(1) is basic description. Define variables and use -summarize-.
(2) and (3) may well require the context of some time-series model(s)
for anyone to do properly as conventional tests such as t-tests _only_
work properly if you can assume serial independence, which on the face
of it is unlikely for such data. Switching to non-parametric tests or
bootstrapping does not solve that problem. There are bootstrapping
fudges designed for time series but they are not really supported in
Stata, as I understand it.
There are time series experts on the list who may want to advise
further. I suspect they would advise a different approach in which you
tried to identify an appropriate model or models for the entire process.
Generating tables and tables of test statistics seems likely to prove
very frustrating without some larger framework, quite apart from the
problem I have already raised that the standard errors are suspect and
the P-values garbage.
Much the same issues arose repeatedly in a thread started recently by
Mahmoud Abd-El-Aal. I and others made similar points there.
As happens so often on the list, I suspect that you would be better off
approaching people in your institution who have expertise in this field.
If you have an adviser or supervisor, that person is surely the person
to ask.
Nick
[email protected]
Wen Xia Ge
My panel data is composed of bond issues of thousands of firms over more
than 10 years. My goal is:
(1) Get the means (medians) of several firm-level variables (V1 to V5)
prior to bond issues and post bond issues, i.e., to get the means
(medians) of V1 to V5 for years -2 to +2 relative to bond issues, and
then
(2) test whether they are significantly different from zero or not in
each year,
(3) test if they change significantly from t-1 to t, from t to t+1.
I would like to present the results in the following format:
Year -2 -1 0 1
2
V1 mean mean mean mean
mean
(t-statistics) (t-statistics) (t-statistics) (t-statistics)
(t-statistics)
V2 mean mean mean mean
mean
(t-statistics) (t-statistics) (t-statistics) (t-statistics)
(t-statistics)
V3 mean mean mean mean
mean
(t-statistics) (t-statistics) (t-statistics) (t-statistics)
(t-statistics)
V4 mean mean mean mean
mean
(t-statistics) (t-statistics) (t-statistics) (t-statistics)
(t-statistics)
V5 mean mean mean mean
mean
(t-statistics) (t-statistics) (t-statistics) (t-statistics)
(t-statistics)
Nick Cox
I don't think you give us any clue on what your goal is, precisely. To
look at changes? That seems to be what all studies of panel data should
do.
To get a better answer than this, you may need to ask a better question.
Otherwise I am reduced to anodyne advice, such as plot your data and
think about building a model.
Wen Xia Ge
My panel data is composed of bond issues of thousands of firms over more
than 10 years. I want to look at the changes of several firm-level
variables prior to bond issues and post bond issues, that is, I want to
look at the changes of several firm-level variables in year t-2, t-1, t,
t+1 and t+2, where bond issues are in year t.
I know we can generate lagged variables (one year lag or two year lag)
in STATA, but I have no clue how to achieve the above goal. Any
suggestions?
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
