0

my data set compromises 7 monthly indices spanning over 20 years. Monthly returns of each index is stored under variable "returns".

I would like to model the return on each asset class separately following a t-distribution, and then impose a dependence structure by using a t copula, which is supposed to give me a multivariate t-distribution. In addition, I would like to model the return process of each asset class using a normal distribution and then impose the dependence structure using a Gaussian copula. This in turn is supposed to give me a multivariate normal distribution. The two approaches are conducted in order calculate CVaR and to see which approach has a higher risk estimate.

Does anyone know how to "combine" a t-distribtution with a t copula in matlab??

Any help is much appreciated.

Best

Dan
  • 45,079
  • 17
  • 88
  • 157
Peter Miller
  • 1
  • Did you try the [`copulafit`](http://www.mathworks.com/help/stats/copulafit.html?refresh=true) function with the `'t'` parameter? – Dan Sep 15 '14 at 08:00
  • What are you actually looking for? `copulafit` gives you parameters of the distribution... so maybe you want to try[`copulapdf`](http://www.mathworks.com/help/stats/copulapdf.html) or one of those functions? – Dan Sep 15 '14 at 09:42
  • What I would like to do is to specify the marginals with a t-distribution for each of my asset classes. I used `fitdist('t')`, but that doesn't seem to make sense, since it gives me the tlocationscale. After that, I would like to impose the dependence structure with a t copula using `copulafit` and `copularnd` as you already pointed out. BUT I struggle with fitting a t-distribution to my asset classes. – Peter Miller Sep 15 '14 at 10:04
  • That's what `copulafit` does...? – Dan Sep 15 '14 at 10:07
  • `Copulafit` is for the t-copula. I want to fit a t-distribution first – Peter Miller Sep 15 '14 at 11:22

0 Answers0