[求助]怎么求广义误差分布的分位数啊

在对第三只基金基金同智做回归时碰到这样的问题：

这是我改过的程序：

' Program to estimate a GARCH(1,1)-t model on the series tongzhi.

'%path = @runpath+"../tongzhi/"
'cd %tongzhi
load tongzhi
series d1 = 0
smpl @first @first
d1 = 1
smpl @all

'get starting values from Gaussian ARCH
equation eq1
eq1.arch tongzhi c
show eq1.output
'declare and innitialize parameters
coef(1) mu = eq1.c(1)
coef(1) omega = eq1.c(2)
coef(1) alpha =eq1.c(3)
coef(1) beta =eq1.c(4)
coef(1) tdf=3
!pi=@acos(-1)

'set up GARCH(1,1)-t likelihood
logl ll_l
ll_l.append @logl logl
ll_l.append res=tongzhi-mu(1)
ll_l.append sig2 =@recode(d1=1,omega(1)/(1-alpha(1)-beta(1)),omega(1)+alpha(1)*res(-1)^2+beta(1)*sig2(-1))
ll_l.append z =res^2/sig2/(tdf(1)-2)+1
ll_l.append logl = @gammalog((tdf(1)+1)/2)-@gammalog(tdf(1)/2)-log(!pi)/2-log(tdf(1)-2)/2-log(sig2)/2 - (tdf(1)+1)*log(z)/2

'estimate and display output
ll_l.ml(showopts,m=1000,c=1e-5)
show ll_l.output
scalar quantiletongzhi1=@qtdist(0.95,tdf(1))
scalar quantiletongzhi2=@qtdist(0.99,tdf(1))
这是数据：

118560.rar (173.05 KB) 本附件包括：

• 毕业论文数据 未定.xls

2007-5-20 17:49:00 上传

[求助]怎么求广义误差分布的分位数啊

程序是执行了

可是要注意的是,估计的系数都不显着.

118634.rar (628 Bytes) 本附件包括：

• tongzhi_garcht.prg

2007-5-21 07:49:00 上传

[求助]怎么求广义误差分布的分位数啊

程序并没有执行完啊，这句话什么意思啊，怎么会缺失一个值呢？

[此贴子已经被作者于2007-5-21 9:57:18编辑过]

LogL: LL_L
Method: Maximum Likelihood (Marquardt)
Date: 05/21/07 Time: 10:32
Sample: 1 199
Included observations: 199
Evaluation order: By observation
Estimation settings: tol= 1.0e-05, derivs=accurate numeric
Initial Values: MU(1)=0.00349, OMEGA(1)=0.00017, ALPHA(1)=0.1633
4, BETA(1)=0.54009, TDF(1)=3.00000
Convergence not achieved after 1000 iterations

Coefficient Std. Error z-Statistic Prob.

MU(1) 0.003982 0.001546 2.575188 0.0100
OMEGA(1) 0.000170 0.000133 1.279660 0.2007
ALPHA(1) 0.191334 0.126748 1.509568 0.1312
BETA(1) 0.530678 0.283426 1.872372 0.0612
TDF(1) 6.072787 3.228616 1.880926 0.0600

Log likelihood 471.9555 Akaike info criterion -4.693020
Avg. log likelihood 2.371636 Schwarz criterion -4.610274
Number of Coefs. 5 Hannan-Quinn criter. -4.659530

Convergence not achieved after 1000 iterations 这句话是怎么回事？我该怎么处理呢

[此贴子已经被作者于2007-5-21 17:22:30编辑过]

LogL: LL_LL
Method: Maximum Likelihood (Marquardt)
Date: 05/21/07 Time: 17:16
Sample: 1 199
Included observations: 199
Evaluation order: By observation
Estimation settings: tol= 1.0e-05, derivs=accurate numeric
Initial Values: NU(1)=2.00000, MU(1)=0.00118, OMEGA(1)=0.00242,
ALPHA(1)=-0.00926, BETA(1)=0.58666
Convergence achieved after 2 iterations
WARNING: Singular covariance - coefficients are not unique

Coefficient Std. Error z-Statistic Prob.

