|
PureBytes Links
Trading Reference Links
|
Herman- Eckhardt gave two examples where the c-test is relevant
to computer backtesting (fractal efficiency and measuring trend
strength with the velocity ratio, as opposed to n-velocity). You
even listed one in your reply (optimizing Profit Stops and Max Loss
Stops). There are an infinite number of other possibilities that are
relevant today.
On optimizing multiple parameters, think of it like this. Take a
simple x by y optimization. You can do it all at once and get an x
by y surface, or you can optimize by stepping through each x value,
one at a time, and all y's; or each y value, one at a time, and all
x's. But then you'll get a bunch of lines that you'll still have to
put together to make the x by y surface anyway.
With more dimensions, you'll need to look at optimization slices to
really visualize what's happening (with a human brain anyway) but the
results will be the same, no matter what order you optimize in and no
matter how many parameters you optimize at one time.
BTW, I think six parameters is way too many. I never optimize more
than two myself. More parameters = less robust.
Mark
--- In amibroker@xxxx, "Herman van den Bergen" <psytek@xxxx> wrote:
> Thanks everybody for the responses.
>
> Downloaded the Eckhardt paper. Does anybody still use angles?
Today,
using
> computers, we would probably use ROC(). I think the C-test was more
relevant
> in the old days when fundamental parameters, which could be in any
number of
> different units, played a bigger role. And then there are those who
use
> astrology... I guess there is still use for the C-test.
>
> Thanks for the link to the Eckhardt Trading Company overview
> http://iasg.pertrac2000.com/SnapshotPT.asp?ID=114 ], that is very
> interesting. I do believe however that for the small trader who
trades
> mechanically in small lots (not millions), long and short, much
greater
> profits are possible. The thing that I admire is that they
conceived
their
> system so many years ago...
>
> wrt Optimizing, I have a gut feeling that not only do the
criteria's
differ
> for trading systems but also for parameters. For example a trading
formula
> with two "period type" parameters that are optimized using 100^2
iterations,
> might be made to fit many price charts with good results. On the
other hand
> if the parameters were Profit Stops and Max Loss Stops, for
example,
they
> would have less of a chance to over-optimize the system. What do
you
think?
>
> Another factor might be the sequence of optimization and whether all
> parameters are optimized in one step or individually. The case of
Thresholds
> was brought up awhile back. Here we have another case where we may
consider
> to optimize separately or in one operation.
>
> Suppose you optimize all six above in one operation, would that be
increase
> or decrease the chance of over optimization? Personally I optimize
one at a
> time and chart the result for each operation. This helps me get a
feel for
> what is happening. In the end, having a better understanding of the
system,
> I may go back to make changes.
>
> Best regards,
> Herman.
>
>
> -----Original Message-----
> From: MarkF2 [mailto:feierstein@x...]
> Sent: Sunday, July 14, 2002 7:47 PM
> To: amibroker@xxxx
> Subject: [amibroker] Re: How to spot over optimizing
>
>
> This is not necessarily about overoptimization, but all systems: the
> methodology MUST have dimensional coherency. See Stocks &
Commodities
> V. 12:5 (218-221): The c-Test by William Eckhardt.
>
> For those who don't know, William Eckhardt was Richard Dennis'
trading
> partner. Here's his documented trading performance over the last
> decade+.
>
> http://iasg.pertrac2000.com/SnapshotPT.asp?ID=114
>
> Mark
>
>
>
>
>
> Your use of Yahoo! Groups is subject to
http://docs.yahoo.com/info/terms/
|