运用stata实证分析建立garch模型遇到的问题获取了一段时间的每日的沪深300指数的收盘价,得到每日的对数收益率.数据处理和分析采用stata软件.分析1.统计性分析sum rt,detail /得到峰度偏度/Percentiles smallest-|||-1% -.0550933 -.0969517-|||-5% -.0330366 -.0845597-|||-10% -.0210769 -0800166 obs 1941-|||-25% -.0089924 -0.792082 sum of wgt. 1941-|||-50% .0012618 Mean .0004669-|||-Largest std. Dev. .019129-|||-75% .0109495 .0786269-|||-90% .0223076 .079472 variance .0003659-|||-95% .0298617 0888131 skewness -.3637401-|||-99% : 0471076 .0893088 Kurtosis 5.669473偏度为-0.3627小于0,峰度为5.669473大于3,初步判定不服从正态分布.sktest rt /进一步检验是否服从正态分布/Percentiles smallest-|||-1% -.0550933 -.0969517-|||-5% -.0330366 -.0845597-|||-10% -.0210769 -0800166 obs 1941-|||-25% -.0089924 -0.792082 sum of wgt. 1941-|||-50% .0012618 Mean .0004669-|||-Largest std. Dev. .019129-|||-75% .0109495 .0786269-|||-90% .0223076 .079472 variance .0003659-|||-95% .0298617 0888131 skewness -.3637401-|||-99% : 0471076 .0893088 Kurtosis 5.669473P=0.0000小于0.05,拒绝服从正态分布的原假设,即rt不服从正态分布.2.平稳性检验line rt t,yline(0) /做图初步看平稳性/Percentiles smallest-|||-1% -.0550933 -.0969517-|||-5% -.0330366 -.0845597-|||-10% -.0210769 -0800166 obs 1941-|||-25% -.0089924 -0.792082 sum of wgt. 1941-|||-50% .0012618 Mean .0004669-|||-Largest std. Dev. .019129-|||-75% .0109495 .0786269-|||-90% .0223076 .079472 variance .0003659-|||-95% .0298617 0888131 skewness -.3637401-|||-99% : 0471076 .0893088 Kurtosis 5.669473dfuller rt /平稳性检验/Percentiles smallest-|||-1% -.0550933 -.0969517-|||-5% -.0330366 -.0845597-|||-10% -.0210769 -0800166 obs 1941-|||-25% -.0089924 -0.792082 sum of wgt. 1941-|||-50% .0012618 Mean .0004669-|||-Largest std. Dev. .019129-|||-75% .0109495 .0786269-|||-90% .0223076 .079472 variance .0003659-|||-95% .0298617 0888131 skewness -.3637401-|||-99% : 0471076 .0893088 Kurtosis 5.669473T统计量为-43.269小于置信度为1%时的临界值,则拒绝有单位根的原假设,即没有单位根,即该序列是平稳的.3.自相关检验corrgram rt /自相关检验/Percentiles smallest-|||-1% -.0550933 -.0969517-|||-5% -.0330366 -.0845597-|||-10% -.0210769 -0800166 obs 1941-|||-25% -.0089924 -0.792082 sum of wgt. 1941-|||-50% .0012618 Mean .0004669-|||-Largest std. Dev. .019129-|||-75% .0109495 .0786269-|||-90% .0223076 .079472 variance .0003659-|||-95% .0298617 0888131 skewness -.3637401-|||-99% : 0471076 .0893088 Kurtosis 5.669473从上图可以看出,rt与l4.rt存在自相关.4.建立arch模型 reg rt l4.rt /做rt和它的四阶滞后回归/Percentiles smallest-|||-1% -.0550933 -.0969517-|||-5% -.0330366 -.0845597-|||-10% -.0210769 -0800166 obs 1941-|||-25% -.0089924 -0.792082 sum of wgt. 1941-|||-50% .0012618 Mean .0004669-|||-Largest std. Dev. .019129-|||-75% .0109495 .0786269-|||-90% .0223076 .079472 variance .0003659-|||-95% .0298617 0888131 skewness -.3637401-|||-99% : 0471076 .0893088 Kurtosis 5.669473predict e,resgen e2=e^2corrgram e2 /回归得到残差再得到残差的平方即方差看方差的自相关性/Percentiles smallest-|||-1% -.0550933 -.0969517-|||-5% -.0330366 -.0845597-|||-10% -.0210769 -0800166 obs 1941-|||-25% -.0089924 -0.792082 sum of wgt. 1941-|||-50% .0012618 Mean .0004669-|||-Largest std. Dev. .019129-|||-75% .0109495 .0786269-|||-90% .0223076 .079472 variance .0003659-|||-95% .0298617 0888131 skewness -.3637401-|||-99% : 0471076 .0893088 Kurtosis 5.669473从上图可以看出我们残差平方存在自相关,可以拟合garch模型.arch rt l4.rt ,arch(1) garch(1)Percentiles smallest-|||-1% -.0550933 -.0969517-|||-5% -.0330366 -.0845597-|||-10% -.0210769 -0800166 obs 1941-|||-25% -.0089924 -0.792082 sum of wgt. 1941-|||-50% .0012618 Mean .0004669-|||-Largest std. Dev. .019129-|||-75% .0109495 .0786269-|||-90% .0223076 .079472 variance .0003659-|||-95% .0298617 0888131 skewness -.3637401-|||-99% : 0471076 .0893088 Kurtosis 5.6694735.模型验证predict r,res gen r2=r^2corrgram r2 /回归得到残差再得到残差的平方即方差看方差的自相关性/Percentiles smallest-|||-1% -.0550933 -.0969517-|||-5% -.0330366 -.0845597-|||-10% -.0210769 -0800166 obs 1941-|||-25% -.0089924 -0.792082 sum of wgt. 1941-|||-50% .0012618 Mean .0004669-|||-Largest std. Dev. .019129-|||-75% .0109495 .0786269-|||-90% .0223076 .079472 variance .0003659-|||-95% .0298617 0888131 skewness -.3637401-|||-99% : 0471076 .0893088 Kurtosis 5.669473从上图可以看出残差平方依然存在自相关,即模型拟合效果不理想.除了reg rt l4.rt之外,也做了arima(1,0,1) arima(2,0,2)等等,然后再拟合garch模型,但是结果还是存在自相关,是我哪里做错了吗?

