t Copula Fit

I am generating 100 t copula radom numbers with Rho=0.4 and Nu =5, using the copularnd function. Then I fit these random numbers back to a t copula. I do this 1000 times and I expect to see the distribution of Nu to be centered around 5. However, the estimated value of Nu takes very high values in many occasions. Any reasons?

Answer 1

[Andrew Newell]

That's just a consequence of the small sample size, and consistent with the 95% confidence estimate ( nuci ) you get for nuhat if you use the following commands:

rho = 4.5; nu = 5; N=100;
U = copularnd('t',0.4,5,N);
[rhohat,nuhat,nuci] = copulafit('t',U)

If you use N=1000, you'll see much tighter bounds on nuhat.