Estimate the correlation parameter of the (bivariate) Clayton copula distribution by maximum likelihood estimation.

biclaytoncop(lapar = "loglink", iapar = NULL, imethod = 1,
             parallel = FALSE, zero = NULL)

Arguments

lapar, iapar, imethod

Details at CommonVGAMffArguments. See Links for more link function choices.

parallel, zero

Details at CommonVGAMffArguments. If parallel = TRUE then the constraint is also applied to the intercept.

Details

The cumulative distribution function is $$P(u_1, u_2;\alpha) = (u_1^{-\alpha} + u_2^{-\alpha}-1)^{-1/\alpha}$$ for \(0 \leq \alpha \). Here, \(\alpha\) is the association parameter. The support of the function is the interior of the unit square; however, values of 0 and/or 1 are not allowed (currently). The marginal distributions are the standard uniform distributions. When \(\alpha = 0\) the random variables are independent.

This VGAM family function can handle multiple responses, for example, a six-column matrix where the first 2 columns is the first out of three responses, the next 2 columns being the next response, etc.

Value

An object of class "vglmff" (see vglmff-class). The object is used by modelling functions such as vglm and vgam.

References

Clayton, D. (1982). A model for association in bivariate survival data. Journal of the Royal Statistical Society, Series B, Methodological, 44, 414–422.

Schepsmeier, U. and Stober, J. (2014). Derivatives and Fisher information of bivariate copulas. Statistical Papers 55, 525–542.

Author

R. Feyter and T. W. Yee

Note

The response matrix must have a multiple of two-columns. Currently, the fitted value is a matrix with the same number of columns and values equal to 0.5. This is because each marginal distribution corresponds to a standard uniform distribution.

This VGAM family function is fragile; each response must be in the interior of the unit square.

Examples

ymat <- rbiclaytoncop(n = (nn <- 1000), apar = exp(2))
bdata <- data.frame(y1 = ymat[, 1], y2 = ymat[, 2],
                    y3 = ymat[, 1], y4 = ymat[, 2], x2 = runif(nn))
summary(bdata)
#>        y1                 y2                  y3                 y4           
#>  Min.   :0.000828   Min.   :0.0008981   Min.   :0.000828   Min.   :0.0008981  
#>  1st Qu.:0.254812   1st Qu.:0.2605758   1st Qu.:0.254812   1st Qu.:0.2605758  
#>  Median :0.508578   Median :0.5055172   Median :0.508578   Median :0.5055172  
#>  Mean   :0.507874   Mean   :0.5064419   Mean   :0.507874   Mean   :0.5064419  
#>  3rd Qu.:0.764826   3rd Qu.:0.7455065   3rd Qu.:0.764826   3rd Qu.:0.7455065  
#>  Max.   :0.999361   Max.   :0.9995598   Max.   :0.999361   Max.   :0.9995598  
#>        x2          
#>  Min.   :0.002145  
#>  1st Qu.:0.270375  
#>  Median :0.526698  
#>  Mean   :0.519788  
#>  3rd Qu.:0.780546  
#>  Max.   :0.999487  
if (FALSE)  plot(ymat, col = "blue")  # \dontrun{}
fit1 <-
  vglm(cbind(y1, y2, y3, y4) ~ 1,  # 2 responses, e.g., (y1,y2) is the 1st
       biclaytoncop, data = bdata,
       trace = TRUE, crit = "coef")  # Sometimes a good idea
#> Iteration 1: coefficients = 1.993075, 1.991763
#> Iteration 2: coefficients = 1.9935812, 1.9935912
#> Iteration 3: coefficients = 1.9935769, 1.9935769
#> Iteration 4: coefficients = 1.993577, 1.993577
coef(fit1, matrix = TRUE)
#>             loglink(apar1) loglink(apar2)
#> (Intercept)       1.993577       1.993577
Coef(fit1)
#>    apar1    apar2 
#> 7.341748 7.341748 
head(fitted(fit1))
#>    y1  y2  y3  y4
#> 1 0.5 0.5 0.5 0.5
#> 2 0.5 0.5 0.5 0.5
#> 3 0.5 0.5 0.5 0.5
#> 4 0.5 0.5 0.5 0.5
#> 5 0.5 0.5 0.5 0.5
#> 6 0.5 0.5 0.5 0.5
summary(fit1)
#> 
#> Call:
#> vglm(formula = cbind(y1, y2, y3, y4) ~ 1, family = biclaytoncop, 
#>     data = bdata, trace = TRUE, crit = "coef")
#> 
#> Coefficients: 
#>               Estimate Std. Error z value Pr(>|z|)    
#> (Intercept):1  1.99358    0.03099   64.34   <2e-16 ***
#> (Intercept):2  1.99358    0.03099   64.34   <2e-16 ***
#> ---
#> Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
#> 
#> Names of linear predictors: loglink(apar1), loglink(apar2)
#> 
#> Log-likelihood: 2439.828 on 1998 degrees of freedom
#> 
#> Number of Fisher scoring iterations: 4 
#> 
#> No Hauck-Donner effect found in any of the estimates
#> 

# Another example; apar is a function of x2
bdata <- transform(bdata, apar = exp(-0.5 + x2))
ymat <- rbiclaytoncop(n = nn, apar = with(bdata, apar))
bdata <- transform(bdata, y5 = ymat[, 1], y6 = ymat[, 2])
fit2 <- vgam(cbind(y5, y6) ~ s(x2), data = bdata,
             biclaytoncop(lapar = "loglink"), trace = TRUE)
#> VGAM  s.vam  loop  1 :  loglikelihood = 186.30248
#> VGAM  s.vam  loop  2 :  loglikelihood = 186.37729
#> VGAM  s.vam  loop  3 :  loglikelihood = 186.37027
#> VGAM  s.vam  loop  4 :  loglikelihood = 186.36697
#> VGAM  s.vam  loop  5 :  loglikelihood = 186.36587
#> VGAM  s.vam  loop  6 :  loglikelihood = 186.36553
#> VGAM  s.vam  loop  7 :  loglikelihood = 186.36542
#> VGAM  s.vam  loop  8 :  loglikelihood = 186.36539
#> VGAM  s.vam  loop  9 :  loglikelihood = 186.36537
#> VGAM  s.vam  loop  10 :  loglikelihood = 186.36537
if (FALSE) plot(fit2, lcol = "blue", scol = "orange", se = TRUE)  # \dontrun{}