חשוב (לדעתי לפחות)
חשוב (לדעתי לפחות) I`m going to weigh in here ... as far as I`m concerned, for *small* accounts (precise definition to follow) nearly all technical trading systems are worthless. Those few technical trading systems I`ve seen that appear to work take a tremendous capital base and a psychological discipline unlikely to be found in someone who has not devoted his or her life to trading. There are a lot of ways to do this wrong and essentially only one way to do this right. First, you acquire a database of past price, volume and open interest data. If the database isn`t nearly flawless, you`ve got one strike against you. You divide that database into two sets, which we call the training set and the testing set. Both these sets should cover *long* periods of time, with *many* different economic and market conditions included. By the way, it isn`t a matter of number of data points -- you need just as many years of tick data as you do daily, weekly or monthly data. And the more individual markets, the better. So, if you`re working with stocks, I`d say you need 100 or more stocks that have been trading for at least 20 years, and preferably far enough back to include the long bear market in the mid 1970s. Next, you search over millions of possible mechanical systems on the training set. This is the easy part if you`re a competent programmer, but it`s next to impossible if you aren`t. The computer does all the work. Expect it to take a fair amount of time, even on the 1.333 GHz systems you can buy today. You should come out of this step with a small handful (no more than five) systems that are promising. Now you do a Monte Carlo simulation of these five systems on the *testing* data set. The numbers you got during the training set search are essentially meaningless -- what matters is how the systems perform on data they have not seen! First, you simulate the systems against the test data and record the exact trades made -- entry date/time and price, exit date/time and price, commissions, an allowance for slippage and the bid-ask spread. Again, if you`re not a competent programmer, this step will be impossible. The list of trades is then fed to a Monte Carlo simulator which will tell you if the systems are worthwhile and what your capital base needs to be. Fortunately, at least one vendor -- CSI Data -- provides such a simulator in their Unfair Advantage package. If you don`t have that, again, you need to be or find a competent programmer. If you don`t go through *all* this work, you`ve got another strike against you. Let`s assume you find one system that passes these rigorous tests. You now need two things: a large enough capital base (the Monte Carlo simulation should tell you this) and psychological discipline. You need these to ride out the drawdowns (again the Monte Carlo should tell you what the drawdowns are likely to be) and to take *every* entry and exit that the system gives you. If you try to second-guess the system, you have another strike against you. So, in conclusion, I believe it is *possible* to make a living as a technical system trader. However, it is *extremely* difficult, and most people who try it without doing the massive amounts of homework I`ve sketched out above are going to lose money that they can`t afford to lose. -- [email protected] (M. Edward Borasky) http://www.aracnet.com/~znmeb How to Stop A Folksinger Cold # 3 ``If you miss the train I`m on...`` Tough.
חשוב (לדעתי לפחות) I`m going to weigh in here ... as far as I`m concerned, for *small* accounts (precise definition to follow) nearly all technical trading systems are worthless. Those few technical trading systems I`ve seen that appear to work take a tremendous capital base and a psychological discipline unlikely to be found in someone who has not devoted his or her life to trading. There are a lot of ways to do this wrong and essentially only one way to do this right. First, you acquire a database of past price, volume and open interest data. If the database isn`t nearly flawless, you`ve got one strike against you. You divide that database into two sets, which we call the training set and the testing set. Both these sets should cover *long* periods of time, with *many* different economic and market conditions included. By the way, it isn`t a matter of number of data points -- you need just as many years of tick data as you do daily, weekly or monthly data. And the more individual markets, the better. So, if you`re working with stocks, I`d say you need 100 or more stocks that have been trading for at least 20 years, and preferably far enough back to include the long bear market in the mid 1970s. Next, you search over millions of possible mechanical systems on the training set. This is the easy part if you`re a competent programmer, but it`s next to impossible if you aren`t. The computer does all the work. Expect it to take a fair amount of time, even on the 1.333 GHz systems you can buy today. You should come out of this step with a small handful (no more than five) systems that are promising. Now you do a Monte Carlo simulation of these five systems on the *testing* data set. The numbers you got during the training set search are essentially meaningless -- what matters is how the systems perform on data they have not seen! First, you simulate the systems against the test data and record the exact trades made -- entry date/time and price, exit date/time and price, commissions, an allowance for slippage and the bid-ask spread. Again, if you`re not a competent programmer, this step will be impossible. The list of trades is then fed to a Monte Carlo simulator which will tell you if the systems are worthwhile and what your capital base needs to be. Fortunately, at least one vendor -- CSI Data -- provides such a simulator in their Unfair Advantage package. If you don`t have that, again, you need to be or find a competent programmer. If you don`t go through *all* this work, you`ve got another strike against you. Let`s assume you find one system that passes these rigorous tests. You now need two things: a large enough capital base (the Monte Carlo simulation should tell you this) and psychological discipline. You need these to ride out the drawdowns (again the Monte Carlo should tell you what the drawdowns are likely to be) and to take *every* entry and exit that the system gives you. If you try to second-guess the system, you have another strike against you. So, in conclusion, I believe it is *possible* to make a living as a technical system trader. However, it is *extremely* difficult, and most people who try it without doing the massive amounts of homework I`ve sketched out above are going to lose money that they can`t afford to lose. -- [email protected] (M. Edward Borasky) http://www.aracnet.com/~znmeb How to Stop A Folksinger Cold # 3 ``If you miss the train I`m on...`` Tough.