time varying VAR
Posted: Tue Dec 13, 2011 8:02 am
HI Tom,
I would like to seek your advice. I have built a time varying VAR model as follows:
linreg wtic
# constant trend bsbc{1 to 4} hh{1 to 4} wtic{1 to 4}
do you think it is better not to include current regressors at time t i.e.
linreg wtic
# constant trend bsbc{0 to 4} hh{0 to 4} wtic{1 to 4}
how do i deal with endogeneity like in Kim's paper - Dealing with Endogeneity in Regression
Models with Dynamic Coefficients
By Chang-Jin Kim
he mentioned that there will be some endogenity problem with the new residuals.
i did a setup with the following with the example with lukepol:
nonlin lsigsqv lsigsqw
compute lsigsqw=%log(||.0001,.0001,.0001,.0001,.0001,.0001,.0001,.0001||)
compute lsigsqv=log(.1)
compute break1=.1, break2=break3=break4=.1
dlm(c=%eqnxvector(ceqn,t),sw=%diag(%exp(lsigsqw)),sv=exp(lsigsqv),$
PRESAMPLE=ERGODIC,y=bsbc,$
method=bfgs,iters=500,type=filter,vhat=wtivhat,yhat=wtiyhat) $
1997:11:28 2011:6:24 xstates vstates
do you think it is better to solve the VAR separately equation by equation like in litterman, doan example? as well as in Lukepol's example?
if i am to solve the var jointly, if i get only signigicant measurement variances for wti but not for HH and BSB or vice-versa. In fact, model specification for each variable were totally different as some have state variances which are significant for trend, constant or their own lags. does it mean time varying is not suitable for HH and bsb or even wti? also, how do i do the set up e.g.
my var:
linreg wtic
# bsb{1 to 4} hh{1 to 4} wtic{1 to 4}
linreg hh
# bsb{1 to 4} hh{1 to 4} wtic{1 to 4}
linreg bsb
# bsb{1 to 4} hh{1 to 4} wtic{1 to 4}
do i formulate a DSGE and subsequently use DLM? any examples would be great. i looked at the example of Commandeur & Koopman, An Introduction to State Space Time Series Analysis for the bivariate model but those totally separate equations unlike mine where the regressors is depdnent variable for another equation. Do i use the example from @VARTVPKSC.src?
Thnak again for your advice.
I would like to seek your advice. I have built a time varying VAR model as follows:
linreg wtic
# constant trend bsbc{1 to 4} hh{1 to 4} wtic{1 to 4}
do you think it is better not to include current regressors at time t i.e.
linreg wtic
# constant trend bsbc{0 to 4} hh{0 to 4} wtic{1 to 4}
how do i deal with endogeneity like in Kim's paper - Dealing with Endogeneity in Regression
Models with Dynamic Coefficients
By Chang-Jin Kim
he mentioned that there will be some endogenity problem with the new residuals.
i did a setup with the following with the example with lukepol:
nonlin lsigsqv lsigsqw
compute lsigsqw=%log(||.0001,.0001,.0001,.0001,.0001,.0001,.0001,.0001||)
compute lsigsqv=log(.1)
compute break1=.1, break2=break3=break4=.1
dlm(c=%eqnxvector(ceqn,t),sw=%diag(%exp(lsigsqw)),sv=exp(lsigsqv),$
PRESAMPLE=ERGODIC,y=bsbc,$
method=bfgs,iters=500,type=filter,vhat=wtivhat,yhat=wtiyhat) $
1997:11:28 2011:6:24 xstates vstates
do you think it is better to solve the VAR separately equation by equation like in litterman, doan example? as well as in Lukepol's example?
if i am to solve the var jointly, if i get only signigicant measurement variances for wti but not for HH and BSB or vice-versa. In fact, model specification for each variable were totally different as some have state variances which are significant for trend, constant or their own lags. does it mean time varying is not suitable for HH and bsb or even wti? also, how do i do the set up e.g.
my var:
linreg wtic
# bsb{1 to 4} hh{1 to 4} wtic{1 to 4}
linreg hh
# bsb{1 to 4} hh{1 to 4} wtic{1 to 4}
linreg bsb
# bsb{1 to 4} hh{1 to 4} wtic{1 to 4}
do i formulate a DSGE and subsequently use DLM? any examples would be great. i looked at the example of Commandeur & Koopman, An Introduction to State Space Time Series Analysis for the bivariate model but those totally separate equations unlike mine where the regressors is depdnent variable for another equation. Do i use the example from @VARTVPKSC.src?
Thnak again for your advice.