The RATS Software Forum

I wish to examine the return spillover effects of various stock markets. The result that MGARCH-BEKK reports does contain parameters like mean, C, A, G and D where C=lower triangular matrix of constant, A=Arch term, G=garch term and D=asymm. parameters. The output does not have return spillover parameter (return is of stock market return series( for my research) under investigation) to measure cross market return spillover like results presented by A, G and D term. May I Know How do I measure return spillover from one market to another like A and G spillover results? Can anyone help? I shall be grateful for the reply.

You need to do a VAR(1) mean model. See, for example,

http://www.estima.com/forum/viewtopic.php?f=11&t=362

The modernized code is several posts down in that thread. What you want is similar, but using MV=BEKK.

I estimated following equation :
garch(model=var1,mv=bekk,hmatrices=h,rvector=r,pmethod=simplex,piters=90,method=bfgs,iters=600,asymmetric) / dlhsid DLJKSE DLKS11 dln225 dlsse dlbse

here six variables are stock market return series. Each return series is multiplied with 100. It displays the output with the following message:

MV-GARCH, BEKK - Estimation by BFGS
NO CONVERGENCE IN 306 ITERATIONS
LAST CRITERION WAS 0.0000000
ESTIMATION POSSIBLY HAS STALLED OR MACHINE ROUNDOFF IS MAKING FURTHER PROGRESS DIFFICULT.
TRY HIGHER SUBITERATIONS LIMIT, TIGHTER CVCRIT, DIFFERENT SETTING FOR EXACTLINE OR ALPHA ON NLPAR.
RESTARTING ESTIMATION FROM LAST ESTIMATES OR DIFFERENT INITIAL GUESSES MIGHT ALSO WORK
Usable Observations 1481

I have tried with different subiertion limits ranging from 5 to 90. But the same above message found. The process also takes lots of time to get the results. Kindly guide. What to do to achieve the convergence. Looking forward for a reply.

That's a really big model to estimate with a BEKK. Since your main interest is in the VAR coefficients, have you tried using a more tightly parameterized GARCH model, such as DCC?

TomDoan wrote:That's a really big model to estimate with a BEKK. Since your main interest is in the VAR coefficients, have you tried using a more tightly parameterized GARCH model, such as DCC?

Thanks Tom. I also wish to analyze volatility spillover. DCC does not give cross market spillover effect in volatilty like BEKK. BEKK estimation becomes difficult in my analysis. How do I proceed to analyze cross market spillover effect? Please suggest. Can choosing a small sample period of nearly 900 or 1000 instead of 1481 solve the problem?

You might try doing a shorter span first, perhaps for a limited number of iterations, then switching to the full data set. Just use the option INITIAL=%BETA on the second GARCH instruction. You could also repeat the same GARCH instruction, again with INITIAL=%BETA after it stalls. The BFGS estimate of the Hessian can sometimes collapse; restarting the process often works to get things moving again. You might also try switching to METHOD=BHHH after BFGS stalls. BHHH isn't a good choice from the start (since it will generally get the shape quite wrong at the guess values), but once you've done 300+ iterations, you're usually close enough to the optimum that the BHHH shape is more reasonable.

Thanks Tom. Now, I got my results successfully. Again thanks a lot for your help. The help feature of RATS is excellent. This deserves all our appreciation.