GARCH波动过大的问题

xuenesta 发表于 2014-11-4 20:20
Estimation results of NGARCH(1,1) model:
estimates:  -0.01844564 0.02974094 0.8907237 0.06205228 ...

模型为
r(t)=u+a(t) a(t)=sigma(t)*norm（0,1）
sigma(t)=b0+b1sigma2（t-1）+b2(a(t-1)-theta*sima(t-1))^2

xuenesta 发表于 2014-11-4 20:29
模型为
r(t)=u+a(t) a(t)=sigma(t)*norm（0,1）
sigma(t)=b0+b1sigma2（t-1）+b2(a(t-1)-theta*sima(t- ...

From your parameters and estimated vol, looks like you are using annualized return (multiply by 252 or something else) to do the estimation. If this is true, this is what causes the problem. If this is not true, then we can take a further look.

best,

Chemist_MZ 发表于 2014-11-5 11:42
From your parameters and estimated vol, looks like you are using annualized return (multiply by 25 ...

不好意思。代码中没有乘以sqrt（252）

Chemist_MZ 发表于 2014-11-5 11:42
From your parameters and estimated vol, looks like you are using annualized return (multiply by 25 ...

不好意思。代码中没有乘以sqrt（252）

xuenesta 发表于 2014-11-5 12:17
不好意思。代码中没有乘以sqrt（252）

No, I mean your original return data not the code. You see your mean return. Is around -1%. Usually if you use daily return, this number will be very small. And I can reproduce the 700% and the whole vol series using annualized return. And if you calculate the annualized return series's std, it is around 500%. That's why I suspect.

What's the return series you use to estimate the model? I see it is read from a "tmp" file in the code, but you didn't attach it. Thanks

You know debug is difficult especially with that little info and I can not run your code:-). So there is a high chance I am wrong.