copBasic (version 2.1.4)

densityCOP: Density of a Copula

Description

Numerically estimate the copula density for a sequence of \(u\) and \(v\) probabilities for which each sequence has equal steps that are equal to \(\Delta(uv)\). The density \(c(u,v)\) of a copula \(\mathbf{C}(u,v)\) is numerically estimated by $$c(u,v) = [\mathbf{C}(u_2,v_2) - \mathbf{C}(u_2,v_1) - \mathbf{C}(u_1,v_2) + \mathbf{C}(u_1,v_1)]/[\Delta(uv)\times\Delta(uv)]\mbox{,}$$ where \(c(u,v) \ge 0\) (see Nelsen, 2006, p. 10; densityCOPplot). The joint density can be defined by the coupla density for continuous variables and is the ratio of the joint density funcion \(f(x,y)\) for random variables \(X\) and \(Y\) to the product of the marginal densities (\(f_x(x)\) and \(f_y(y)\)): $$c\bigl(F_x(x), F_y(y)\bigr) = \frac{f(x,y)}{f_x(x)f_y(y)}\mbox{,}$$ where \(F_x(x)\) and \(F_y(y)\) are the respective cumulative distribution functions of \(X\) and \(Y\), and lastly \(u = F_x(x)\) and \(v = F_y(y)\).

Usage

densityCOP(u,v, cop=NULL, para=NULL, deluv=.Machine$double.eps^0.25,
                truncate.at.zero=TRUE, the.zero=0, sumlogs=FALSE, ...)

Arguments

u

Nonexceedance probability \(u\) in the \(X\) direction;

v

Nonexceedance probability \(v\) in the \(Y\) direction;

cop

A copula function;

para

Vector of parameters or other data structure, if needed, to pass to the copula;

deluv

The change in the sequences \(\{u, v\} \mapsto \delta, \ldots, 1-\delta; \delta = \Delta(uv)\) probabilities;

truncate.at.zero

A density must be \(c(u,v) \ge 0\), but because this computation is based on a rectangular approximation and not analytical, there exists a possibility that very small rectangles could result in numerical values in R that are less than zero. This usually can be blamed on rounding. This logical if TRUE truncates computed densities to zero, and the default assumes that the user is providing a proper copula. A FALSE value is used by the function isfuncCOP;

the.zero

The value for “the zero” where a small number might be useful for pseudo-maximum likelihood estimation using sumlogs;

sumlogs

Return the \(\sum{\log c(u,v; \Theta)}\) where \(\Theta\) are the parameters in para; and

...

Additional arguments to pass to the copula function.

Value

Value(s) for \(c(u,v)\) are returned.

References

Joe, H., 2014, Dependence modeling with copulas: Boca Raton, CRC Press, 462 p.

Nelsen, R.B., 2006, An introduction to copulas: New York, Springer, 269 p.

See Also

simCOP, densityCOPplot, kullCOP

Examples

# NOT RUN {
# Joe (2014, p. 164) shows the closed form copula density for the Plackett.
"dPLACKETTcop" <- function(u,v,para) {
   uv <- u*v; upv <- u + v; eta <- para - 1
   A <- para*(1+eta*(upv - 2*uv)); B <- ((1+eta*upv)^2 - 4*para*eta*uv)^(3/2)
   return(A/B)
}
dPLACKETTcop(0.32, 0.74,            para=1.3) # 0.9557124
densityCOP(  0.32, 0.74, cop=PLcop, para=1.3) # 0.9557153

# }
# NOT RUN {
# Joe (2014, p. 165) shows the corner densities of the Plackett as Theta.
# copBasic uses numerical density estimation and not analytical formula.
eps <- .Machine$double.eps
densityCOP(0,0, cop=PLcop, para=4) # 3.997073  (default eps^0.25)
densityCOP(1,1, cop=PLcop, para=4) # 3.997073  (default eps^0.25)
densityCOP(1,1, cop=PLcop, para=4, deluv=eps)     # 0 (silent failure)
densityCOP(1,1, cop=PLcop, para=4, deluv=eps^0.5) # 4.5
densityCOP(1,1, cop=PLcop, para=4, deluv=eps^0.4) # 4.002069
densityCOP(1,1, cop=PLcop, para=4, deluv=eps^0.3) # 3.999513
# So we see that the small slicing does have an effect, the default of 0.25 is
# intented for general application by being away enough from machine limits. 
# }
# NOT RUN {
# }
# NOT RUN {
# Joe (2014, p. 170) shows the closed form copula density for "Bivariate Joe/B5" copula
"dJOEB5cop" <- function(u,v,para) {
   up <- (1-u)^para; vp <- (1-v)^para; eta <- para - 1
   A <- (up + vp - up*vp); B <- (1-u)^eta * (1-v)^eta; C <- (eta + A)
   return(A^(-2 + 1/para) * B * C)
}
densityCOP(0.32, 0.74, cop=JOcopB5, para=1.3)  # 0.9410653
dJOEB5cop( 0.32, 0.74, para=1.3)               # 0.9410973 
# }