I can only give the two models without seasonals because I can not find seasonal option in Eviews menu. Yuniarto Hadiwibowo ----- Original Message ---- From: Allin Cottrell <cottrell(a)wfu.edu&gt; To: Gretl users <gretl-users(a)lists.wfu.edu&gt; Sent: Thu, January 21, 2010 11:14:55 AM Subject: [Gretl-users] vecm comparisons Well, one more thing: for fuller comparison, it might be nice to have the output from other software for the two models above without the seasonals: vecm 2 1 LRM LRY IBO IDE --rc vecm 2 1 LRM LRY IBO IDE And yet one more thing: gretl automatically prints the cross-equation covariance matrix. If you could get that out of other software I'd like to see it too. Allin Cottrell Vector Error Correction Estimates Date: 01/21/10 Time: 13:14 Sample (adjusted): 1974Q3 1987Q3 Included observations: 53 after adjustments Standard errors in ( ) & t-statistics in [ ] Cointegrating Eq: CointEq1 LRM(-1) 1.000000 LRY(-1) -0.969116 (0.14297) [-6.77827] IBO(-1) 5.402772 (0.54610) [ 9.89345] IDE(-1) -4.140325 (1.12517) [-3.67975] C -6.478051 (0.88530) [-7.31738] Error Correction: D(LRM) D(LRY) D(IBO) D(IDE) CointEq1 -0.299784 0.026943 0.003921 0.020001 (0.07169) (0.06160) (0.02270) (0.01492) [-4.18156] [ 0.43742] [ 0.17273] [ 1.34035] D(LRM(-1)) -0.220041 0.267268 0.002698 0.023956 (0.14064) (0.12084) (0.04454) (0.02927) [-1.56453] [ 2.21182] [ 0.06058] [ 0.81833] D(LRY(-1)) 0.076984 -0.021191 0.150092 0.033433 (0.17305) (0.14868) (0.05480) (0.03602) [ 0.44485] [-0.14253] [ 2.73894] [ 0.92819] D(IBO(-1)) 0.178382 -0.127891 0.356503 0.294057 (0.49305) (0.42361) (0.15613) (0.10263) [ 0.36179] [-0.30190] [ 2.28337] [ 2.86535] D(IDE(-1)) -1.357771 -0.791761 0.043718 0.133585 (0.61581) (0.52909) (0.19500) (0.12818) [-2.20485] [-1.49647] [ 0.22419] [ 1.04220] R-squared 0.366690 0.196647 0.276511 0.368517 Adj. R-squared 0.313914 0.129701 0.216220 0.315894 Sum sq. resids 0.036050 0.026611 0.003615 0.001562 S.E. equation 0.027405 0.023546 0.008678 0.005704 F-statistic 6.948066 2.937390 4.586285 7.002902 Log likelihood 118.0648 126.1095 179.0107 201.2499 Akaike AIC -4.266597 -4.570172 -6.566441 -7.405656 Schwarz SC -4.080720 -4.384295 -6.380564 -7.219780 Mean dependent 0.007757 0.003340 -0.001114 -0.000384 S.D. dependent 0.033086 0.025239 0.009802 0.006897 Determinant resid covariance (dof adj.) 4.90E-16 Determinant resid covariance 3.30E-16 Log likelihood 643.8520 Akaike information criterion -23.35290 Schwarz criterion -22.42352 LRM LRY IBO IDE LRM 0.000751 0.000380 -8.51E-05 -4.80E-06 LRY 0.000380 0.000554 -3.52E-06 -1.79E-05 IBO -8.51E-05 -3.52E-06 7.53E-05 1.12E-05 IDE -4.80E-06 -1.79E-05 1.12E-05 3.25E-05 Vector Error Correction Estimates Date: 01/21/10 Time: 13:12 Sample (adjusted): 1974Q3 1987Q3 Included observations: 53 after adjustments Standard errors in ( ) & t-statistics in [ ] Cointegrating Eq: CointEq1 LRM(-1) 1.000000 LRY(-1) -0.975655 (0.14438) [-6.75778] IBO(-1) 5.408588 (0.55145) [ 9.80798] IDE(-1) -4.162443 (1.13619) [-3.66350] C -6.417559 Error Correction: D(LRM) D(LRY) D(IBO) D(IDE) CointEq1 -0.281469 0.037469 -0.003902 0.019960 (0.07528) (0.06486) (0.02366) (0.01576) [-3.73889] [ 0.57772] [-0.16493] [ 1.26671] D(LRM(-1)) -0.236567 0.258051 0.010221 0.024003 (0.14293) (0.12314) (0.04492) (0.02992) [-1.65511] [ 2.09557] [ 0.22753] [ 0.80228] D(LRY(-1)) 0.079759 -0.019068 0.148606 0.033478 (0.17397) (0.14988) (0.05467) (0.03641) [ 0.45848] [-0.12723] [ 2.71806] [ 0.91937] D(IBO(-1)) 0.111450 -0.167095 0.385608 0.294132 (0.50223) (0.43269) (0.15784) (0.10513) [ 0.22191] [-0.38617] [ 2.44301] [ 2.79789] D(IDE(-1)) -1.365951 -0.792514 0.045036 0.133979 (0.61888) (0.53319) (0.19450) (0.12954) [-2.20713] [-1.48636] [ 0.23154] [ 1.03425] C 0.008956 0.001031 -0.001354 -0.000417 (0.00390) (0.00336) (0.00123) (0.00082) [ 2.29534] [ 0.30684] [-1.10408] [-0.51041] R-squared 0.373829 0.201320 0.295388 0.368572 Adj. R-squared 0.307215 0.116354 0.220430 0.301398 Sum sq. resids 0.035643 0.026456 0.003521 0.001562 S.E. equation 0.027538 0.023725 0.008655 0.005764 F-statistic 5.611876 2.369414 3.940685 5.486883 Log likelihood 118.3652 126.2641 179.7113 201.2522 Akaike AIC -4.240197 -4.538269 -6.555144 -7.368006 Schwarz SC -4.017145 -4.315217 -6.332092 -7.144954 Mean dependent 0.007757 0.003340 -0.001114 -0.000384 S.D. dependent 0.033086 0.025239 0.009802 0.006897 Determinant resid covariance (dof adj.) 5.16E-16 Determinant resid covariance 3.19E-16 Log likelihood 644.7542 Akaike information criterion -23.27374 Schwarz criterion -22.23283 LRM LRY IBO IDE LRM 0.000758 0.000383 -8.25E-05 -4.90E-06 LRY 0.000383 0.000563 -1.08E-06 -1.84E-05 IBO -8.25E-05 -1.08E-06 7.49E-05 1.14E-05 IDE -4.90E-06 -1.84E-05 1.14E-05 3.32E-05 _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users

