Markov Switch Model TVTP

[TimTan]

Hi everyone,

Sorry this will be a kind of stupid question, I am using Eviews 8.0 (should be most updated) and I am not that familiar in reading the output information.


I try to run a Markov Switch TVTP, and It seems that both of my regimes results are fine, but my transition matrix ........... are having problems? (Output results are below.) Does this mean that all my results are void because of transition matrix? (In case of same data doing fine in MSAR model, is this likily to be a case of bad Probability regressors ???

Thanks

Convergence achieved after 22 iterations

Variable Coefficient Std. Error z-Statistic Prob.

Regime 1

C -2.548807 0.881127 -2.892666 0.0038

Regime 2

C 1.737608 0.194046 8.954613 0.0000

Common

AR(1) -0.208022 0.057056 -3.645929 0.0003
AR(2) -0.170526 0.056672 -3.008992 0.0026
AR(3) -0.167335 0.056525 -2.960349 0.0031
AR(4) 0.753223 0.056699 13.28467 0.0000
LOG(SIGMA) 0.584460 0.062001 9.426685 0.0000

Transition Matrix Parameters

P11-C 2.052101 2.707579 0.757910 0.4485
P11-TW_ECOCIA -0.165481 0.181409 -0.912197 0.3617
P21-C 1.584755 2.448984 0.647107 0.5176
P21-TW_ECOCIA -0.220942 0.132846 -1.663139 0.0963

Mean dependent var 1.480750 S.D. dependent var 4.697781
S.E. of regression 2.049150 Sum squared resid 676.0415
Durbin-Watson stat 1.812625 Log likelihood -354.4341
Akaike info criterion 4.350406 Schwarz criterion 4.554951
Hannan-Quinn criter. 4.433421

p.s. any tips or suggestion on how to do Markov Switch and TVTP on Eviews for Newcomer?

[EViews Glenn]

What problem are you seeing in the transition matrix parameters?

[TimTan]

Hi Glenn:

Thank you for your prompt reply, it seems that the results from both regimes (and common terms) are all significantly differs from zero (0), but that the Transition Matrix parameters' results have very high prob. ........ as such ......... I am not sure does this imply that no transition matrix and led to a void of other results as well........ not so sure..........can ask for some help???

Thanks,

Tim.

[EViews Glenn]

The transition matrix parameters are in logistic parameter space and are a bit more difficult to interpret. If they were *all* zero it would imply equal state transition probabilities in all regimes, not no transitions at all.

I wouldn't be that concerned as transition probabilities are hard to estimate, but If you are concerned about this, you should do Walds for a specific alternative.

[TimTan]

Thanks for your kind help, Glenn. I felt much better now.

The reason I was worry was that my expected duration for 1 regime was short (about 2 quarters) and the other regime was very long (1000+ quarters) which excess my total observations (about 150+ quarters)........... so.......... I guess some how it is not transiting ....... I will try to check my GDP data

[EViews Glenn]

That's a separate issue from insignificance. It is certainly possible that the transition matrix you end up with is not sensible. What you are getting is that one regime is (near) an absorbing state. This may make sense, or it may not.

You might want to play around with different starting values to see if you get changes in the results. These models often have multiple roots.