PureBytes Links
Trading Reference Links
|
AM> I think probably the best way to do this is to generate a string of
AM> random numbers in a file, each number corresponding to a trading bar
AM> (meaning you need dates and maybe times), and read the file into
AM> a TS data stream like data2. Now you have a sequence of random
AM> numbers that will not change from run to run, so you can make valid
AM> comparisons of strategies using random entries or exits.
When optimizing a strategy's entries/exits based on a non-changing set
of random numbers, then in fact you might be curve-fitting your
entries/exits to that particular set of random numbers. A workaround
might be running some 100 optimixations for each of the variable
combinations being optimixed, and *averaging* the 100 observations for
each variable combinations. If optimizing over one variable, for
example, this can be achieved by adding an extra, "dummy", variable
that changes from 1 to 100 so you'll get 100 *different* random runs
for each variable value that is being optimized. Then you export the
optimization results into Excel, and do your averaging there.
Best regards,
Ivo Karindi
ivo@xxxxxxxxx
|