fbpx

How Long is a Trading System valid for?

How Long is a Trading System valid for

Published May 22nd, 2020

Q. When you go live with a system, how long do you actually expect it to last for and continue to give you a trading edge. Do you anticipate that it may only be good for a certain amount of time, whether that’s a couple years? Or, if it’s truly robust, should it continue to work for maybe ten or more years?

A. If you have a truly robust system it should continue to work extensively. Now, that’s not to say that you can’t learn more things about the market, you can’t learn more things about your system, you can’t learn more things about yourself, and therefore you can make adjustments to that system. I think every strategy is a work in progress. We rarely make adjustments but we certainly have made adjustments in the past to our systems because we have learned new things or technology has changed, which has enabled us to test our systems in different ways. A very good example of that is I can now strip out resource stocks from industrial stocks, as an example. I can test those different segments of the market because they’re very, very, different kinds of stocks.

Most resource stocks act very differently from industrial stocks. We even see that in the fundamental world. There’s a lot of fundamental investors (Warren Buffett being one) who will not touch resource stocks, as an example. They will only touch industrial stocks. With the technology and data that we now have, which we didn’t have ten years ago, we can now test that and we may be able to draw conclusions that, “Oh, you know what? I really shouldn’t be trading resource stocks with my trend following strategy because they act very differently to industrial stocks.” Now to some extent that is a little bit of curve fitting, but again I stress that so long as we have a significant sample of trades across all sorts of different stocks, and we’re using the robust system, then that should be okay.

The more you curve fit, the more you optimise, the more likely that the system is going to fail. You tend to hear people saying that all trading systems will break. That’s generally because they’ve optimised it to start with. It’s an ongoing process of maybe making a slight adjustment here or a slight adjustment there. You’ve got people like Salem Abraham, you’ve got Eckhart, you’ve got Dunn, you’ve got all these guys that have been trading trend based systems they’ve been trading the same systems, slightly adjusted here and there, for thirty years. There’s no reason why a good, solid, robust system, cannot last a whole lifetime.

I think the important thing here is that too many people are trying to find the best system, and that’s the wrong way to approach it. What you’ve got to do is try and break your system. The more stress you can place on your system when you’re testing it, to try and break it, you’ll get a lot more out of it. So for example, testing your parameter stability. Let’s use an example in my book ‘Unholy Grails’, we use an index filter which defines the broader market. Now let’s say that index filter is a hundred days. What we want to test is what the result is when we change that by a certain amount. What if we change that to ninety or make it a hundred and ten, or make it a hundred and fifty. How much would that change the results?

So you can increase the single variants by changing all the inputs in your signal, and seeing what impact that has. You might say, “What if I move my signals input parameters by 15%?”, “What if I move them by 20%?”, what kind of an impact is that going to have on the bottom line? The other thing you can do to try and break your system is increase the data variants as an example. So let’s say you’ve got a system that –so some of my systems for example uses the opening price to exit a position. What happens if I randomize that open price by 10%? And test back for twenty years? Say the opening price is adjusted up or down randomly by 10%, what impact is that going to have? That’s the whole idea of trying to break the system and trying to disprove it, rather than prove it. If you can’t break it, if you can’t disprove it, chances are you’re probably on to something that’s really good and reasonably robust, and will continue to work reasonably well into the future.

Join The Chartist free trial