6.371 Double EMA???
- Thread starter tampa
- Start date
DEMA(X,N) = 2* EMA(X,N) - EMA(EMA(X,N),N)
The combination of twice the EMA minus EMA-of-EMA gives this filter some interesting properties. First, it has a much heavier weight on recent data than a plain EMA. Recent data is weighted nearly twice as heavily as in regular EMA. This weight falls off more rapidly with time than with regular EMA, though. The weight on the data exactly N-periods before the present is actually 0 and the weight on the data older than N periods is actually negative. These negative weights hit a valley and then fade exponentially to -0 as we look deeper into the past.
The combination of positive weights on recent data (before N-periods ago) and negative weights on older data (more than N-periods ago) creates a kind of momentum projection. Thus, DEMA has lower lag than EMA.
tampa, I've got a meeting set up with Mathis and Rundy in London next Saturday and hopefully I'll be able to learn a little more about their groundbreaking work. Unfortunately they won't be officially speaking at the conference and I had to use the influence of a mutual colleague to arrange the meeting. They are very publicity shy - typical academicians.
http://www.sta-uk.org/ifta_2002.htm
http://www.sta-uk.org/ifta_2002.htm
I know that there ain't no Holy Grail, and I know that a lot of things look good - at least for a spell. But this is incredable. The report on their site is just astonishing.
I would trade it without hesitation, if I could get my hands on a software package that could do the work.
BTW - I am green with envy over your meeting
Sounds like we have two definitions of "Double EMA" I'm refering to the one developed by Patrick Mulloy in the Jan 94 TASC and summarized on p121-123 of "TA from A to Z" by Achelis. Anyone know for sure which one is used by Mathis and Rundy?
RE: fractions: Unlike simple moving averages (SMAs), the math on all EMAs works with any number of periods greater than 1 (weird stuff happens if N<1). The exponential percentage (which is the weight on the most recent data sample) is 2/(N+1). So, if N = 6.371, then this percentage would be: 27.13%.
RE: fractions: Unlike simple moving averages (SMAs), the math on all EMAs works with any number of periods greater than 1 (weird stuff happens if N<1). The exponential percentage (which is the weight on the most recent data sample) is 2/(N+1). So, if N = 6.371, then this percentage would be: 27.13%.
Absolutely! Its all described in Achelis. Page 122 describes the spreadsheet for Double EMA, and p208 shows the spreadsheet for the regular EMA calcs that needed to do the two components of DEMA.
Happy smoothing,
Traden4Alpha