Talk:Moving average (disambiguation)

From Wikipedia, the free encyclopedia

Contents 1 Never tells what it is 2 Moving average article is too narrow 3 Rolling average 4 Influenced by 5 hmm exp weighted average

[edit] Never tells what it is

Hmm, this seems a bit off. Why should we disambig to detailed uses without even simply saying what a simple moving average is? Any ideas on how to keep the disambig but cover what a MA is? - Taxman ^Talk 15:55, 21 December 2005 (UTC)

[edit] Moving average article is too narrow

As far as I am aware there is no special mathematical requirement for a moving average to be defined as averaging over previous data in a timeseries.

A 'backwards'-evaluated moving average always induces a shift in the data. I often prefer to to use a centred moving average, taking both 'past' and 'future' data into account, which induces no shift. There is also no restriction on using a moving average (however evaluated) to try to smooth any arbitrary function.

Essentially moving averages are used for smoothing, and this is a much broader application than the article currently implies, which is too heavily 'weighted' toward stockmarket applications. — DIV (128.250.204.118 09:08, 3 April 2007 (UTC))

[edit] Rolling average

I'm pretty sure the term "Rolling average" means exactly the same thing. I'm thinking this should go into the introduction. --Starwed 15:43, 10 June 2007 (UTC)

[edit] Influenced by

"An SMA can lag to an undesirable extent, and can be disproportionately influenced by old data points dropping out of the average."

The average is not influenced by data points dropping out of it, but by data points that have not yet dropped out.

84.147.231.109 22:51, 22 June 2007 (UTC)

[edit] hmm exp weighted average

Everything after "The weight omitted by stopping after k terms is" to "k=3.45(n+1)" is wrong. The series (1+a)^k + (1+a)^k+1 + ... leaves out the denominator of the omitted portion. The correct formula for the fraction of the omitted portion is (1-a)^k not (1-a)^k/a. For 99.9% weight: k = Ln(.001) / (Ln(1-a). For large value of N (>100), The natural log of 1-(2/N) approaches -2/N, therefore K =3.45 * (N+1) for large N is correct (with 99.9% weight).