How to exclude bilateral tail

Hello all,

I am using this code to generate random numbers and I would like to know how I can exclude the bilateral tail (extreme values). I would like to set a range of 0.05:0.95. Anyone could help me with this?

Thanks in advance,

Arthur

%MEAN
mu=[0.0021 0.46 0.307 1.57 0.235 -2.17];
%/** convert correlation matrix to covariance matrix **/
R = [1.0000	0.2008	0.1561	0.0530	0.4697	0.3865
    0.2008	1.0000	0.7806	-0.3242	0.6049	-0.0772
    0.1561	0.7806	1.0000	-0.8276	0.3765	-0.4989
    0.0530	-0.3242	-0.8276	1.0000	0.0297	0.7812
    0.4697	0.6049	0.3765	0.0297	1.0000	0.5794
    0.3865	-0.0772	-0.4989	0.7812	0.5794	1.0000];

%/** standard deviations of each variable **/
c = [0.17035e-1 0.48517e-1 0.20748 0.36239 0.39731 0.20431e+1];
D = diag(c);
Sigma = D*R*D  %/** covariance matrix **/
rand('seed')  % For reproducibility
R = mvnrnd(mu,Sigma,10000)

[dpb]

This doesn't seem to make any sense --

>> sum(R(:)>0.95)/numel(R)
ans =
    0.1768
>> sum(R(:)<0.05)/numel(R)
ans =
    0.3843
>> 

A very large percentage of your obsverations overall are outside those ranges with those parameters; for some columns almost everything will be rejected...

>> sum(R>0.95)/size(R,1)
ans =
         0         0    0.0030    0.9490    0.0380    0.0710
>> sum(R<0.05)/size(R,1)
ans =
    0.9980         0    0.1100         0    0.3310    0.8670
>> 

You would end up w/ an array almost singular; certainly very ill-conditioned.

What in the world are you trying to model? (It seems a form of a bounded distribution would be more appropriate than trying to truncated an unbounded one for starters.)

[Arthur Santos]

I'm trying to generate random numbers (following a normal distribution) including the uncertainties (standard errors and correlation matrix). However, I wanted numbers excluding extreme values (tails).

[dpb]

Or do you mean the 0.95 to be 1-0.05 tails, not an absolute value of 0.95?

That would be a much different result although again the resulting distribution will not be normally distributed.

[Arthur Santos]

I would like to exclude extreme values by taking 0.05 from each tail... did you understand me?