做滚动GARCH时遇到的一件怪事

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

一共有1612个数据，见附件。
当使用全部数据时，是可以得出结果的，而且结果比较合理。代码如下，用的rugarch包
aa=read.csv(file="dd.csv",header=T)
spec=ugarchspec(variance.model = list(model="fGARCH",submodel="GARCH",garchOrder = c(1, 1),variance.targeting=0),
mean.model = list(armaOrder = c(1, 1),include.mean = F),distribution.model="norm")
shfit=ugarchfit(spec,aa[,2])###产生的aa有两列，第二列是数据####
shfit

部分结果如下
*---------------------------------*
*          GARCH Model Fit        *
*---------------------------------*

Conditional Variance Dynamics   
-----------------------------------
GARCH Model     : fGARCH(1,1)
fGARCH Sub-Model        : GARCH
Mean Model      : ARFIMA(1,0,1)
Distribution    : norm

Optimal Parameters
------------------------------------
        Estimate  Std. Error  t value Pr(>|t|)
ar1    -0.858165    0.111992  -7.6628        0
ma1     0.878201    0.103933   8.4497        0
alpha1  0.027117    0.003523   7.6968        0
beta1   0.971883    0.003608 269.3639        0
omega   0.000000          NA       NA       NA

Robust Standard Errors:
        Estimate  Std. Error  t value Pr(>|t|)
ar1    -0.858165    0.075884 -11.3090        0
ma1     0.878201    0.072014  12.1949        0
alpha1  0.027117    0.004604   5.8899        0
beta1   0.971883    0.004127 235.5091        0
omega   0.000000          NA       NA       NA

LogLikelihood : 4344.185

但是，我去掉第一个数之后就结果不收敛了，代码如下
spec=ugarchspec(variance.model = list(model="fGARCH",submodel="GARCH",garchOrder = c(1, 1),variance.targeting=0),
mean.model = list(armaOrder = c(1, 1),include.mean = F),distribution.model="norm")
shfit=ugarchfit(spec,aa[-1,2])###产生的aa有两列，第二列是数据####
shfit

跑出来的结果如下
Warning message:
In .fgarchfit(spec = spec, data = data, out.sample = out.sample,  :
ugarchfit-->warning: solver failer to converge.


就是第一个数去掉之后就跑不出来了，当然，我也试过将默认算法“solnp”修改为“hybrid”，则可以运算。但是因为我是做滚动GARCH，将算法修改之后又出现新的问题是若干个问题如下：
1: In arima(data, order = c(modelinc[2], 0, modelinc[3]),  ... :
  possible convergence problem: optim gave code = 1
2: In arima(data, order = c(modelinc[2], 0, modelinc[3]),  ... :
  possible convergence problem: optim gave code = 1
3: In arima(data, order = c(modelinc[2], 0, modelinc[3]),  ... :
  possible convergence problem: optim gave code = 1
4: In arima(data, order = c(modelinc[2], 0, modelinc[3]),  ... :
  possible convergence problem: optim gave code = 1
5: In arima(data, order = c(modelinc[2], 0, modelinc[3]),  ... :
  possible convergence problem: optim gave code = 1
6: In arima(data, order = c(modelinc[2], 0, modelinc[3]),  ... :
  possible convergence problem: optim gave code = 1
7: In arima(data, order = c(modelinc[2], 0, modelinc[3]),  ... :
  possible convergence problem: optim gave code = 1
8: In arima(data, order = c(modelinc[2], 0, modelinc[3]),  ... :
  possible convergence problem: optim gave code = 1
9: In arima(data, order = c(modelinc[2], 0, modelinc[3]),  ... :
  possible convergence problem: optim gave code = 1
10: In arima(data, order = c(modelinc[2], 0, modelinc[3]),  ... :
  possible convergence problem: optim gave code = 1
11: In arima(data, order = c(modelinc[2], 0, modelinc[3]),  ... :
  possible convergence problem: optim gave code = 1


综上，我想请教各位大神：
1.为何全部数据能跑，去掉第一个数据就出现问题。。。
2.怎么解决ARMA的收敛问题。。。。

ps：写文章真苦逼，乍看简单，实际操作总是有意想不到的困难。祝大家国庆长假快乐。
dd1.xls (111 KB)

2017-10-3 16:53:07 上传

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0 回帖

关键词：GARCH ARCH RCH ARC distribution

由于上传不了CSV格式的，我只能转成Excel格式上传

作者您好，请问这个问题您解决了吗