Craig, You brought up the peak/trough ratio suggestion -- I noticed fairly quickly in the system I was experimenting with that not all mountain ranges are equal, and that appears to make a difference. In other words, I think that a shallow preceding trough tends to lead to different subsequent price behavior than does a deeper trough -- yet my system didn't take trough characteristic into account. I'm not sure yet whether a shallow trough is better or worse than a deeper one, but I do think it statistically makes a difference, and has to be further explored.
I'll try out your other suggestions.
One additional observation: I use fml(" ") fairly extensively, and sometimes one fml(" ") calls another -- remember the space limitations, etc., in each custom formulation.... Well, as I run the system test right now, I can actually count along with the test for *each* iteration -- one-one thousand, two-one thousand, three-one thousand, etc.... Thus, optimization across about 4000 iterations/combinations takes over an hour!!! On a 350 MHz Pentium II. My gosh. The manual does say that custom function calls will take longer, but this is ridiculous. I wish Windows 98 had some good disk caching algorithm, or I could do something.... Any ideas??
Thanks,
Nicholas |
