Add an itau argument to BiCopSelect, RVineCopSelect and RVineStructureSelect

[tvatter]

To implement this, I suggest the following:

• preprocessors: add check_twoparams
• BiCopSelect, RVineCopSelect, RVineStructureSelect: add a method argument (default method = "mle”) => only affects the one-parameter families… taken care of by adding check_twoparams to the preproc
• fit.ACopula: add a method argument (default method = “mle”)
• fit.FirstTreeCopulas and fit.TreeCopulas: add a method argument (default method = “mle”)

You can compare the speed of using itau instead of mle using the examples from RVineCopSelect/RVineStructureSelect:

system.time(RVM <- RVineStructureSelect(daxreturns, c(1,3:6)))
system.time(RVM <- RVineStructureSelect(daxreturns, c(1:6), method = "itau"))
system.time(RVM <- RVineStructureSelect(daxreturns, c(1, 3:6), method = "itau"))

system.time(RVM1 <- RVineCopSelect(simdata, familyset = c(1, 3, 4, 5 ,6), Matrix))
system.time(RVM1 <- RVineCopSelect(simdata, familyset = c(1:6), Matrix, method = "itau"))
system.time(RVM1 <- RVineCopSelect(simdata, familyset = c(1, 3, 4, 5 ,6), Matrix, method = "itau"))

Note that it generates a warning (as expected) when method = "itau" and familyset = 1:6.

As an additional suggestion, in RVineStructureSelect/fit.ACopula, it's probably not worth calling BiCopSelect and BiCopHfunc1/2 when familyset == 0.

[tnagler]

Thanks a lot!

[tnagler]

Given its importance in financial applications, I also added support for the t-copula (see fc4451b). In this case par2 is considered to be fixed with value 5.

Here are some timings on my machine:

library("VineCopula")
data("daxreturns")

system.time(RVM <- RVineStructureSelect(daxreturns, 1:6))
#>    user  system elapsed 
#> 118.232   0.000 118.091 
system.time(RVM <- RVineStructureSelect(daxreturns, 1:6, method = "itau"))
#>    user  system elapsed 
#>   3.976   0.000   3.979 

system.time(RVM1 <- RVineCopSelect(daxreturns, familyset = 1:6, RVM$Matrix))
#>    user  system elapsed 
#> 119.076   0.508 119.548 
system.time(RVM1 <- RVineCopSelect(daxreturns, familyset = 1:6, RVM$Matrix, method = "itau"))
#>    user  system elapsed 
#>   3.576   0.004   3.573 

This is awesome!

[tnagler]

I added a little more flexible by using profile likelihood for par2 instead of keeping it fixed (344fe92). New timings:

system.time(RVM <- RVineStructureSelect(daxreturns, 1:6, method = "itau"))
#>    user  system elapsed 
#>   11.756   0.112  11.875
system.time(RVM <- RVineStructureSelect(daxreturns, c(1, 3:6), method = "itau"))
#>    user  system elapsed 
#>   2.816   0.020   2.830