Saving GARCH model parameter estimates

[tx_ecnmst]

I am trying to estimate a bivariate GARCH-M but cannot get convergence for my model. I would like to try using the genetic algorithm (or similar) to get initial parameter estimates and then use BFGS to obtain final estimates. How do I save the estimated parameters from step one and how do I put those into a form that can be used with the 'INITIAL =' option in GARCH? I tried using the 'vechmat=' option but it is not available for GENETIC algorithm. Thank you.

[TomDoan]

INITIAL=%BETA will do that, but you should really just use GENETIC as your PMETHOD, rather than doing first SIMPLEX then GENETIC. (GENETIC does a broad search on its own, so doing simplex first doesn't really help much). It appears to still have convergence issues. Have you tried doing a simpler model like CC (at least to start).

[tx_ecnmst]

Thank you, Tom.

[tx_ecnmst]

Hi Tom,

I am attempting to estimate the impact of the risk premium on nominal interest rates in a bivariate VECM-GARCH-M (code and data attached).The empirical model as set up relates the two endogenous variables to the covariance of nominal rates and inflation (HHS(1,2)) but the theory says that the risk premium is the covariance between the real rate and inflation. But there is simple linear relationship (unless my math is off) such that COV(r,pi) = COV(i,pi)-VAR(pi), where i is nominal, r real and pi = inflation. So I need to modify the current model specification to have the mean model depend not on the HHS parameters, but a linear combination of them, so that each VAR equation will have something like HHS(2,1)-HHS(1,1) as an independent variable.

Thanks,

Bill

[TomDoan]

Use the HADJUST option to create the series you need as part of the GARCH instruction:

https://estima.com/webhelp/topics/garch ... model.html

[tx_ecnmst]

Thanks, Tom. That worked great.

Code: Select all

set covr = 0.0
dec symm[series] hhs(2,2)
clear(zeros) hhs
equation(REGRESSORS) infeq dpi 13
# d3{1 to 13} realr3{1} hhs(1,1) covr
equation(REGRESSORS) nomeq d3 13
# dpi{1 to 13} realr3{1} hhs(1,1) covr
group garchm infeq nomeq

nlpar(exactline,derive=fourth,mutate=best,alpha=0.5)
garch(model=garchm,p=1,q=1,method=bfgs,pmethod=simplex,piters=500,$
mvhseries=hhs,rvectors=rd,hmatrices=hh,stdresids=rstd,$
factorby=eigen,iters=1000,subiters=1000,derives=dd,$
hadjust=(covr=(hhs(2,1)-hhs(1,1))),mv=cc)