1
$\begingroup$

If you'll look up for the definition of a (discrete) white noise process on the web, you'll find some sources that say:

"white noise is defined by zero mean, finite and constant variance, zero autocorrelation"

and roughly the same amount of them that say:

"white noise is defined by zero mean, finite variance, zero autocorrelation".

In the latter (e.g. the Wikipedia page), "white noise" encompasses also the case of heteroskedastic white noise - where the variance isn't constant. Obviously, stationarity is dropped.

Which of these two is "the" standard definition? Am I missing something?

Note: I submitted the same question in math.stackexchange. Whenever I get an answer, I'll delete the other one.

$\endgroup$
2
$\begingroup$

There is no common definition of white noise.

You have two definitions. And there exists more definitions. For example, white noise is a stationary stochastic process with constant spectral density (and thus infinite variance).

https://www.encyclopediaofmath.org/index.php/White_noise

$\endgroup$
1
$\begingroup$

A white noise process has constant variance. So the second definition is incomplete.

Ref: Armitage encyclopedia of Biostatistics. The def there for discrete time requires that the autocorrelation looks like: $\gamma_{r,s} = \sigma \delta_{r,s}$.

$\endgroup$
  • $\begingroup$ Unfortunately, there's plenty of (qualified) sources that omit the constancy of variance. Someone even calls weak white noise the heteroskedastic one...My opinion is that there's no consensus on the definition of "white noise", and that there's not a theorical motivation behind it🙁. $\endgroup$ – Lo Scrondo May 30 '17 at 11:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.