[求助]garch模型的选择标准

ojsapp: You can first ignore heteroscedasticity and work only on mean equation and determine the optimal lags in ARIMA by AIC, for instance. then based on your "optimal" specification of mean equation, proceed to work with the GARCH part. Note that in practice GARCH(1,1) has been adequate for many processes. If you read some journal articles in, say, J Applied of Econometrics, J of Econometrics, and the like, people do not think about GARCH(p,q), where p>3, q>=2. For concreteness, if you end up with a GARCH(3,3), that may be . This can be seen as a intuition. You can choose the one with highest maximum likelihood as your favourite specification for conditional variance models.

[此贴子已经被作者于2006-4-23 18:31:54编辑过]

hgy8857: I can not agree with you, regarding GARCH, poeple do not rely on R-square for mdoel selection

-------------increasing number of volatility models, either univariate or multivariate, have been developed such that model selection turns out to be imperative for practitioners. Model comparison can be done from different viewing angles such as highest likelihood method, model parsimony, AIC or BIC, model's forecasting performance, misspecification tests of the models, etc.(see--------------------

To zhaosweden:

Could you give a complete time-series example of ARIMA and GARCH estimation? My email is ojsapp@yahoo.com.cn.

Thanks.

谢谢楼上的帮助，麻烦系统解释一下！

楼上的仁兄们，可以帮忙拿中文解释一下吗？

先谢谢了！

我也想在此处提个问题，请给位高手帮忙解答！ THANKS IN ADVANCE！

我在选择模型的时候，选择了EGARCH-m 模型，结果得出的结果是

Coefficient Std. Error z-Statistic Prob.

LOG(GARCH) 0.001225 0.000437 2.800177 0.0051
R_SEVEN(-1) 0.872264 0.048997 17.80248 0.0000
R_SEVEN(-2) -0.891094 0.048811 -18.25613 0.0000
MA(1) -0.867198 0.023645 -36.67505 0.0000
MA(2) 0.890844 0.019611 45.42680 0.0000

Variance Equation

C(6) -0.904941 0.529500 -1.709048 0.0874
C(7) 0.782494 0.327633 2.388326 0.0169
C(8) 0.927965 0.073690 12.59289 0.0000

GED PARAMETER 1.098847 0.233156 4.712924 0.0000

R-squared -0.007741 Mean dependent var -0.002650
Adjusted R-squared -0.115233 S.D. dependent var 0.094873
S.E. of regression 0.100190 Akaike info criterion -2.879337
Sum squared resid 0.752852 Schwarz criterion -2.618892
Log likelihood 129.9321 Hannan-Quinn criter. -2.774640
Durbin-Watson stat 2.260319
其间R2是负值，而logL 统计量和AIC SIC统计都是最优的，我想请问是怎么回事？

顺便提一句，如果单独用均值方程的话，R是0.300294。

我想是不是在garch模型里面所用的是极大似然估计，只需看似然值就可以了？

can not see the leverage effect in ur model

杠杆效应项系数不显著，故然去掉了。

有没有杠杆效应在模型中跟R2是负值有关系吗？我不是很懂，还请明示！

谢谢！

[em17][em17]