|
PureBytes Links
Trading Reference Links
|
On Sat, 24 Mar 2001 23:06:41 -0800, you wrote:
>I think when applying Jurik's JMA you must think of it in terms of
>non-linear statistical analysis. IE: Whereas a straight line can be drawn
>through a data set using a least squares regression, JMA is a non-linear
>regression using a proprietary error correcting technique. Infact, if you
>look at "R-squared" as being a measurement of how "well" data fits a
>linear regression line, (I believe Mark took this approach) you might also
>perceive that Jurik's CFB is a non-linear R-Squared measurement,
>especially if you do a "CFB ratio" and use 0 and 1 as your min and max
>values and compare this to the TS R-squared indicator. (or even ADX for
>that matter, although I prefer R-squared to ADX.)
Your comment tends to run the risk of confusing the approximation
_model_ (e.g. straight line, any non-linear approach, periodicity,
etc) and the _goal function_ (e.g. r-squared = min, mean error = min,
max error = min, etc) here, while both of them can be modified
_independently_ from each other in any optimization task, like e.g.
approximation.
But you completely missed my point that immanent integration effect in
most indicators can be compensated by differentiation (, where both
effects address the approximation _model_:
>rudolf stricker <lists@xxxxxxxxxxx> wrote:
>
>> Lag in many or most indicators is a consequence of some immanent
>> integration effects. Therefore, lag can be reduced considerably by
>> applying some amount of differentiation to the indicator.
mfg rudolf stricker
| Disclaimer: The views of this user are strictly his own.
|