NU(1) 0.906328 NA NA NA
MU(1) 0.012924 NA NA NA
OMEGA(1) -3.12E-05 NA NA NA
ALPHA(1) -0.002935 NA NA NA
BETA(1) 0.357079 NA NA NA

Log likelihood 342.3151 Akaike info criterion -3.390102
Avg. log likelihood 1.720177 Schwarz criterion -3.307356
Number of Coefs. 5 Hannan-Quinn criter. -3.356612

为什么有的基金数据比如基金科翔不能估计出来结果啊数据如下：

118777.rar (11.01 KB) 本附件包括：

• kexiang.wf1

2007-5-21 20:14:00 上传

为什么有的基金数据比如基金科翔不能估计出来结果啊

[此贴子已经被作者于2007-5-21 21:06:43编辑过]

是不是初值出了问题啊

WARNING: Singular covariance - coefficients are not unique

单协方差系数不是唯一的，这是怎么回事啊呵呵真是有点晕了，前面都好好的，后面的好几个基金都出现童言的问题，这该怎么办啊？请大师指点

[此贴子已经被作者于2007-5-21 21:06:05编辑过]

你都不检验数据?

第103个样本你算对了吗?

log(1.2173)-log(1.2506)=?

程序中ll_l.ml(showopts,m=1000,c=1e-5)

可将c=1e-5改为c=1e-7做比较

以下是引用yiyo900在2007-5-22 8:09:00的发言：

你都不检验数据?

第103个样本你算对了吗?

log(1.2173)-log(1.2506)=?

程序中ll_l.ml(showopts,m=1000,c=1e-5)

可将c=1e-5改为c=1e-7做比较

hehe,不好意思，我看看

[此贴子已经被作者于2007-5-22 10:30:04编辑过]

基金科翔的确是我看错了。可是基金兴安的数据我检查了一遍，t分布下仍然没有结果，但ged分布下却有结果，也按照你的要求将此c=1e-5改成了c=1e-7可是结果仍然如下：

LogL: LL_L
Method: Maximum Likelihood (Marquardt)
Date: 05/22/07 Time: 10:32
Sample: 1 199
Included observations: 199
Evaluation order: By observation
Estimation settings: tol= 1.0e-07, derivs=accurate numeric
Initial Values: MU(1)=0.00085, OMEGA(1)=0.00010, ALPHA(1)=0.4825
5, BETA(1)=0.44058, TDF(1)=3.00000
Convergence achieved after 2 iterations
WARNING: Singular covariance - coefficients are not unique

Coefficient Std. Error z-Statistic Prob.

MU(1) 0.001307 NA NA NA
OMEGA(1) 0.000149 NA NA NA
ALPHA(1) 0.502401 NA NA NA
BETA(1) 0.541528 NA NA NA
TDF(1) 3.082311 NA NA NA

Log likelihood 464.8282 Akaike info criterion -4.621389
Avg. log likelihood 2.335820 Schwarz criterion -4.538643
Number of Coefs. 5 Hannan-Quinn criter. -4.587899

数据如下： 119012.rar (38.57 KB) 本附件包括：

• jjsj.xls

2007-5-22 10:36:00 上传

[此贴子已经被作者于2007-5-22 10:35:58编辑过]

基金金元、汉博也是同样的情况，t分布没有结果，而ged分布却是有结果。数据我已经检查过了，c也改过了，难道出了别的问题？

LogL: LL_L
Method: Maximum Likelihood (Marquardt)
Date: 05/21/07 Time: 21:01
Sample: 1 199
Included observations: 199
Evaluation order: By observation
Estimation settings: tol= 1.0e-05, derivs=accurate numeric
Initial Values: MU(1)=0.00251, OMEGA(1)=9.2e-05, ALPHA(1)=0.2798
0, BETA(1)=0.58117, TDF(1)=3.00000
Convergence achieved after 3 iterations
WARNING: Singular covariance - coefficients are not unique

Coefficient Std. Error z-Statistic Prob.

MU(1) 0.002925 NA NA NA
OMEGA(1) -0.000125 NA NA NA
ALPHA(1) 0.176879 NA NA NA
BETA(1) 0.812159 NA NA NA
TDF(1) 6.068057 NA NA NA

Log likelihood 462.9452 Akaike info criterion -4.602464
Avg. log likelihood 2.326358 Schwarz criterion -4.519718
Number of Coefs. 5 Hannan-Quinn criter. -4.568975

[此贴子已经被作者于2007-5-22 10:54:59编辑过]