If you have GARCH toolbox, this is the simplest way :

[NL,NC]=size(data);
for n=1:NC
Return(:,n) = log(data(2:NL,n)./data(1:NL-1,n))*100;
end

[coeff,errors,LLF,innovations,sigmas] = garchfit(Return);

% Then, for 5 periods forecast horizon

[sigmaForecast,meanForecast] = garchpred(coeff,return,5);

For other UGARCH structure, you can use garchset and garchget fonctions.

For MGARCH you can use ucsd_garch toolbox :

[此贴子已经被作者于2007-2-14 4:10:38编辑过]

一般用包含截矩项的AR(1)刻画的比较多，其实主要还是要把截矩项加上去

不过GARCH模型在金融里面的地位已经逐步被SV模型取代了，只不过后者目前估计还比较困难