it is more reliable than rules from pure data mining. there is nothing new on wall street. what intrigues me about emintrade' L5 is that the entry price is always L5% higher than open. How does human behavior play into a number (here L5) regardless of the open price of tomorrow? I wonder if emintrade can explain a little bit.