|
PureBytes Links
Trading Reference Links
|
> For this application (finding the dominant cycle in the range
> 5...50, say), I would question the use of a bypass filter of (15,20).
> This will chop out a lot of signal at the Len=5 and len=50 time
> periods.
I just picked 15 and 20 out of the air for demonstration
purposes. It happens to do a pretty good job, at least to my
eye. Depends on what level of cycle you're looking for.
And as I said, I don't remember the length -> frequency (bar
length) mapping, but I don't believe it's one-to-one. I'm not at
all sure that a 15,20 bypass passes only 15-20 bar cycles (and
surrounding cycle lengths, given the slow falloff of the xaverage
filter). But I've spent way too much time on this already so I'm
not inclined to experiment with it. :-)
> And further detrending seems unecessary.
Without the LRS there is still a lot of trend component. You'd
have to decide whether that's good for your application or not.
> I'd think the following is likely more
> adequate and accurate for the purpose of determining dominate cycle:
> LRS = LinRegSlope(Price - Xaverage(Price, 50), 5);
> Here, we assume 5 is the min. cycle of interest and 50 is the max.
Seems very noisy to me, but maybe it works well if you want those
higher-freq cycles.
Gary
|