GARCH-BEKK with initial values vector
Posted: Mon Apr 10, 2017 4:57 am
Hi,
I would like to estimate a GARCH BEKK without exogenous regressors and use those coefficient estimates as input for another GARCH BEKK estimation with exogenous regressors. I saw that there's an option 'initial', but I am struggling with the way to input previously estimated coefficients. I used %BETA, but the size of that vector is concerning me. It includes the estimates of the VAR(1) model coefficients I've estimated as mean model and the BEKK coefficients, but I would like to use the values in another model via 'initial' where I include dummies in the VAR(1) and external regressors in the GARCH BEKK. I think I would need to adjust the %BETA vector to include extra zeros in order to compute a vector of the same size of the new model I intend to estimate. How would I be able to do that?
I was thinking as well that I could estimate the VAR(1) with dummies separately (just as I did for the VAR(1) without dummies), and include those estimates in the vector and then add the %BETA coefficients of just the GARCH BEKK C, A and B matrices. How would I be able to do that?
I'm doing this because the GARCH model won't converge (even when adjusting the simplex iterations and BFGS iterations), so I want to change the initial values to help it converge.
It concerns this code (the code for the VAR(1) is standard):
Many thanks in advance for your help!
Best,
Myrthe
I would like to estimate a GARCH BEKK without exogenous regressors and use those coefficient estimates as input for another GARCH BEKK estimation with exogenous regressors. I saw that there's an option 'initial', but I am struggling with the way to input previously estimated coefficients. I used %BETA, but the size of that vector is concerning me. It includes the estimates of the VAR(1) model coefficients I've estimated as mean model and the BEKK coefficients, but I would like to use the values in another model via 'initial' where I include dummies in the VAR(1) and external regressors in the GARCH BEKK. I think I would need to adjust the %BETA vector to include extra zeros in order to compute a vector of the same size of the new model I intend to estimate. How would I be able to do that?
I was thinking as well that I could estimate the VAR(1) with dummies separately (just as I did for the VAR(1) without dummies), and include those estimates in the vector and then add the %BETA coefficients of just the GARCH BEKK C, A and B matrices. How would I be able to do that?
I'm doing this because the GARCH model won't converge (even when adjusting the simplex iterations and BFGS iterations), so I want to change the initial values to help it converge.
It concerns this code (the code for the VAR(1) is standard):
Code: Select all
*****************************************************************************************
*
* Estimate GARCH-BEKK without dummies
* with VAR(1) without dummies
*
garch(p=1,q=1,model=var1model,mv=bekk,rvector=rr,hmatrices=hh,pmethod=simplex,piters=20,iters=1000,robusterrors)
*
* need variables with coefficients to use later for initial_values in next GARCH-BEKK
*
compute coefficients = %BETA
display coefficients
*
*****************************************************************************************
*
* Estimate GARCH-BEKK with external event regressors
* with VAR(1) without dummies
*
* use initial values from previous GARCH-BEKK without dummies ---> how??
*
garch(p=1,q=1,model=var1model,mv=bekk,rvector=rr_ev,hmatrices=hh_ev,initial = coefficients, iters=1000,robusterrors,xreg)
# prod_announce strat_moves fin_client neg_events
*
******************************************************************************************
* Estimate GARCH-BEKK with company related events dummies
* with VAR(1) with seasonality dummies
*
* use initial values from previous GARCH-BEKK without dummies ---> how??
*
garch(p=1,q=1,model=var1model_seas,mv=bekk,rvector=rr_ev,hmatrices=hh_ev,initial = coefficients, iters=1000,robusterrors,xreg)
# prod_announce strat_moves fin_client neg_events
*
*****************************************************************************************
Best,
Myrthe