--- [email protected] wrote:
> I have a time series with 40 obs. Based on the underlying process of the
> series I would like to generate 5000 (for example) observations, so then I
> can generate a robust (in the sense of being based on many observations)
> concrete probability distribution . In order to do that what I though I could
> do is:
>
> - First, distribution testing of my original series, and then monte carlo
> simulation.
>
> Question: is there any command that allows me to test which is the best
> distribution that fits my data, or do I need to test this distribution by
> distribution (e.g. normality tests, chi-square...and so on).
>
> - Is there any other command that allows me to generate observations (based
> on the underlysing process of a time seies)?
> (e.g. Can bsample, or bootstrap be of any help?)
I think what you are proposing has already been invented before
(this happens to most of my genius inventions too): it sounds a
lot like parametric bootstrap. A basic introduction in this area
can be found in section 6.5 in (Efron and Tibshirani 1993).
With time series data finding the distribution from which to
draw your observations is probably the least of your problems.
A more serious problem with time-series models is serial
correlation. Efron and Tibshirani (1993) give in chapter 8 some
pointers on ways of doing the bootstrap with time series data.
Bradley Efron and Robert J. Tibshirani (1993) "An introduction
to the bootstrap". Boca Raton: Chapman & Hall/CRC.
Hope this helps,
Maarten
-----------------------------------------
Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands
visiting address:
Buitenveldertselaan 3 (Metropolitan), room Z434
+31 20 5986715
http://home.fsw.vu.nl/m.buis/
-----------------------------------------
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
