|
PureBytes Links
Trading Reference Links
|
On Sun, 11 Feb 2001 09:30:39 -0500, you wrote:
>The problem when you advance past moving averages isn't lag. The real
>problem is "stationarity", how accurate are past data points at predicting
>future data points? This is the common problem that we have all seen: even
>if we find a perfect moving average with the perfect time period on past
>data, how good will it be at predicting future data? So even if you find a
>moving average or other filtering method that "has no lag", it doesn't mean
>that you can forecast the future. Markets do not have "stationarity".
I personally find your "stationarity" condition a little bit too
restrictive to identify a predictable process. Also non-stationary or
even (low-dimensional) chaotic processes can be predicted, at least
for short time windows (which may be sufficient for successful
trading).
The extrapolation (or generalization) capabilities imo do not depend
primarily from the selected model or type of indicator (e.g. moving
average) but much more from the "optimization process" (and its goal
function), which is used to adjust the free parameters of the model to
the given data. Of course this "optimization process" is a very simple
one in standard moving average calculation, with no generalization
effort, like multi-crosswise correlation etc.
mfg rudolf stricker
| Disclaimer: The views of this user are strictly his own.
|