Hi, I've been a member here for about 7 years and like most of you there are several indicators I follow, I have one that is my favorite that I have created, and then there are 1 or 2 others that seem to add "a lot" to the accuracy of the first indicator. Obviously, "a lot" is subjective, and I, being somewhat OCD, would like to use the system tester to find how much, and also, how much "weight" I should assign the secondary indicator.

For example let's look at the CTI which is a fantastic indicator included with metastock.

CTI:=(((ADI/(ADI+DCI))-0.5)*100)*(p1/100) +
(((NNH/(NNH+NNL))-0.5)*100)*(p2/100) +
(((USV/(USV+DSV))-0.5)*100)*(p3/100) +
ROC(MKT,p4,%)*(p5/100);
Mov(CTI,p6,E)

In this case, different weights are assigned to the different components, specified by the p1/100, p2/100, p3/100 etc.
First question, is there a way to use the system tester to optimize these to find out what are the optimum weights to use for these indicators??? They would all have to add up to 1, and that's where I'm having problems with coding.
Can you specify somehow that p1/100 + p2/100 ... p6/100 = 1 or opt1 + opt2 ... +opt6 = 1???

Thanks for any help or direction!
Biff

Hi yes, here is the code for the Indicator, however I'd like to use it in the system tester to optimize the percentage that I could assign each variable.
Code Below****

{This indicator is a combination of the ratios of the Advancing issues vs Declining issues, Advancing Volume vs. Declining Volume, New Highs vs. New Lows, and the rate of change of the underlying exchange. The result is smoothed by an exponential moving average. The default weighting of the ratios and the length of the average is aimed at measuring a relatively moderate time frame.}

p1:=Input("Advancing and Declining Issues %",0,100,25);
p2:=Input("New Highs and New Lows %",0,100,30);
p3:=Input("Up Volume and Down Volume %",0,100,20);
p4:=Input("Periods in Rate of Change",1,200,12);
p5:=Input("Rate of Change %",0,100,25);
p6:=Input("Exponential Moving Average periods",1,200,12);
ADI:=Security("C:\MetaStock Data\BM Data\X.NYSE-A",C);
USV:=Security("C:\MetaStock Data\BM Data\X.NYSE-A",V);
DCI:=Security("C:\MetaStock Data\BM Data\X.NYSE-D",C);
DSV:=Security("C:\MetaStock Data\BM Data\X.NYSE-D",V);
NNH:=Security("C:\MetaStock Data\BM Data\X.NYSE-H",C);
NNL:=Security("C:\MetaStock Data\BM Data\X.NYSE-L",C);
MKT:=Security("C:\MetaStock Data\BM Data\.NYA",C);

CTI:=(((ADI/(ADI+DCI))-0.5)*100)*(p1/100) +
(((NNH/(NNH+NNL))-0.5)*100)*(p2/100) +
(((USV/(USV+DSV))-0.5)*100)*(p3/100) +
ROC(MKT,p4,%)*(p5/100);
Mov(CTI,p6,E)

This problem is where Neural Networks (NN) come into play and where they do their most powerful work. In lieu, using the EST "optimisation" (brute-force curve-fitter), we have 6 Optx variables, so express them as a percentage of the whole:



You'll need to walk-foward your test results.



wabbit [:D]

[edit] You're also going to need one kick-ass CPU and a lot of time to run this through the EST.

opt(1-5) : [1...100] => 100^5 = 1e+10 passes, multiplied by your EMA lengths (opt6), say, [1..100] again => 100^6 = 1e+12 passes (1,000,000,000,000) and remember: we have required the use of a PREV function!