MS-VAR with time varying probabilities

[Aktar]

Hi Tom,

I would like to identify common business cycle of a set of countries (as in krozlig 2001), and test the predictive power of leading indicators in the identification of this common cycle. The best approach may be a MS-VAR-TVTP model. Do the programm of filardo 1994 enjoy this feature? Because it use the MSVAR setup procedure.

Regards

[TomDoan]

Filardo's paper was univariate, though the extension to a VAR is fairly straightforward. It could take a very long time to estimate though. Since I have no idea what paper Krolzig(2001) is (PLEASE take the time to give a real reference) I have no idea whether that would help.

[Aktar]

The paper is : "International Business Cycles: Regime Shifts in the Stochastic Process of Economic Growth"

and the standard code for ms-var seems to be:

Code: Select all

nonlin(parmset=varparms) mu phi sigmav
nonlin(parmset=msparms) theta
gset pt_t  gstart gend = %zeros(nexpand,1)
gset pt_t1 gstart gend = %zeros(nexpand,1)
frml msvarf = log(%MSVARProb(t))
@msvarinitial gstart gend

@msvarsetup(lags=1,states=3,switch=mh)
# dusa djap dfrg duk dcan daus
@msvarinitial gstart gend
@msvarEMgeneralsetup
do emits=1,50
   @msvaremstep gstart gend
   disp "Iteration" emits "Log Likelihood" %logl
end do emits
set p1 gstart gend = (pstar=%msvarmarginal(emptsm(t),0)),pstar(1)
set p2 gstart gend = (pstar=%msvarmarginal(emptsm(t),0)),pstar(2)
set p3 gstart gend = (pstar=%msvarmarginal(emptsm(t),0)),pstar(3)

maximize(trace,parmset=varparms+msparms,start=%(p=%mslogisticp(theta),pstar=%MSVARInit()),$
reject=%minvalue(MSVARTransProbs)<0.0,method=bfgs,iters=400) msvarf gstart gend

[TomDoan]

You could certainly try to combine the two ideas. However, it's very likely that you'll get a model with quite a few modes and it's possible that none will be easy to interpret.