What is copulas exactly, in probability and finance terms

AI Thread Summary
A copula is a joint probability distribution that models the correlation between random variables, utilizing uniform cumulative distribution functions (CDFs) and a covariance matrix. The discussion highlights the complexity of formulating different copulas, particularly bivariate Gaussian and t-copulas, and contrasts them with Bayesian network modeling, which can handle more variables but faces discretization challenges. The conversation also touches on simulation methods for copulas, including the use of Cholesky decomposition and R packages for implementation. Participants express interest in understanding the source code of R packages, noting that they are open source and can be accessed by unpacking tar files. Overall, copulas play a significant role in finance, as discussed in recent literature.
colstat
Messages
56
Reaction score
0
Hi, all
I have been reading about copula, but still very confused.

What exactly is a copula? My understanding is: there are couple of components
1. uniform cdf marginal
2. a covariance matrix

What exactly is this thing? Why am I calculating the marginals and what does it have to do with the covariance matrix?

I am reading on the bivariate Gaussian copulas and t-copulas.
 
Physics news on Phys.org
Copula is just a joint probability distribution. The beauty is that copula models correlation. The downside is it is difficult to formulate different couplas. I like baysian network modelling of joint distributions. Baysian network handles more variables, not only bivariates; the downside is discretization. If you are reading the most recent Significance magazine you will find out all copula can do to the financing industry.
 
Thank you zlin034. What do you use to simulate copulas, Gaussian and student-t?

I have two ideas for bivariate Gaussian:
1. integrate the density from Wiki, here
http://en.wikipedia.org/wiki/Copula_(probability_theory)#Gaussian_copula

or this,
http://www.vosesoftware.com/ModelRi...n_ModelRisk/Copulas/Vose_Bivariate_Copula.htm

2. use Cholesky-decomposition \Sigma=A'A,
then, generate iid standard normal random variables V = (V1, V2)',
then, get Xi from A*V, for i=1,2.
then, get ui= \Phi(Xi), for i=1,2.
 
I am not using R, but even in R there's an algorithm right? Is there a way to see what they did in the package?
 
R is open source right? Please read the source code from the package
 
The compressed R packages have file extension .tar, they are called tar balls.

If you open the tar balls, you can see all sources codes are ASCII text files.
 
Back
Top