I have three random variables X, Z, and Y, and I know their distribution functions F(X), F(Z), and F(Y). However, the dependence structure among them is not known. I want to calculate the conditional distributions F(Y|X) and F(X|Y) given X, Z, and Y.
I'm using the copula and gamlss.dist packages in R to generate a Gumbel copula with theta=2 and create a multivariate distribution (mvdc). The code snippet below generates a specific probability using the generated copula:
pacman::p_load(copula, gamlss.dist)
# Generate Gumbel copula with theta=2
thetaVal <- 2
copula <- gumbelCopula(thetaVal, dim = 3)
copPre <- mvdc(copula, c("SHASH", "norm", "norm"),
list(
list(mu = 46, sigma = 5, nu = 2, tau = 1), # X SHASH
list(mean = 46, sd = 1), # Y normal
list(mean = 23, sd = 1))) # Z normal
pMvdc(c(47, 45, 22), copPre)
The last line of code gives the probability that (X ≤ 47),(Y ≤ 45), and (Z ≤ 22).
I would like guidance on calculating the conditional distributions F(Y|X) and F(X|Y) using the provided values for X, Z, and Y and the copula model.
Thanks in advance for your assistance!
To get a conditional probability like Pr(X <= x | Y <= y), you can use the formula Pr(X <= x, Y <= y) / Pr(Y <= y):
library(copula)
library(gamlss.dist)
# Generate Gumbel copula with theta=2
thetaVal <- 2
copula <- gumbelCopula(thetaVal, dim = 3)
copPre <- mvdc(copula, c("SHASH", "norm", "norm"),
list(
list(mu = 46, sigma = 5, nu = 2, tau = 1), # X SHASH
list(mean = 46, sd = 1), # Y normal
list(mean = 23, sd = 1))) # Z normal
pMvdc(c(47, 45, 22), copPre)
# Pr(X <= x | Y <= 45)
x <- 50
pMvdc(c(x, 45, Inf), copPre) / pnorm(45, 46, 1)
# or:
pMvdc(c(x, 45, Inf), copPre) / pMvdc(c(Inf, 45, Inf), copPre)
For Pr(X <= x | Y = y) you have to use the cCopula function, which is, to me, a bit complicated to understand.
# Pr(X <= x | Y = y)
x <- 50
y <- 47
cCopula(
cbind(pnorm(y, 46, 1), pSHASH(x, 46, 5, 2, 1)),
indices = 2,
copula = copula
)