The RATS Software Forum

You have monthly data. That, by itself, often changes the "GARCH" behavior. Also, there is no reason a priori to believe that DCC is a good explanation. Have you done a CC model with the same data?

Dear Tom,


I want your help considering the DCC-GARCH estimation.
Can you please refer to some codes in order to extract conditional volatility, dynamic conditional correlation and dynamic conditional covariance as series after estimating the DCC-GARCH model?


Regards

That's the same for any type of multivariate GARCH model, not just DCC:

https://estima.com/docs/RATS%209%20User ... f#page=328

Hello

I have estimated the conditional correlation using DCC-GARCH and got insignificant DCC-alpha while DCC-beta is strongly significant. I have tried to estimate CCC instead but also got insignificant correlation. Could you please help and advise what could be the reason and possible solution?

Many thanks!

It sounds like you have series which have relatively small contemporaneous correlation. If that's true, you'll never get a DCC model to work well.

Thank you very much the prompt reply!

It is true, the correlation is very small between the variables. Does it mean that CCC is also not working because of a such small correlation? is there any solution possible?

Why do you need a "solution"? Sometimes zero is the right answer.

My research aims to find out whether the correlation is time-varying or constant. Since both models (DCC and CCC) provide insignificant estimates i don't know how to correctly interpret these results, and explain why CCC gives insignificant correlation estimate (what is the reason?). Maybe there is something i could do?

If the correlation is small, it's very hard to tell CC and DCC apart---the log likelihood is very flat across the range of correlations near zero.

You can do a Tse test for CC and a fluctuations test on the CC model to see whether there's any reason to believe that the CC is inadequate.

Dear Tom

I need your help and advise. I am estimating VAR-DCC-GARCH. I have 3131 observations. Information criteria in a VAR show 5 lags, but when i run multivariate diagnostic tests (Li-McLeod and Hosking) on standardized residuals the test statistics reject the null of no serial correlation. However, the test statistics fail to reject the null in the squared standardized residuals. Does these results mean that the conditional mean is misspecified? If this true, how can i correct it?

Looking forward to your response.

Thank you!

If it's a really strong rejection, then it sounds like you have some major outliers very near each other (say, five entries apart). On the other hand, with 3000+ data points, you will almost never find a GARCH model which will pass the diagnostics at the conventional .05 level.

Dear Tom

Thank you for your help and time.

With regards to multivariate diagnostic tests (Li-McLeod and Hosking) for DCC-GARCH model that is estimated for 3131 data points, do you believe the problem with rejection is due to test nature or GARCH model as such?

Please share if you have any ideas how to test the model with this much data comprising sub periods with diverse characteristics

Thank you very much!

If there are "subperiods with diverse characteristics" (other than the typical high and low volatility periods that would be common in a GARCH model) then you would either need to figure out whether there is some way to account for those in the model, or break the estimates into more homogeneous subperiods. The fact that it's hard to get any GARCH model to pass all diagnostics when you have 3000 data points isn't an excuse to ignore possible problems with the model. The question is whether there is anything that's "fixable". With 3000+ data points, a .04 autocorrelation is statistically significant, though probably economically insignificant, particularly if it's on a longer lag (like 5 or 6).

Hey Tom,
I am working on DCC-Garch with VARMA variance. I new to RATS and time series analysis.
1) I am getting this error
MV-DCC GARCH with VARMA Variances - Estimation by BFGS
NO CONVERGENCE IN 44 ITERATIONS. FINAL NORMED GRADIENT 5197.30853
SUBITERATIONS LIMIT EXCEEDED.
ESTIMATION POSSIBLY HAS STALLED OR MACHINE ROUNDOFF IS MAKING FURTHER PROGRESS DIFFICULT
TRY DIFFERENT SETTING FOR EXACTLINE, DERIVES OR ALPHA ON NLPAR
RESTARTING ESTIMATION FROM LAST ESTIMATES OR DIFFERENT INITIAL GUESSES/PMETHOD OPTION
MIGHT ALSO WORK
With Heteroscedasticity/Misspecification Adjusted Standard Errors
Usable Observations 3736
Log Likelihood -30122.4401

when running code with VARMA variance.
my question is what VARMA variance does. And when I'm running without it there is convergence. the same is happening in the CC model. how it can be corrected. I have tried to increase iterations. but still its same.
the following codes were used:
GARCH(P=1,Q=1,model=var1,mv=dcc,variances=varma,ITERS=7000,PITERS=10,pmethod=simplex,ROBUST,hmatrices=hh,rvectors=rd) / RIBOV RMOEX RSENSEX RSHCOMP RJALSH
@regcrits
set z1 = rd(t)(1)/sqrt(hh(t)(1,1))
set z2 = rd(t)(2)/sqrt(hh(t)(2,2))
set z3 = rd(t)(3)/sqrt(hh(t)(3,3))
set z4 = rd(t)(2)/sqrt(hh(t)(4,4))
set z5 = rd(t)(3)/sqrt(hh(t)(5,5)
@bdindtests(number=20) z1
@bdindtests(number=20) z2
@bdindtests(number=20) z3
@bdindtests(number=20) z4
@bdindtests(number=20) z5

And, my second question is the @MVARCH test is still showing there is ARCH effect. is the model not correct or there is another diagnostic test which can be used to justify the results are correct.

You can read more about VARIANCES=VARMA at

https://estima.com/ratshelp/index.html? ... tput_VARMA

It does tend to be hard to use as the number of variables increases. Until you have a model which converges you shouldn't worry too much about the tests for residual ARCH. Make sure you have an adequate mean model (checking for serial correlation in the mean) before worrying too much about that.