The RATS Software Forum

Hi, everyone,
When i run bivariate GARCH with t distritution with maximize method, i will use command
logtdensity (U,V,nu)
where 'nu' stand for degrees of freedom, how to calculate it? If
nu=no. of observations-no. of parameters to be estimated
as we know, when the 'nu>100', the t distribution converges to normal distribution. In my case, there are large number of observations, see 10000, the 'nu' would be more than 9000. But i noticed that a paper with the same model as mine, and his degrees of freedom is just 7, though his no. of observations is more than 3000. Does it mean i use the wrong way to calculate 'nu'?
Looking forward to getting answer from you experts!

Appreciate any help!

That's the shape of the t-distribution for the residuals, which allows for fatter tails than a Normal. (A Normal would be a special case with nu=infinity). If you use the option DISTRIB=T, the value will be estimated. In most applications, it tends to be between 4 and 10. (It can't be less than 2, and anything above around 20 is almost indistinguishable from a Normal).

TomDoan wrote:That's the shape of the t-distribution for the residuals, which allows for fatter tails than a Normal. (A Normal would be a special case with nu=infinity). If you use the option DISTRIB=T, the value will be estimated. In most applications, it tends to be between 4 and 10. (It can't be less than 2, and anything above around 20 is almost indistinguishable from a Normal).

Thanks for ur help. Yes, i can use Sharp to set the degrees of freedom if i use GARCH command, but what if i use Maximize function (since Maximize is more appropriate in estimating multi-variate GARCH to some extend)? how to estimate the degrees of freedom(nu) for Logtdensity(U,V,nu)?

Thanks very much.

Just add NU to your parameter set. Make sure you give it a non-zero initial value. (nu=10 tends to work fairly well).