Dear Tom,
My model structure is as VAR-MGARCH-in Mean and I use DCC(one step estimation) model. My prof. wants to see the IRF of these 6 variate DCC-mean model.
I look for the codes but I realized that it is imposible for "mean model" and also Mgarch model which use other than VEC/BEKK model structure.
Am I right on this issue?
If I am right , is there any paper or sth. else that highlight this use? Because, I have to give reference.
DCC Impulse_Response Function
Re: DCC Impulse_Response Function
See Engle and Sheppard(2001), "Theoretical and Empirical Properties of Dynamic Conditional Correlation Multivariate GARCH", NBER Working Paper 8554. That's not specifically about IRF's, but discusses the problem with closed form variance forecasts with DCC. The mean-model IRF's for your model are complicated by the "M" effect, which means that the IRF's depend not just upon the mean model dynamics but also the variance model dynamics.
Re: DCC Impulse_Response Function
Dear Tom,
Thanks for your quick reply. I got that article that tlaks about forecast of DCC and its bias. But I can not able to figure out its relation with impulse response function. Can you open it a bit, please
So can we say,
1. It is impossible/ meaningless to get IRF of DCC model . I can not even write a code.
or
2. Yes, we can get IRF but its calculation is very difficult to find. I can write a code...
Because I can not able to get any article that use VAR-MGARCH-mean using DCC than applying impulse-response.
Thanks for your quick reply. I got that article that tlaks about forecast of DCC and its bias. But I can not able to figure out its relation with impulse response function. Can you open it a bit, please
So can we say,
1. It is impossible/ meaningless to get IRF of DCC model . I can not even write a code.
or
2. Yes, we can get IRF but its calculation is very difficult to find. I can write a code...
Because I can not able to get any article that use VAR-MGARCH-mean using DCC than applying impulse-response.
Re: DCC Impulse_Response Function
An impulse response function is a difference between a forecast with a particular shock and a forecast without one. If there's no simple way to do a forecast, there will be no simple way to do an impulse response function. In addition, an impulse response function in a non-linear model will generally depend upon the size of the shock and on the initial conditions so there isn't a generic "impulse response function".
See Section 18.2.2 Structural Analysis in Kilian and Lutkepohl's soon-to-be-published book: http://www-personal.umich.edu/~lkilian/book.html. That's not specifically for your case, but discusses the general problems.
See Section 18.2.2 Structural Analysis in Kilian and Lutkepohl's soon-to-be-published book: http://www-personal.umich.edu/~lkilian/book.html. That's not specifically for your case, but discusses the general problems.
Hafner Herwartz 2006
Dear Tom,
I am again here for asking the same question about impulse-response function for DCC.
If you can look to the past posts and and my model which is a "in-mean model" you said it is nearly impossible to find out the impulse-response of DCC model with mean equation.
"The mean-model IRF's for your model are complicated by the "M" effect, which means that the IRF's depend not just upon the mean model dynamics but also the variance model dynamics."
I make some suggestions and also make some quotations from the papers you suggested
But my prof. insists on the IRFs of the model and his last comment is:
"These days nonlinear impulse responses(which depend on the size of the shock and on the initial conditions) are becoming easy to compute thus such exercise is not "impossible".
Therefore, I will be so pleased If you give me suggestions about what he wants and in which test I can perform his request. Because I can not able to find anything about his request and I am in a hole
Thanks in advance
Kind regards.
I also find these papers but can not be sure whether there is a code for them.
http://citeseerx.ist.psu.edu/viewdoc/do ... 1&type=pdf
https://www.omicsonline.org/open-access ... ?aid=90840
I am again here for asking the same question about impulse-response function for DCC.
If you can look to the past posts and and my model which is a "in-mean model" you said it is nearly impossible to find out the impulse-response of DCC model with mean equation.
"The mean-model IRF's for your model are complicated by the "M" effect, which means that the IRF's depend not just upon the mean model dynamics but also the variance model dynamics."
I make some suggestions and also make some quotations from the papers you suggested
But my prof. insists on the IRFs of the model and his last comment is:
"These days nonlinear impulse responses(which depend on the size of the shock and on the initial conditions) are becoming easy to compute thus such exercise is not "impossible".
Therefore, I will be so pleased If you give me suggestions about what he wants and in which test I can perform his request. Because I can not able to find anything about his request and I am in a hole
Thanks in advance
Kind regards.
I also find these papers but can not be sure whether there is a code for them.
http://citeseerx.ist.psu.edu/viewdoc/do ... 1&type=pdf
https://www.omicsonline.org/open-access ... ?aid=90840
Re: DCC Impulse_Response Function
The second edition of the GARCH course has quite a bit more about simulation methods than the first.
From what you've written and the papers you've cited, I'm not at all sure what you actually mean by "impulse response function". In a VAR-GARCH model, there are two possible types of IRF's---the IRF's for the mean and the IRF's for the variance. The DCC paper is all about IRF's for the variance, which are actually quite simple for a straight DCC model---the structure of the mean model doesn't matter at all. It's the IRF's for the mean that are complicated (a lot) by the fact that you have an M term. It's not that it's especially difficult to do simulations of the model (you can just simulate the errors using the DCC-GARCH model in isolation, then feed those in to solving the mean model), but that the results are likely to be so dependent upon initial conditions and shock sizes that there won't be much value in the results. (It may also take a very large number of draws to get stable estimates, even given specific initial conditions and shock sizes).
From what you've written and the papers you've cited, I'm not at all sure what you actually mean by "impulse response function". In a VAR-GARCH model, there are two possible types of IRF's---the IRF's for the mean and the IRF's for the variance. The DCC paper is all about IRF's for the variance, which are actually quite simple for a straight DCC model---the structure of the mean model doesn't matter at all. It's the IRF's for the mean that are complicated (a lot) by the fact that you have an M term. It's not that it's especially difficult to do simulations of the model (you can just simulate the errors using the DCC-GARCH model in isolation, then feed those in to solving the mean model), but that the results are likely to be so dependent upon initial conditions and shock sizes that there won't be much value in the results. (It may also take a very large number of draws to get stable estimates, even given specific initial conditions and shock sizes).