interesting idea jj. a long term research goal for me now. have a system that does well on trend days and tries to keep itself out of choppy days, and a system that does the opposite.
That's a noble idea, but the problem is unless they are equally profitable it makes no sense to run both systems. Whichever is most profitable should receive all of your buying power. Only when you can't put any more money into a system due to slippage or other factors does it make sense to add another system (or at least until your system ceases to be as profitable as your other one). If I'm missing something here, please point it out.
I believe you are missing something. If the two systems' holding period returns are anything less than perfectly positively correlated, it is worthwhile adding some of the lesser system to the better system. Refer to Markowitz, MPT etc. And to common sense too
I don't see it being nearly that simple. Sure, if you have systems A and B, they show negative correlation and are roughly equivalent in terms of ROI, then sure, reduce your risk/drawdown and run both systems side-by-side. However, if system A is much more profitable than system B, and you are comfortable with and prepared for the drawdowns that system A exhibits in backtesting, why put money into system B to grow at an inferior rate? I think you are assuming the system to be roughly of the same caliber in terms of profitability OR a situation where your "more profitable" system provides you with periods of drawdown that you cannot handle. Or am I, too, missing something?
I think the idea is this: suppose you have system A that returns 40% and has 10% drawdown. You also have system B that returns 30% and has 15% drawdown. Because the strategies are negatively correlated, if you run both, you may see 35% return with 5% drawdown. Adjusted for risk, this new system is superior to each of its constituents. Magic (the Markowitz style), as waxwing alluded above.
Right, I would put that in the category of "roughly equivalent profitability", when taking into account the respective drawdowns. The general problem is just a risk/return tradeoff. The point I was trying to make is that the notion that the more negatively correlated systems, the better, is not an absolute truth by any means,
kohanz, You make an important point that is widely misunderstood - you should seek uncorrelated systems (ie: =0). Two systems with a correlation of -1 will tend to generate returns that approaches 0 over time. bolter
i think i was misunderstood when i said 'do the opposite'. i meant system A trades well on trend days and stays out of choppy days. and system B trades well on choppy days and stays out of trend days. A if X && !Y, B if Y && !X kind of thing. say system A averages 2000$ a day, system B averages 200$ a day. say for sake of argument the market switches from trend days to choppy every other day. ideally A would trade on the even days and make no attempt to trade on the odd days, running system B over top will not effect the capital usage of A since it will only get triggers on the odd days, and make no attempts on the even days. adding system B will increase expected daily profit by 10% (not a bad improvement), assuming the 50/50 split of trend/chop, otherwise could be more or less. in real life it wouldn't be that simple, at best I could have indicators which label the day as trendy or choppy, and choose one to run. possibly running both if the day is hard to define to try to allocate my capital to whichever bot thinks it has the best chance, the logic being if one thinks it can make money, then it is also preventing the other bot from losing money by using capital. i think due to the nature of these hypothetical systems it will be rare they step on each other's toes. at least thats the goal. backtesting is an easy way to see if the systems compliment each other or if B just reduces the expected profit of running purely A. since frost runs a counter trend system, it may be advantageous to start designing/testing a trend system which compliments it during periods where capital is not being utilized. having the second system may not only produce better capital utilization, but also reduce drawdowns of the counter trend system running alone, hopefully.
How can you tell if a day will be choppy or trendy in advance? Wouldn't being able to predict that be the "holy grail" in and of itself?
no need. if i did go the route of using an indicator to decide between systems, just choose the system with highest expected profitability during 'undecided' areas. in the case of not using a global day indicator, ideally the nature of both of the systems will decide for us, meaning B will 'tend' to trade when A doesn't want to trade => better overall capital utilization.