In partnership with:
CME Group

Best of TTU – Identifying the Right Parameter Set

In any systematic trading strategy choosing the right combination of parameters for your model or approach is critical. So today, I wanted to share a valuable takeaway from a conversation with Bill Dreiss, where we discuss his thoughts, experience and approach to identifying parameter sets which I think you will find to be rather different to most managers.  Bill and I discussed this as well as many other super interesting aspects of trading from his 40+ year career and I think these lessons will benefit you immensely.  If you want to listen to the full podcast episode then just click here.  click here

Identifying the right Parameter Set for Risk Management


Bill:  The big trend followers, and certainly trend following dominates the CTA space, you’d think they’d all just be cannibalizing themselves. I think this gets back to what I was saying, that I’m a technical trader that believes in fundamentals. So there’s a world out there that can’t be controlled or that’s beyond the reach of market psychology or market methodology and I think that world is subject to forces - generally longer term forces - that are pretty much universal and timeless.  


Niels:  Human behavior, yeah. 

Bill:  That go back in history as far as you want to go and that will go into the future as far. 

Niels:  You know, I agree with all that, Bill, and it’s interesting because there’s obviously still so much resistance, it’s fair to say, by a lot of people, certainly on the investor side, to embrace this and you always have to justify why trend following works. Even if it has a year or two of under average performance then it’s their case for why it has stopped working and it’s never going to work again. So certain things don’t change. 

I want to go back also to another point which I think differentiates you a lot compared to the managers that we see out there. Maybe you can explain more that where I’m going with this is that I know that you, or at least part of your system is not looking at parameters, meaning you’re not trying to optimize a certain parameter set while, if you use moving averages, or price breakout, whatever it might be, clearly a big part of the research is really identifying the right parameter sets to use. Explain to me about that and why you’ve chosen this way of looking at it. 

Bill:  Well, of course the idea of data fitting has been the nemesis of anyone who’s tried to design systems. So one of the attractions to the fractal approach was that you’re dealing again with very fundamental patterns, but you’re dealing with patterns as opposed to numbers. You’re dealing with pictures instead of the numerical approach. So in the first place, if you adjust the algorithm the way I’ve described it, is not a matter of optimizing on any kind of numerical parameters. It’s a matter of setting up a certain structure and then, in a sense, graphically utilizing that structure to translate that into patterns.  

"... went through and tried all the different possibilities and picked out the best one."


Now there’s certainly data fitting in the sense that you’re fitting what patterns that you think are significant versus those that you don’t. You’ve obviously got to have some choice there. For instance, trading weekly charts versus daily charts, that’s obviously a parameter that you’ve selected. But these might be numerical parameters but they’ve been selected on a qualitative criterion. They haven’t been selected because I went through and tried all the different possibilities and picked out the best one. It was a much broader type of judgement that was made.  

So the advantage, again is that, in terms of designing a system, you’re not really focusing. You’re coming into it with an analysis that’s based upon, shall we say, qualitative judgements about how the markets work and so on and so forth. Then that’s implemented more or less directly without having to go through a lot of optimization, testing, and so on and so forth. Obviously you’re going to backtest your methodology and if it doesn’t look like it works you’re not going to use it. So right there you are data fitting.  

Nobody uses a system that doesn’t test out. You certainly can’t avoid that kind of parameterization completely, but you can certainly keep it to a minimum and also avoid being deluded by it. I think the worst thing that comes from parameterization is you tend to think you’ve got something that’s a lot more, say magical, than is actually the case.

Niels:  When it comes to risk management, though, parameters are difficult not to apply: how much should I risk, etc. etc. Obviously people usually come to these conclusions based on research and so on and so forth. How do you, just sort of broadly, frame that in terms of just the parameter side of things, not exactly the risk management, we’ll talk about that a little bit latter? 

Bill:  Yeah, well but risk management is pretty much common sense. Everybody’s risk management is the same. Risk management is driven by the realities of the business. To some extent it’s a matter of choice. In other worlds, I’ve decided to operate in a certain, shall we say level of leverage. You find that the level I operate in is probably towards the high end among CTAs. If you get much higher than that then you are out of business.  


Most CTAs, as they get big, they tend to cut their leverage. So the typical CTA is probably about half of what I use. But again, that to me is pretty standard. In other words, I can just glance at somebody’s performance record or whatever – just basic stats, and I know what region of risk or of exposure that they’re operating in.  

If you look at those people, the people who are established and have been around for a while, they’re all in the same ballpark.

You look at me versus DUNN or various other people with similar shall we say standard deviations – monthly standard deviations or similar drawdowns, or whatever, we’re all doing the same thing. We’ve got about the same margin to equity. 

So I think that, to some extent, money management’s another issue. Once your system generates a series of trades, which may or may not be correlated. Typically, you’re going to have about 40% winners and 60% losers and on and on. That’s why I say that different people have different systems and, certainly from a marketing point of view, it’s nice to say, “Talk about how special your system is.” But you’re really pretty much driven by what the markets are doing and what they offer you in terms of possibilities.  

Someone who… A well designed system is going to do a reasonably good job of capturing what the market offers, but it’s more of either you’ve got one or you don’t, it’s not a matter that there’s a lot of distinction in terms of what the outcome is. Once again, the outcome is pretty much driven by the level of the amount of leverage that you take. If you normalize to that leverage, then anybody who’s been around for a long time is going to have pretty much the same performance. 

We have a risk management concept that overlays the portfolio that's based on marginal utility. So, we're harvesting profits along the way which is very different than what we did learn in the original Turtle trading programs. We're still doing the same things, just a little bit differently than we used to.

Also, if you enjoyed this short insightful takeaway from a past episode of this show then you will love the free book I’m giving away right’s called “The Many Flavors of Trend Following” and it includes some of my best insights on this perhaps the Most Dependable and Consistent, yet Often Overlooked Investment Strategy...Get you free copy here