|
PureBytes Links
Trading Reference Links
|
Kent:
>Is what you are trying to do really the best thing to do? Even the
>JMA has parameters that users can use to tune (or "optimize") it
>for various applications. Shouldn't your MA have the same options?
>Don't you want to use it differently in different circumstances?
The current version does, but I understand the original one didn't.
The current version has an additional parameter that allows you to
adjust the inertia of the filter; not for purposes of fitting, but
for other reasons such as making more definite crossovers between
two JMAs of different lengths. My filter has an "interia" variable
also, which can easily be made into a parameter.
>Why isn't Max$ or Max Sharpe Ratio just as good as RMSE for judging
>fitness? There are dozens of objective methods you could use to
>determine goodness. Only you can determine which one is best. And
>that will be a very subjective decision.
Well, for evaluating a filter, regardless of the application, there
are precise mathematical ways to quantify goodness of fit, which can
be fairly simple for filtering a static set of data, but the measure
is a bit more complex when filtering a time series in real time
without the benefit of knowing future data. This is a DSP problem
independent of trading. I'm approaching this from a data signal
processing (DSP) point of view, not a trader's point of view. The
prolem is even more complex when the filter doesn't have consistent
frequency-dependent behavior, like these adaptive filters have (a
lowpass filter with occasional bursts of highpass behavior). So I'm
trying to get a handle on what the *mathematical* definitions of
"goodness" might be in such a case.
-Alex
|