GARCH significance, what is the precise number?

2 visualizaciones (últimos 30 días)
Yoshiko Kawashima
Yoshiko Kawashima el 16 de Nov. de 2011
Comentada: amine guiri el 9 de Jul. de 2015
When performing GARCH, EGARCH and JRR Matlab produces : Value-Standard Error and t-stat. how do we find out if the t-stat is significant e.g. should it be very big or small and what is the precise number for 5% significance.I exactly need to know how they call something significance?

Respuesta aceptada

Wayne King
Wayne King el 16 de Nov. de 2011
Hi, You have to evaluate the t-statistic on the appropriate t-distribution. The t-distribution is characterized by one parameter, the degrees of freedom. So for example, in the following GARCH model, the degrees of freedom are 8.
model = garch('Constant',0.0001,'Garch',0.75,...
'Arch',0.1,'Offset',0.5,'Distribution',struct('Name','t','DoF',8));
You can use tcdf() and tinv() to evaluate significance. If you wanted only 0.05 of the probability outside that number in a two-tailed test.
dof = 8;
tlow = tinv(0.05/2,dof);
thigh = tinv(1-(0.05/2),dof);
Gives you your upper and lower critical values. The t-distribution is symmetric around zero, so they differ only in sign.
If you observed a t-statistic of 2.5, to find the p-value
2*(1-tcdf(2.5,dof))
The above is for a two-tailed test of significance.

Más respuestas (1)

Wayne King
Wayne King el 16 de Nov. de 2011
Sorry, I may have misinterpreted your original question. If you are talking about the output of garchfit:
Parameter Value Error T Statistic
----------- ----------- ------------ -----------
C -0.0013526 0.0013368 -1.0119
K 0.00027647 0.00014944 1.8500
GARCH(1) 0.77871 0.10106 7.7057
ARCH(1) 0.073557 0.028595 2.5724
Then the T-statistic is Value/Error, so how many standard errors is the Value away from the null value of zero. You can use a t-statistic of 2 as an indication of 0.05 significance in this instance.
  2 comentarios
Yoshiko Kawashima
Yoshiko Kawashima el 16 de Nov. de 2011
right that is what I meant. So if the T Statistics is bigger than 2 then it is significant. is that correct?
amine guiri
amine guiri el 9 de Jul. de 2015
I HAVE THE SAME QUESTION , it's is significant if it's bigger than 2 FOR 0.05 ?

Iniciar sesión para comentar.

Categorías

Más información sobre Conditional Variance Models en Help Center y File Exchange.

Etiquetas

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by