运用stata实证分析建立garch模型遇到的问题
获取了一段时间的每日的沪深300指数的收盘价,得到每日的对数收益率.数据处理和分析采用stata软件.
分析
1.统计性分析
sum rt,detail    //得到峰度偏度//

偏度为-0.3627小于0,峰度为5.669473大于3,初步判定不服从正态分布.
sktest rt       //进一步检验是否服从正态分布//

P=0.0000小于0.05,拒绝服从正态分布的原假设,即rt不服从正态分布.
2.平稳性检验
line rt t,yline(0)        //做图初步看平稳性//

dfuller rt        //平稳性检验//

T统计量为-43.269小于置信度为1%时的临界值,则拒绝有单位根的原假设,即没有单位根,即该序列是平稳的.
3.自相关检验
corrgram rt        //自相关检验//

从上图可以看出,rt与l4.rt存在自相关.
4.建立arch模型
 reg rt l4.rt       //做rt和它的四阶滞后回归//

predict e,res
gen e2=e^2
corrgram e2 //回归得到残差再得到残差的平方即方差看方差的自相关性//

从上图可以看出我们残差平方存在自相关,可以拟合garch模型.
arch rt l4.rt ,arch(1)  garch(1)

5.模型验证
predict r,res  
gen r2=r^2
corrgram r2  //回归得到残差再得到残差的平方即方差看方差的自相关性//

从上图可以看出残差平方依然存在自相关,即模型拟合效果不理想.
除了reg rt l4.rt之外,也做了arima(1,0,1)  arima(2,0,2)等等,然后再拟合garch模型,但是结果还是存在自相关,是我哪里做错了吗?