[R-SIG-Finance] fCopula and mnormt

[Joe W. Byers]

>
>

I have found several technical inconsistencies with covariance matrices and
correlation matrices parameters in the functions and documentations.

for example rcopulae.gauss
> print(rcopula.gauss)
function (n, Sigma = equicorr(d, rho), d = 2, rho = 0.7) 
{
    d <- dim(Sigma)[1]
    if (sum(diag(Sigma)) != d) 
        stop("Sigma should be correlation matrix")
    mnorm <- rmnorm(n, Sigma = Sigma)
    matrix(pnorm(mnorm), ncol = d)
}

has sigma as a parameter but wants a correlation matrix.  rmnorm in mnormt is
the same:
> print(rmnorm)
function (n, Sigma = equicorr(d, rho), mu = rep(0, d), d = 2, 
    rho = 0.7) 
{
    d <- dim(Sigma)[1]
    A <- t(chol(Sigma))
    X <- matrix(rnorm(n * d), nrow = n, ncol = d)
    mu.matrix <- matrix(mu, nrow = n, ncol = d, byrow = TRUE)
    return(t(A %*% t(X)) + mu.matrix)
}

This inconsistency is really confusing and has caused me stress in determining
if I am actually providing the correct parameter to the function.

rmvnorm in mvtnorm clearly requires a covariance parameter when the default
method is used.
function (n, mean = rep(0, nrow(sigma)), sigma = diag(length(mean)), 
    method = c("svd", "chol")) 
{
    if (nrow(sigma) != ncol(sigma)) {
        stop("sigma must be a square matrix")
    }
    if (length(mean) != nrow(sigma)) {
        stop("mean and sigma have non-conforming size")
    }
    method <- match.arg(method)
    if (method == "svd") {
        ev <- eigen(sigma, sym = TRUE)$values
        if (!all(ev >= -sqrt(.Machine$double.eps) * abs(ev[1]))) {
            warning("sigma is numerically not positive definite")
        }
        sigsvd <- svd(sigma)
        retval <- t(sigsvd$v %*% (t(sigsvd$u) * sqrt(sigsvd$d)))
    }
    if (method == "chol") {
        retval <- chol(sigma, pivot = T)
        o <- order(attr(retval, "pivot"))
        retval <- retval[, o]
    }
    retval <- matrix(rnorm(n * ncol(sigma)), nrow = n) %*% retval
    retval <- sweep(retval, 2, mean, "+")
    retval
}
<environment: namespace:mvtnorm>

This function is much cleaner, easier to understand thus less confusing because
it differentiates between two generating methods.  I only wanted to point this
because package:mnormt is a dependency of many of the Rmetrics packages.  I also
may be really off base with this comment, but I was taught to use clear and
consistent parameter names as much as possible.

If this is determined a fix that needs to be done, I would love to assist with
any modifications of the code or documents.

Thank you
Joe