R help - Apr 2013 - lognormal sampleing using covariance matrix

Dear All,
 
wondering if someine can access the link to the randsamp code referenced in the
R-help archive here:
http://www.mail-archive.com/r-help@stat.math.ethz.ch/msg75645.html ? I have
tried but for whatever reason I can not get trough. My problem seems to
be similar to what the author originally described there, so having access to it
would be great. Else, if you have any thougths on sampling from a distribution
assumed to be log-normal with expected means + SD of
 
a=5.04  + 0.56
b=1.89 + 0.19
c=0.43  + 0.08
d=0.44  + 0.07
 
and covariance matrix of:
 
       a                    b                       c                    d
a   0.31
b   0.015              0.04           
c  -0.019           0.000007            0.0064   
d  0.0069           -0.0069             -0.000067         0.0048      
 
would be greatly appreciated. Can this be done with a function?I been working on
this for months now but can not crack it. I can not find a package that has a
function to deal with this. thanks,
 
Andras
	[[alternative HTML version deleted]]

Andras,
  I am no expert but you could start with correlated normal random variables and
then convert to lognormal.

You can use rmvsnorm from the {fCopulae} package to generate the correlated
normals.

Here is a call that I have used in the past:

  set.seed(12345)
  numSims<-10 # number of multivariate simulations
  nVars<-3 # number of variables per simulation
  sd<-.3 # Standard deviation
  corrMatrix<-matrix(c(1,.2, .1, .2, 1, .2, .1, .2, 1), nrow = 3)  #
Correlation matrix
  sims<-rmvsnorm(numSims, dim=nVars, Omega=corrMatrix*sd^2)
  sims

               1            2           3
  1   0.17565865  0.000961992  0.23987213
  2   0.21283981  0.576746426  0.54851754
  3  -0.03279099  0.102383699 -0.17187533
  4  -0.13604915  0.125701955 -0.44056873
  5   0.18176624 -0.184257203 -0.49189846
  6  -0.54538679  0.131041175  0.52003437
  7   0.18902957 -0.222728926 -0.17124538
  8  -0.08285523 -0.114034555  0.15543724
  9  -0.08524792  0.312371313  0.23281669
  10 -0.27579660  0.032647156 -0.05873216
  attr(,"parameters")
  attr(,"parameters")$xi
  [1] 0 0 0

  attr(,"parameters")$Omega
        [,1]  [,2]  [,3]
  [1,] 0.090 0.018 0.009
  [2,] 0.018 0.090 0.018
  [3,] 0.009 0.018 0.090

  attr(,"parameters")$alpha
  [1] 0 0 0

Note that in my example above the mean is 0 but can be changed using the
"mu" parameter. See the docs for fCopulae.

This should help you convert the normal random variables to lognormal:
   http://en.wikipedia.org/wiki/Log-normal_distribution

Good luck,
KW

> Dear All,
>  
> wondering if someine can access the link to the randsamp code referenced in
the R-help archive here: http://www.mail-archive.com/r-help at
stat.math.ethz.ch/msg75645.html ? I have tried but for whatever reason I can not
get trough. My problem seems to be similar to what the author originally
described there, so having access to it would be great. Else, if you have any
thougths on sampling from a distribution assumed to be log-normal with expected
means + SD of
>  
> a=5.04  + 0.56
> b=1.89 + 0.19
> c=0.43  + 0.08
> d=0.44  + 0.07
>  
> and covariance matrix of:
>  
>        a                    b                       c                    d
> a   0.31
> b   0.015              0.04           
> c  -0.019           0.000007            0.0064   
> d  0.0069           -0.0069             -0.000067         0.0048      
>  
> would be greatly appreciated. Can this be done with a function?I been
working on this for months now but can not crack it. I can not find a package
that has a function to deal with this. thanks,
>  
> Andras
--