If you read my other post (about starting tools & stuff) then you'll know I'll be trading very soon. Now since everyone tells me I should watch out, start with small lots, ... I've been thinking. Anyways if it works out well: no problem, but I have to say that I admit there's always a risk of failing. Now if this happens I'll cut my losses early enough and I might try a mechanical approach. I wouldn't be trading then: It would be core (or is it position?) trading: a few times a month, a year, the longterm kind of trading. Now I "developed" a trading system and I backtested it with wealth-lab, results vary between 150%-1000% (mostly around 800%) a year. Trading the longer term, I don't think I have to reckon with the slippage, ... I also intend to "trade" only the large cap stocks (QQQ, MSFT, SUNW). With an average volume of > 10 million shares. With this system I got a 589 % gain on SUNW in 1 year. I think that's fairly good. Of course, I'm only going to use this mechanical approach if thinks don't work out my way the "normal trading method". Am I being naive here, or did I forget something, because 600 % isn't nothing.? (And if I used margin/leverage that would be MUCH more)
Halcyon, 600% with what kind of risk? What are the drawdowns. How often is the system in the market? How many trades did the system make, (thereby allowing you to make statistical inferences.) nitro
You just have to try it and see. I am glad to see you are considering starting small. Many systems will back-test, but will not make money. It just works out that way. If you find a system that works, good for you. Just remember that no system keeps working forever. You will eventually have to change with the market. Good luck
Halcyon - Backtesting results can often be illusionary. Besides the usual issues like "could you really get executed at the prices the system is using", there's are basic validity of test issues. The first validity issue has to do with overoptimization on the backtest data. If you hone system parameters on a backtest dataset, then you really need to forward test it on an even longer time period. The backtest results should be taken with a grain of salt. It's the forward test results that matter. Then there's the issue of market structure during both the backtest and forward test cases. Certain systems that worked during the pre-2000 wild bull days crashed and burned when trading conditions changed. So optimally, both the backtest period and the forward test period should be long enough and encompass enough different market conditions to give you high confidence. Non-trading illustration - when I was doing spook work years ago, one of the teams developed an image processing system designed to identify camo'ed tanks in photographs. That tested it and tested it and ultimately achieved a 98% identification rate. The first time it was deployed in action, it failed miserably. It turned out that the system had been tested and optimized on only photos taken in bright sunlight. When photos on overcast days were fed to it, it failed. There's also the issue of statistical sample validity involving the generated signals. This involves both the quantity and magnitude of the results generated by the signals. In the first case, if you got a 600% profit but from only 5 signals over a two year test period - it's promising but is statistically inadequate to draw a reliable conclusion from because 5 signals isn't a large enough sample. Maybe it's actually picking tops and bottoms and riding them, but you can't tell. It might immediately start consistently failing. Now in another case maybe you got 100 signals over a two year forward test period from a system, so you might decide the sample size was sufficiently statistically valid to draw conclusions from. But if there was a significant skew in individual signal results there's a big question mark. For example, let's say your test results show a big summary net profit. But out of the 100 signals generated, 98 of them were effectively washes (part moderately profitable offset by the other part moderately unprofitable) but 2 trades were big winners that accounted for all of the overall net profit results. That's where being conscious of the sensitivity of the individual generated signal results is important. Do you conclude that you've got a big time winning system or a mediocre system that accidently hit a couple of flukes? Good luck.
if your results seem "too good to be true," then chances are the system will kill you when you actually apply it. There are many reasons why - previous posts on the subject have examined some of them. Or you can look at it this way: What are the chances that SUNW, MSFT, and the QQQ will, anytime soon, trade at all similarly to how they traded through most of the '90s, and then during the crash? You may be reluctant to give the details of your system here, but if it's based on trend-following, either breakout patterns or perhaps some indicator like MACD or DMI, maybe with some kind of parabolic or trailing exit, and if it pyramids at all, then the same settings that produced exponentiating returns on '90s data in favorite tech stocks may very well destroy an account very quickly, even if the next few years do not, as seems rather likely at this point, offer up a relatively choppy, non-trending market. A system that could dependably average 25% ROA/year, over a long period of time, in a range of non-correlated markets, with no or negligible RoR (risk of ruin), would be considered a tremendous achievement.
The returns are quite high for a system. How much data did you back test this over? A lot of systems that have high returns have a "marble...as my father puts it" in the system that produces a loss that is much more than all of the gains combined and than some. Make sure you don't have this marble in your bag. Robert Tharp
I agree with ArchAngel in that the forward testing is the most important aspect of the system development. A system that works as well on forward data as the backtested data is what you are looking to find. Good luck, it is not easy to find. Limit the number of technical variables to have a chance at finding it, otherwise it will never happen.
I, too, have begun working with automated trading. Using Trade Station I can back test my stategies to get an idea of how they might perform under similar market conditions. But what is forward testing? Does it refer to basically paper trading your system for a while?
you forward test the same way you did back test. It is a bit like paper trading, but it's paper trading by the machine with wich you back tested (ideally). This is very important because it will show you that the back test was an utopia and had bugs :eek: after the forward test you should go in live test. that is to check the actual slippage of the system and how many trades you get at the prices the system tells you should.. many surprises there too. we have good threads about back testing on ET. Do a search on them... also, a very important back test to do : out of sample data : check the system BEFORE the data you use to build it AND with OTHER MARKETS. there are all kind of reasons why back test will fail in real life. All these reasons make most traders say it does not work.. It's silly. It's like the stupid saying 'I've never met a rich technical analyst'. However, what is true is the extreme difficulty to build systems and back test them properly. I can not even count the number of systems I just scratched to the trash bin. It is really extremely difficult However, it's a great way to learn how to trade because it does teach you a few very important things : - have a method - test it - have the discipline to follow it. If you use mechanical systems or not, later, these traits will help you make money. tntneo