The vita distribution consists of mostly joe copulas. The first two vine trees:
We found some thresholds were the default nt approach seemed to perform poorly.
The simulation design crossed two sample sizes \(n=500,1000\) with two thresholds \(k=7, 10\). In each condition 500 discrete samples were generated. From each discrete sample, 1000 bootstrap samples were drawn.
Summaries across parameters, with absolute RB in the second table:
## # A tibble: 4 x 4
## # Groups: n [2]
## n k rbboot rbdefault
## <dbl> <fct> <dbl> <dbl>
## 1 500 7 -0.465 -12.0
## 2 500 10 -0.0575 -14.9
## 3 1000 7 -2.79 -12.3
## 4 1000 10 -1.68 -14.4
## # A tibble: 4 x 4
## # Groups: n [2]
## n k ABSrbboot ABSrbdefault
## <dbl> <fct> <dbl> <dbl>
## 1 500 7 3.11 12.0
## 2 500 10 2.83 14.9
## 3 1000 7 3.11 12.3
## 4 1000 10 3.40 14.4
Overall coverage
## # A tibble: 4 x 2
## method `mean(covered)`
## <fct> <dbl>
## 1 boot.percentile 0.848
## 2 default 0.794
## 3 boot.z 0.848
## 4 boot.bc 0.841
And further subdivided
## # A tibble: 16 x 4
## # Groups: n, k [4]
## n k method `mean(covered)`
## <dbl> <dbl> <fct> <dbl>
## 1 500 7 boot.percentile 0.848
## 2 500 7 default 0.793
## 3 500 7 boot.z 0.849
## 4 500 7 boot.bc 0.835
## 5 500 10 boot.percentile 0.92
## 6 500 10 default 0.867
## 7 500 10 boot.z 0.920
## 8 500 10 boot.bc 0.917
## 9 1000 7 boot.percentile 0.735
## 10 1000 7 default 0.682
## 11 1000 7 boot.z 0.734
## 12 1000 7 boot.bc 0.725
## 13 1000 10 boot.percentile 0.889
## 14 1000 10 default 0.836
## 15 1000 10 boot.z 0.888
## 16 1000 10 boot.bc 0.887
Plots for each parameter: