Last week on Statalist Susan Lewis asked for assistance in computing a
trailing 60-day standard deviation from a number of units' timeseries (that
is, from a panel). Various inventive strategies were proposed, including a
number that used the interaction of 'by' prefix and 'sum' (the running sum
function) to apply the stats textbook formula for the variance: sum of
squares minus the square of the mean. Although these were nifty solutions,
I thought they were in general a very poor way to solve that problem from a
numerical analyst's standpoint, since it is well known that (especially in
single precision) that formula can be very imprecise. Since I did not have
a better mousetrap to offer at the time, I kept quiet.
Now, after some analysis of the problem, here is a better mousetrap, which
will also catch a variety of related wildlife. Nick Cox and I wrote
'statsmat', which will generate descriptive stats (including some that
summarize won't). A long time ago (Stata 5 days) Nick wrote 'movsumm' which
would compute one of summarize's statistics in a moving-window context--but
not in a panel. It came to my mind that some bashing of these two elements
of code might bring forth something useful. And voila...
'MVSUMM': module to generate moving-window descriptive statistics in time
series or panel
mvsumm computes a moving-window descriptive statistic for tsvar
which must be a time series variable under the aegis of tsset.
If a panel calendar is in effect, the statistic is calculated for
each time series within the panel. The moving-window statistic
is placed in a new variable, specified with the generate()
option. The statistics available include minimum, maximum, other
key percentiles, mean and standard deviation: one of these
and/or other statistics returned by {cmd:summarize}, or easily
computable from what it returns, may be specified. aweights or
fweights may be specified. Although mvsumm works with unbalanced
panels (where the start and/or end points differ across units),
it does not allow gaps within the observations of a time series;
that is, the value of an observation for a given period may be
missing, but the observation itself must be defined. Gaps in time
series may be dealt with via the tsfill command.
Distribution-Date: 20020723
Author: Nicholas J. Cox, University of Durham
Support: email N.J.Cox@durham.ac.uk
Author: Christopher F Baum, Boston College
Support: email baum@bc.edu
Available from SSC via ssc install mvsumm. We thank Nick Winter for
suggestions and assistance finding the bugs in an earlier version.
