z****l
发帖数: 72 | 1 Hi there,
I'm doing time series analysis and trying to develop a method to describe
the property of the underlying process.As you know, there are many
approaches in the literature.
My point is to get some indications in short time series. My question is how
short the series is short? Anyone knows how to quantify it?
I'd like to test against the power spectrum.
thx | O*********r
发帖数: 1835 | 2 Let me try to offer my two pennies...
I think how short is considered short all depends on what process you intend
to study. For example, if you study seasonal cycle of, say, U.S.
temperature and you only have 3 months of data, it's for sure "too short"
because you don't even have samples of a full cycle (which is 1 year). But
on the other hand, if you study diurnal cycle or tidal phenomena, 3 months
of continuous observations are more than enough. So, there is no
universal criterion; it a | z****l
发帖数: 72 | 3 Ok, Good answer.
I'll try it in the following way.
Say, we have two different sorts of time series. I compute the power spectra
for both cases with a proper length, with which I can distinguish them.
Practically speaking, the number of points needed for the power spectrum is
much bigger than that is needed in my own method.
I conclude that I can distinguish the dynamics in short series. What do you
think? :-) |
|
