|
PureBytes Links
Trading Reference Links
|
Trey,
>From the historical article Dennis found, the 2nd one that is 8 years newer
is the preferred
one just based on the fact that it is a true coefficient (nice properties).
Any advantage from the added effects of the earlier one would be lost in
confusion of use.
Not to say that the earlier one couldn't be better for analysis of effects,
but for practical use and
understanding the second one appears more robust.
But by all means try both, its your money, lol.
Phil
&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
Dennis,
That's what I found as well. I'm still not sure which version is
correct. You're correct about the perfect U value. I guess I overlooked
the meaning of that. Thanks for the input.
Trey
-----Original Message-----
From: DH [mailto:catapult@xxxxxxxxxxxxxxxxxx]
Sent: Tuesday, March 08, 2005 3:29 PM
To: Omega List
Subject: Re: Theil's U
Looks like there are a couple of versions of the U statistic.
http://www.math.yorku.ca/Who/Faculty/Monette/Ed-stat/0308.html
That said, I'm not sure it gives you much useful information about a
filter. A formula that didn't filter the data at all, Filter = Price,
would have a perfect U value of zero. My idea of a perfect filter is one
that gets rid of the noise without lagging the signal and I don't think
the U statistic of price vs. filter measures that very well. You'd
probably do better comparing your realtime filter to one that is allowed
to look at future data, e.g. a centered moving average in the simplest
case. Then you could use any number of statistical measures to assign a
performance rating to your realtime filter.
--
Dennis
|