Become an expert in R — Interactive courses, Cheat Sheets, certificates and more!
Get Started for Free

asymCopula

Construction of copulas using Khoudraji's device


Description

Creates an object representing a copula constructed using Khoudraji's device (Khoudraji, 1995). The resulting R object is either of class "khoudrajiBivCopula", "khoudrajiExplicitCopula" or "khoudrajiCopula".

In the bivariate case, given two copulas C_1 and C_2, Khoudraji's device consists of defining a copula whose c.d.f. is given by:

C_1(u_1^{1 - a_1}, u_2^{1 - a_2}) C_2(u_1^{a_1}, u_2^{a_2})

where a_1 and a_2 are shape parameters in [0,1].

The construction principle (see also Genest et al. 1998) is a special case of that considered in Liebscher (2008).

Usage

khoudrajiCopula(copula1 = indepCopula(), copula2 = indepCopula(dim = d),
                shapes = rep(NA_real_, dim(copula1)))

Arguments

copula1, copula2

each a copula (possibly generalized, e.g., also a "rotCopula") of the same dimension d. By default independence copulas, where copula2 gets the dimension from copula1.

shapes

numeric vector of length d, with values in [0,1].

Details

If the argument copulas are bivariate, an object of class "khoudrajiBivCopula" will be constructed. If they are exchangeable and d-dimensional with d > 2, and if they have explicit p.d.f. and c.d.f. expressions, an object of class "khoudrajiExplicitCopula" will be constructed. For the latter two classes, density evaluation is implemented, and fitting and goodness-of-fit testing can be attempted. If d > 2 but one of the argument copulas does not have explicit p.d.f. and c.d.f. expressions, or is not exchangeable, an object of class "khoudrajiCopula" will be constructed, for which density evaluation is not possible.

Value

A new object of class "khoudrajiBivCopula" in dimension two or of class "khoudrajiExplicitCopula" or "khoudrajiCopula" when d > 2.

References

Genest, C., Ghoudi, K., and Rivest, L.-P. (1998), Discussion of "Understanding relationships using copulas", by Frees, E., and Valdez, E., North American Actuarial Journal 3, 143–149.

Khoudraji, A. (1995), Contributions à l'étude des copules et àla modélisation des valeurs extrêmes bivariées, PhD thesis, Université Laval, Québec, Canada.

Liebscher, E. (2008), Construction of asymmetric multivariate copulas, Journal of Multivariate Analysis 99, 2234–2250.

Examples

## A bivariate Khoudraji-Clayton copula
kc <- khoudrajiCopula(copula2 = claytonCopula(6),
                      shapes = c(0.4, 0.95))
class(kc) # "kh..._Biv_Copula"
kc
contour(kc, dCopula, nlevels = 20, main = "dCopula(<khoudrajiBivCopula>)")

## A Khoudraji-Clayton copula with second shape parameter fixed
kcf <- khoudrajiCopula(copula2 = claytonCopula(6),
                       shapes = fixParam(c(0.4, 0.95), c(FALSE, TRUE)))
kcf. <- setTheta(kcf, c(3, 0.2)) # (change *free* param's only)
validObject(kcf) & validObject(kcf.)

## A "nested" Khoudraji bivariate copula
kgkcf <- khoudrajiCopula(copula1 = gumbelCopula(3),
                         copula2 = kcf,
                         shapes = c(0.7, 0.25))
kgkcf # -> 6 parameters (1 of 6 is 'fixed')
contour(kgkcf, dCopula, nlevels = 20,
        main = "dCopula(<khoudrajiBivC.(nested)>)")

(Xtras <- copula:::doExtras()) # determine whether examples will be extra (long)
n <- if(Xtras) 300 else 64 # sample size (realistic vs short for example)

u <- rCopula(n, kc)
plot(u)

## For likelihood (or fitting), specify the "free" (non-fixed) param's:
##           C1:  C2c C2s1    sh1  sh2
loglikCopula(c(3,   6, 0.4,   0.7, 0.25),
             u = u, copula = kgkcf)

## Fitting takes time (using numerical differentiation) and may be difficult:

## Starting values are required for all parameters
f.IC <- fitCopula(khoudrajiCopula(copula2 = claytonCopula()),
                  start = c(1.1, 0.5, 0.5), data = pobs(u),
                  optim.method = "Nelder-Mead")
summary(f.IC)
confint(f.IC) # (only interesting for reasonable sample size)

## Because of time,  don't run these by default :

## Second shape parameter fixed to 0.95
kcf2 <- khoudrajiCopula(copula2 = claytonCopula(),
                        shapes = fixParam(c(NA_real_, 0.95), c(FALSE, TRUE)))
system.time(
f.ICf <- fitCopula(kcf2, start = c(1.1, 0.5), data = pobs(u),
                   optim.method = "Nelder-Mead")
) # ~ 7-8 sec
confint(f.ICf) # !
coef(f.ICf, SE=TRUE)

## With a different optimization method
system.time(
f.IC2 <- fitCopula(kcf2, start = c(1.1, 0.5), data = pobs(u),
                   optim.method = "BFGS")
)
printCoefmat(coef(f.IC2, SE=TRUE), digits = 3) # w/o unuseful extra digits


if(Xtras >= 2) { # really S..L..O..W... --------

## GOF example
optim.method <- "Nelder-Mead" #try "BFGS" as well
gofCopula(kcf2, x = u, start = c(1.1, 0.5), optim.method = optim.method)
gofCopula(kcf2, x = u, start = c(1.1, 0.5), optim.method = optim.method,
          sim = "mult")
## The goodness-of-fit tests should hold their level
## but this would need to be tested

## Another example under the alternative
u <- rCopula(n, gumbelCopula(4))
gofCopula(kcf2, x = u, start = c(1.1, 0.5), optim.method = optim.method)
gofCopula(kcf2, x = u, start = c(1.1, 0.5), optim.method = optim.method,
          sim = "mult")

}## ------ end { really slow gofC*() } --------

## Higher-dimensional constructions

## A three dimensional Khoudraji-Clayton copula
kcd3 <- khoudrajiCopula(copula1 = indepCopula(dim=3),
                        copula2 = claytonCopula(6, dim=3),
                        shapes = c(0.4, 0.95, 0.95))

n <- if(Xtras) 1000 else 100 # sample size (realistic vs short for example)
u <- rCopula(n, kcd3)
splom2(u)
v <- matrix(runif(15), 5, 3)
dCopula(v, kcd3)

## A four dimensional Khoudraji-Normal copula
knd4 <- khoudrajiCopula(copula1 = indepCopula(dim=4),
                        copula2 = normalCopula(.9, dim=4),
                        shapes = c(0.4, 0.95, 0.95, 0.95))
knd4
stopifnot(class(knd4) == "khoudrajiCopula")
u <- rCopula(n, knd4)
splom2(u)
## TODO :
## dCopula(v, knd4) ## not implemented

copula

Multivariate Dependence with Copulas

v1.0-1
GPL (>= 3) | file LICENCE
Authors
Marius Hofert [aut] (<https://orcid.org/0000-0001-8009-4665>), Ivan Kojadinovic [aut] (<https://orcid.org/0000-0002-2903-1543>), Martin Maechler [aut, cre] (<https://orcid.org/0000-0002-8685-9910>), Jun Yan [aut] (<https://orcid.org/0000-0003-4401-7296>), Johanna G. Nešlehová [ctb] (evTestK(), <https://orcid.org/0000-0001-9634-4796>), Rebecca Morger [ctb] (fitCopula.ml(): code for free mixCopula weight parameters)
Initial release
2020-12-07

We don't support your browser anymore

Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.