Tuesday, December 16, 2008

Size matters

Above are results for the HLC Low system for IWM, Qs and SPY.
Optimimized inputs are as follows: IWM 7,14 ; Qs 4,13; SPY 2,13
Average % profitable is 70%.
I haven't explored the importance of volume in the systems I've previously profiled, but here's a quick look at what happens when we add a volume component to the HLC Low system.
TS2000i code for the 3 performance reports below now reads:

Inputs: Len1(2), Len2(9);
If Close=HighestFC(close, Len1) and low=highestFC(low, Len1) and high=highestFC(high, Len1) and Volume=highestFC(Volume,Len1)
Then Sell This Bar at Close;
If BarsSinceEntry=Len2
Then ExitShort this bar at Close

The code highlighted in blue is the sole addition to the previous HLC Low code.

We could tweak the system a little further by making the volume length a separate input variable (Len3), but I'll leave that testing to my more motivated readers.

Note that the Inputs for IWM have been optimized to 2,9. The top optimized value is actually 2,13, which yields an additional $1000, but from a risk management perspective I prefer this max consecutive win/loss ratio in lieu of the enhanced return.

Optimized Inputs for the Qs are 1,12.

Optimized Inputs for the SPY are 1,14.
The average % profitable for the 3 examples has now risen to 83% with a considerably improved max consecutive win/loss ratio.
This system is just something to play around with and should be regarded only as a jumping off point for further volume impact explorations. Nevertheless, the win/loss ratios and % profitable numbers are very impressive and may warrant the time and effort of additional testing.
Keep in mind these are all SHORT ONLY examples. The long side trades perform no where near as well, no doubt reflecting the downtrend momentum of most of the test period. This can also be noted in Monday equity curve profile, where the first dozen trades produced only marginal returns since the markets were still uptrending at that point.


Bain said...

Please consider the following conceptual example:
Would you rather trade a system that over the last 5 years has a profit factor of 9 and profitable trade percentage over 75% but only performs well on SPY? Or would you rather trade a system that over the last 5 years has a profit factor of 2 and profitable trade percentage around 55% but that works on a much, much larger basket of stocks?

My concern is that if a system is only successful on one stock, can it still be a legitimate system, or does it suggest a coincidence?

Also, are you aware of any way for Tradestation to test a basket of stocks at the same time so that the results could quantify the robustness of the system? Is seems awfully tedious to test one at a time but I am not aware of a better way.

Your perspective on this would be greatly appreciated.


bzbtrader said...

This is a huge question and the answer really revolves around one's attitude towards risk management and their confidence in their system(s). I've detailed my own personal bias on these matters many times in previous posts. . which is why I focus on IWM, the Qs, and more recently GE. There are many traders who focus solely on SPY or DIA (or the futures) using a variety of systems. Having spent over 5000 hours programming and testing TS systems, I've yet to find a wiz-bang system that works well over an array of sectors.
Regarding the basket testing question. . best to check with TS support directly on this as I'm not up on the current features of TS8.4.

bzbtrader said...

The other approach to system trading I neglected to mention is using a basket of systems to trade just a few stocks. This is my MO and I find it delivers the level of risk control that I require. As I also mentioned in many previous posts, it's not necessary to actually trade these specific systems, but to use them as a basket to gauge momentum for premium decay strategies or swing trading.

Bain said...

Thanks a lot for your time. Many of your comments are reassuring - granted I've spent a little less time programming (as in one tenth). I will strongly consider testing multiple systems on fewer stocks and I always find your archived posts helpful and/or thought provoking. Thanks again.