Charles --
Actually, I've received quite a lot of helpful response from another
group, some of which are leading to the solution I sought. If
you're interested, I can send a more mathematical description with
graphs, etc., of the various time series in question, but it appeared to
me that this was not the place for such. Please let me know.
And thanks for your suggestions.
Robert
BTW, it's Kolmogorov.
At 04:22 AM 4/19/2010, you wrote:
Since nobody else has suggested
anything I suggest a few points. ( The lack of response is not altogether
surprising since you don't define your objective with much scientific
clarity.)
One issue is whether you want to test some properties of the series or
whether you want to test for similarity in the sense of correlation.
If the first, then one would have thought you would want to measure the
autocorrelations and partial autocorrelations and then compare them with
the mean in some arbitrary way.
If the second then you might consider cumulating the observations and
using something like the Kolmorgorov-type tests to see whether the series
are similar.
Alternatively and very crudely you might take first differences and
calculate the Baumol efficiency criterion which = mean - k *
standard deviation (k being arbittrarily chosen to reflect a weighting or
trade-off between the "risingness" reflected in the mean and
the unsmoothness reflected in the standard deviation. This is an old
measure suggested to measure investment performance but it might get you
started even though it ignores the statistical properties of the time
series.
Charles Ward
(1)
On 16 April 2010 17:09, robert pisani
<r.pisani@mac.com>
wrote:
- Two simple time series questions:
- 1- I want a measure of a time series that describes how closely
it
- resembles a given collection of time series. Thus,
the time series
- collection may be contained in column 2 to n of an Excel file,
with
- each column containing one of the collection, with the time series
to
- be measured in column 1.
- 2- Define a measure of a time series that measures how close it is
to
- a smoothly rising (but not necessarily monotonically rising)
- series. Thus using such a measure, one could rank a
collection of
- time series so that those most smoothly rising and fastest
rising
- rank highest/lowest using the measure. Thus the time series to
be
- measured may reside in column 1 of an Excel spread sheet, with
the
- second column being the numbers 1, 2, 3, . . . etc.
- The proximity of the given series to the collection or to
"smooth
- rising" should be measured in some rms sense. I
realize that these
- requirements are not completely defined. I'm looking for
ideas.
- _______________________________________________
- Gretl-users mailing list
-
Gretl-users@lists.wfu.edu
-
http://lists.wfu.edu/mailman/listinfo/gretl-users
_______________________________________________
Gretl-users mailing list
Gretl-users@lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users