Re: A question on time series

From: "Beliavsky" <beliavsky@xxxxxxx>
Date: 26 Dec 2006 06:13:03 -0800

MCI wrote:

Suppose I have a time series of 1000 daily observations of random
variable x. I want to see if the standard deviation of change of x is
proportional to level of x,

I suggest a different approach from what you describe below.

Assuming x is always positive, you can compute three types of "change"
time series:

(1) r(t) = x(t) - x(t-1)
(2) r(t) = (x(t)-x(t-1))/x(t-1)
(3) r(t) = log(x(t)/x(t-1))

If the distributions of changes computed by (2) or (3) are closer to
normal (using one of the standard tests of normality) than those
computed by (1), your hypothesis is confirmed.

.

Prev by Date: Re: Roulette: what to do after 100 'red'-spins
Next by Date: GLM: a bernoulli or a binomial response?
Previous by thread: definition of measurement of agreement
Next by thread: Re: A question on time series
Index(es):
- Date
- Thread

Relevant Pages

Re: A test for randomness?
... you do not know the mean and standard deviation ... the 14 sized sample fits the expected distribution. ... you want to use a test on normality to see if the sample ... There are a WHOLE lot of selection biases that could result ...
(sci.stat.math)
Re: A question on time series
... MCI wrote: ... I want to see if the standard deviation of change of x is ... so I construct a rolling window: ... To begin with you have to cleanse the 1000 observations of any ...
(sci.stat.math)