关于ltolerance选项的设置问题？

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

连老师，我在估计ARCH模型的时候，命令为arch x y, earch(1) egarch(1) ltolerance(0.0001),但是对数似然函数值之差小于我的设定0.0001时程序还在运行，这是为什么？另外，我发现arch命令运行太慢，有没有使其运行快的方法。

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏1 回帖

关键词：ltolerance Tolerance ance OLE Era 设置 选项 ltolerance

这应该不会呀，我此前用MLE估计模型时，用过这个选项，都是没有问题的。
你把运行的结果截个图给我看看。

arch x y , earch(1) egarch(1) het(z) ltolerance(0.0001)

Number of gaps in sample:  8  
(note: conditioning reset at each gap)


(setting optimization to BHHH)
Iteration 0:   log likelihood =  586.42721  
Iteration 1:   log likelihood =  597.54868  
Iteration 2:   log likelihood =  603.71421  
Iteration 3:   log likelihood =  607.30761  
Iteration 4:   log likelihood =   609.5603  
(switching optimization to BFGS)
Iteration 5:   log likelihood =   610.3975  
Iteration 6:   log likelihood =  610.52171  
Iteration 7:   log likelihood =  610.81827  
Iteration 8:   log likelihood =  610.89797  
Iteration 9:   log likelihood =  610.90555  
Iteration 10:  log likelihood =  610.90643  
Iteration 11:  log likelihood =  610.90911  
Iteration 12:  log likelihood =  610.90912  *** 应该在此处就应该停止了

Iteration 13:  log likelihood =  610.90922  
Iteration 14:  log likelihood =  610.90924   
(switching optimization to BHHH)
Iteration 15:  log likelihood =  610.90924  
Iteration 16:  log likelihood =  610.90925  (backed up)   
Iteration 17:  log likelihood =  610.90925  (backed up)
Iteration 18:  log likelihood =  610.90925  (backed up)
Iteration 19:  log likelihood =  610.90925  (backed up)
(switching optimization to BFGS)
Iteration 20:  log likelihood =  610.90925  (backed up)
.................

(setting optimization to BHHH)
Iteration 0:   log likelihood =  439.03055  
Iteration 1:   log likelihood =  450.28815  
Iteration 2:   log likelihood =  465.40009  
Iteration 3:   log likelihood =  467.02226  
Iteration 4:   log likelihood =  467.16146  
(switching optimization to BFGS)
Iteration 5:   log likelihood =  467.67814  
Iteration 6:   log likelihood =  467.82723  
Iteration 7:   log likelihood =   467.8609  
Iteration 8:   log likelihood =  467.89386  
Iteration 9:   log likelihood =  467.89716  
Iteration 10:  log likelihood =  467.89747  
Iteration 11:  log likelihood =  467.89962  
Iteration 12:  log likelihood =  467.89999  
Iteration 13:  log likelihood =  467.90005  
Iteration 14:  log likelihood =  467.90007  (backed up)
(switching optimization to BHHH)
Iteration 15:  log likelihood =   467.9001  
Iteration 16:  log likelihood =  467.90011  (backed up)
Iteration 17:  log likelihood =  467.90011  (backed up)
Iteration 18:  log likelihood =  467.90011  (backed up)
Iteration 19:  log likelihood =  467.90011  (backed up)
(switching optimization to BFGS)
Iteration 20:  log likelihood =  467.90011  (backed up)
Iteration 21:  log likelihood =  467.90011  (backed up)
Iteration 22:  log likelihood =  467.90011  (backed up)
Iteration 23:  log likelihood =  467.90011  (backed up)
Iteration 24:  log likelihood =  467.90011  (backed up)
Iteration 25:  log likelihood =  467.90011  (backed up)
Iteration 26:  log likelihood =  467.90011  (backed up)
Iteration 27:  log likelihood =  467.90011  (backed up)
Iteration 28:  log likelihood =  467.90011  (backed up)
Iteration 29:  log likelihood =  467.90011  (backed up)
(switching optimization to BHHH)
Iteration 30:  log likelihood =  467.90011  (backed up)
Iteration 31:  log likelihood =  467.90011  (backed up)
Iteration 32:  log likelihood =  467.90011  (backed up)
Iteration 33:  log likelihood =  467.90011  (backed up)
Iteration 34:  log likelihood =  467.90011  (backed up)
(switching optimization to BFGS)
Iteration 35:  log likelihood =  467.90011  (backed up)
Iteration 36:  log likelihood =  467.90011  (backed up)
Iteration 37:  log likelihood =  467.90011  (backed up)
Iteration 38:  log likelihood =  467.90011  (backed up)

hehe,我看了help ，可能是由于“If the optimizer becomes stuck with repeated  "(backed up)" messages, the gradient probably still contains substantial values, but an uphill direction cannot be  found for the likelihood.  
”
所以我就设定gtolerance(999)选项，但是可能不是全局最优，或者设定最大的iterate次数，如iterate(50),但是发现与设定gtolerance(999)选项的结果差异很大。这时候应该怎么办？

秋日私语 发表于 2011-7-11 13:12
arch x y , earch(1) egarch(1) het(z) ltolerance(0.0001)

Number of gaps in sample:  8  
(note: conditioning reset at each gap)


(setting optimization to BHHH)
Iteration 0:   log likelihood =  586.42721  
Iteration 1:   log likelihood =  597.54868  
Iteration 2:   log likelihood =  603.71421  
Iteration 3:   log likelihood =  607.30761  
Iteration 4:   log likelihood =   609.5603  
(switching optimization to BFGS)
Iteration 5:   log likelihood =   610.3975  
Iteration 6:   log likelihood =  610.52171  
Iteration 7:   log likelihood =  610.81827  
Iteration 8:   log likelihood =  610.89797  
Iteration 9:   log likelihood =  610.90555  
Iteration 10:  log likelihood =  610.90643  
Iteration 11:  log likelihood =  610.90911  
Iteration 12:  log likelihood =  610.90912  *** 应该在此处就应该停止了
A: 表面上看，似乎是0.00001，但实际上可能是四舍五入的结果，有可能是 0.000013。

Iteration 13:  log likelihood =  610.90922  
Iteration 14:  log likelihood =  610.90924   
(switching optimization to BHHH)
Iteration 15:  log likelihood =  610.90924  
Iteration 16:  log likelihood =  610.90925  (backed up)   
Iteration 17:  log likelihood =  610.90925  (backed up)
Iteration 18:  log likelihood =  610.90925  (backed up)
Iteration 19:  log likelihood =  610.90925  (backed up)
(switching optimization to BFGS)
Iteration 20:  log likelihood =  610.90925  (backed up)
.................

我设置的是0.0001，而不是0.00001.
另外
我就设定gtolerance(999)选项，但是可能不是全局最优，或者设定最大的iterate次数，如iterate(50),但是发现与设定gtolerance(999)选项的结果差异很大。这时候应该怎么办？

这只能作为无奈之举，因为无法保证全局最优。
可能的处理方式如下：
1. 回到源头看看模型设定是否有问题？
可以选择一个子样本，如某个时间段，测试一下模型是否收敛。
先设定最简单的 GARCH(1, 1)  模型，如果收敛，再逐步分析复杂一点的模型。
2. 看看数据本身是否有问题。
例如，是否包含了严重的离群值或录入错误的观察值。
3. 如果上述设定都没有问题，那就考虑使用一个小一点的收敛判据，比如 0.0001.

thanks. 呵呵，我试试