Re: st: Heteroskedasticity still presents after using DCC-MGARCH

["Ranad P." <ranad112011@gmail.com>]

Hi Tirthankar

Many thanks for the info.

Yes, I checked the PACF of  square of the prices to see if there is
heteroskedasticity on prices or not. Actually I am working on prices,
"the returns" is just a way to mention it, sorry about this confusion.
I also use LM test to test for ARCH effects

My model selection and specifications:
-I tried to types of mean equation
 1):  I use VAR for the mean equation after checking for lag length
based on AIC.  It is VAR(1) based on AIC
 2) I use ARIMA for each mean equation .

- For the volatility equation, I tried from GARCH (5,5) to GARCH
(1,1), I even tried to change order of GARCH terms following the
significant in PACF of square residuals.

In addition to checking PACF, I sometimes use the Box-Ljung type
portmanteau tests and LM test but the test always reject the null
hypothesis of homoskedasticity.

Thank you for the links. I am reading it now.

Best,
Ranad


On Thu, Sep 27, 2012 at 5:34 AM, Tirthankar Chakravarty
<tirthankar.lists@gmail.com> wrote:
> A few points:
>
> - You say that you are checking the square of the prices for
> heteroskedasticity, but are modelling the returns?
>
> - Also, while the PACF is a visual indicator, you need to perform
> actual tests, like the Box-Ljung type portmanteau tests on the squared
> residuals to be able to say something statistically definite. You
> might wish to start here:
> http://onlinelibrary.wiley.com/doi/10.1111/1467-9892.00061/abstract
>
> - You need to say what your model specification looks like.  You have
> not told us your model selection device -- have you tried adding
> further lags to your volatility equations?
>
> T
>
> On Wed, Sep 26, 2012 at 11:33 PM, Ra Nad <ranad112011@gmail.com> wrote:
>> Hi all,
>>
>> I have a problem  running the DCC-MGARCH model.Heteroskedasticity
>> still presents after using DCC-MGARCH no matter what mean equation or
>> order of ARCH/GARCH that I use
>>
>> I tested several series of returns and found no autocorrelation but
>> heteroskedasticity in the returns . So I use DCC-MGARCH model to
>> estimate their correlation.After running the DCC-MGARCH, I checked the
>> residuals for heteroskedasticity and found heteroskedasticity still
>> presents. The special thing is, whatever lag order or mean equation I
>> use, when I check for PACF of the squared residuals, they are still
>> significant at the same lags like the PACF of the squared return.
>>
>> I don't know if my explanation is clear or not so I give an example below:
>>
>> When checking the PACF of  squared prices A,B, and C, I found the PACF
>> is significant at lag 1,2,3,9 for squared A, the PACF is significant
>> at lag 2 for squared B, and at lag 3,5 for squared C. After running
>> the DCC-MGARCH  with several types of mean equation such as VAR,
>> ARIMA, I check for the residual's PACF and found PACF of squared A are
>> still significant at lag 1,2,3,9 , PACF of squared B is still
>> significant at lag 2, the same happens for squared C. So basically the
>> DCC-MGARCH did not remove any ARCH effect at all.
>>
>> The problem happens will every return series that I have. I also tried
>> to use ARCH or GARCH for each returns individually, but the
>> residuals's PACF after using ARCH or GARCH or DCC-MGARCH look just
>> exact the same like the PACF of the squared returns.
>>
>> Do you know why this problem occurs? and What should I do to fix it?
>> Many thanks for any suggestion.
>>
>> Best,
>> Ranad
>> *
>> *   For searches and help try:
>> *   http://www.stata.com/help.cgi?search
>> *   http://www.stata.com/support/statalist/faq
>> *   http://www.ats.ucla.edu/stat/stata/
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/