Simulating multivariate normal distribution with mean of e(B) and variance/covariance e(V)

[Matt Manierre]

Hi everyone,

I am working on implementing a parametric bootstrap simulation where I have B, a vector with several parameter estimates from a regression model, and e(V), a variance/covariance matrix. The book I'm using says that "each entry in the simulated sample is a random draw from a multivariate normal distribution with mean [e(b)] and variance/covariance [e(V)]." I've subbed in stata matrix names where appropriate.

How would I implement this type of analysis in stata? The bootstrap command seems to be inappropriate in this scenario since it is non-parametric.

[Clyde Schechter]

Maybe you want to use the -drawnorm- command. See the help file for details.This will enable you to create a data set containing synthetic data with the desired distribution.

[Scott Merryman]

And for completeness, to convert from a covariance matrix to a correlation matrix:

. mata
------------------------------------------------- mata (type end to exit) --
:  cov = (1, 1, 8.1 \ 1, 16, 18 \ 8.1 ,18, 81)

: cov
[symmetric]
         1     2     3
    +-------------------+
  1 |    1              |
  2 |    1    16        |
  3 |  8.1    18    81  |
    +-------------------+

: corr =invsym(sqrt(diag(cov)))*cov*invsym(sqrt(diag(cov)))

: corr
[symmetric]
         1     2     3
    +-------------------+
  1 |    1              |
  2 |  .25     1        |
  3 |   .9    .5     1  |
    +-------------------+

: end

[Matt Manierre]

Scott,

That's it!! I must have been searching the wrong terms because that is a dead ringer for what I need. Thank you!