|
PureBytes Links
Trading Reference Links
|
Keep in mind that a MA is a low-(frequency)-pass filter. That is, it passes
through most variations of longer duration than the MA span but removes most
(or all) of shorter-period variations. To use one for revealing a cycle,
therefore, one should start with a MA having a span of, perhaps, a quarter
of the length of the cycle of interest and compare its result with that of
successively longer-period MAs. What you want is the shortest-period MA
which produces the desired degree of smoothing (removal of short-period
variations) withOUT significantly reducing the size of the desired cycle.
You can take this one step further by also calculating a MA with span equal
to the cycle of interest and then displaying the difference - short period
MA minus long period MA. This gives you a rough band-pass filter.
Carroll Slemaker
----- Original Message -----
From: "Monte C. Smith" <mcs@xxxxxxxxxxx>
To: "Gary Funck" <gary@xxxxxxxxxxxx>
Cc: <omega-list@xxxxxxxxxx>
Sent: Friday, September 29, 2000 5:30 PM
Subject: Re: Exponential moving average
> Gary Funck wrote:
> >
>
> > Sounds interesting. If you don't mind sharing the "discovery", would
> > you please explain how you use the MA lag to determine the cycle period?
> >
> > [...]
>
> Sure, but it's worth noting at the outset that I soon learned that my
> 'dicovery' was a well known phenomenon with some folks.
> There was a trader once who started using a neural net to generate buy
> and sell signals for the emini. Sometimes it gave good signals,
> sometimes it cost him money. He started trying to see into the black box
> (the NN) by various means, in an attempt to learn when to take signals
> and when to pass. He himself had not selected the NN's inputs nor had he
> trained it.
> At one point he decided to run a MA crossover optimization -- if I
> remember, it was on the equity curve -- and tested for best ROA. Before
> running the optimization, he had neglected to filter the opto
> instructions, so he wound up with a long list of combinations whose
> length parameters were 'invalid,' i.e., the short and long MAs were
> inverted, so the MA labeled 'fast' actually contained more periods than
> the MA labeled 'slow.'
> After the run was completed, the best ROA was found when the inverted
> 'fast' MA crossed above the inverted 'slow' MA. In reality, this meant
> that the best ROA was achieved when the slower MA crossed above the
> faster MA.
> It turned out that the reason this bearish crossover signal delivered
> the best ROA was because the crossovers were the same distance apart as
> the price cycle period. More precisely, the crossovers were occurring at
> the half-cycle, as Mark Jurik pointed out.
> So the LAG of the MAs can identify the price cycle period this way, and
> position you so you are 'in tune' with the cycle (remembering of course
> that the cycle could still change or disappear altogether).
> I'm sure there are a number of people on the List that could elaborate
> on this, and hopefully someone will, but that's it in a nutshell.
>
> Regards,
> Monte
>
|