Page 1 of 1

BAYESTST unit root test

Posted: Sat Jul 02, 2016 9:52 am
by akdeniz_coskun
Dear all,
I try to modify Nakajima (2011) codes for my paper. The data used in the original paper has unit root. Most of the TVP-VAR papers that I have read so far, don’t contain unit-root test statistics results. I replicate the codes with both unit root and stationary series. The series that have unit root give better results.

I think TVP-VAR estimation is based on the Bayesian techniques, so may I use BAYESTST unit root test instead of the classical unit root tests.
I have uploaded the the codes and excel file. May I interpret as all series stationary with this Bayesian unit root test ?
One of the my series (namely IP) has trend, so I have calculated it with trend. The others has been calculated without trend.

Could you give me some examples that employ this unit root tests before the estimation of the TVP-VAR model, please? (I haven't found yet)

Code: Select all

Bayesian Unit Root Test (IP with trend)
 Squared t     Schwarz Limit     Small Sample Limit   Marginal Alpha
    8.641              7.870               1.873           0.1195

Bayesian Unit Root Test (CPI)
 Squared t     Schwarz Limit     Small Sample Limit   Marginal Alpha
  347.406             15.368               9.371           0.0000

Bayesian Unit Root Test (M1)
 Squared t     Schwarz Limit     Small Sample Limit   Marginal Alpha
   20.010             13.465               7.468           0.0075

Bayesian Unit Root Test (CBRT)
 Squared t     Schwarz Limit     Small Sample Limit   Marginal Alpha
   42.705              6.437               0.441           0.0000

Bayesian Unit Root Test (FKE)
 Squared t     Schwarz Limit     Small Sample Limit   Marginal Alpha
   11.210              7.556               1.559           0.0311

Bayesian Unit Root Test (CREDIT)
 Squared t     Schwarz Limit     Small Sample Limit   Marginal Alpha
   94.586             14.965               8.968           0.0000

Re: BAYESTST unit root test

Posted: Sun Jul 03, 2016 2:46 pm
by TomDoan
akdeniz_coskun wrote:Dear all,
I try to modify Nakajima (2011) codes for my paper. The data used in the original paper has unit root. Most of the TVP-VAR papers that I have read so far, don’t contain unit-root test statistics results. I replicate the codes with both unit root and stationary series. The series that have unit root give better results.

I think TVP-VAR estimation is based on the Bayesian techniques, so may I use BAYESTST unit root test instead of the classical unit root tests.
I have uploaded the the codes and excel file. May I interpret as all series stationary with this Bayesian unit root test ?
One of the my series (namely IP) has trend, so I have calculated it with trend. The others has been calculated without trend.
No. @BAYESTST isn't really a serious proposal for testing for unit roots (in practice) because it uses an unrealistic flat prior on the deterministic components. (Sims' original idea had no deterministics, which makes it even more unrealistic).
akdeniz_coskun wrote: Could you give me some examples that employ this unit root tests before the estimation of the TVP-VAR model, please? (I haven't found yet)
No, because it's irrelevant. If the model has adequate dynamics to avoid the "spurious regression" problem, then there is no reason the data can't be non-stationary.

Re: BAYESTST unit root test

Posted: Sun Jul 03, 2016 5:57 pm
by akdeniz_coskun
Dear Tom,

Thank you so much for your reply, I totally misunderstood the procedure. Thanks for correcting my thought about the test.

Regards.