Hello all, I'd like to request some help with comparison of VECM output across programs, if people happen to have copies of the relevant proprietary software. I'm interested in a couple of models from Johansen's VECM book. In gretl-speak these are: open denmark.gdt vecm 2 1 LRM LRY IBO IDE --rc --seasonals vecm 2 1 LRM LRY IBO IDE --seasonals That, we use the Danish data from Johansen's book and estimate a couple of VECMs, one with a constant restricted to the cointegrating space and one with an unrestricted constant (gretl's default), and both including (orthogonalized) seasonal dummies. (In gretl, the leading "2" is the lag order, which comes down to 1 in differences, and the "1" is the cointegration rank.) I've attempted to do this with stata, RATS 6.0 and R. Stata won't include the seasonals when a restricted constant is included (why not?) but it'll do the other model. RATS, it turns out, is useless for VECMs unless you also buy CATS (which I have not done); if anyone has CATS it would be nice to see output from that. R offers a melange of packages that "sort of" do VECMs, but so far as I can see, nothing coherent and fully functional. Anyway, I'm particularly interested to see what Eviews and PcGive do with these models, if anyone can oblige. Thanks in advance! Well, one more thing: for fuller comparison, it might be nice to have the output from other software for the two models above without the seasonals: vecm 2 1 LRM LRY IBO IDE --rc vecm 2 1 LRM LRY IBO IDE And yet one more thing: gretl automatically prints the cross-equation covariance matrix. If you could get that out of other software I'd like to see it too. Allin Cottrell _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users

Hi, I attached the output of JMULTI.

Dear Henrique, yes, you are right. The JMulti gui is a little bit different to gretl -- that's what confused me a little bit. The attached file should be correct now.

I've attempted to do this with stata, RATS 6.0 and R. Stata won't include the seasonals when a restricted constant is included (why not?)

for VECMs unless you also buy CATS (which I have not done); if anyone has CATS it would be nice to see output from that.

a melange of packages that "sort of" do VECMs, but so far as I can see, nothing coherent and fully functional.

Anyway, I'm particularly interested to see what Eviews and PcGive do with these models, if anyone can oblige. Thanks in advance!

On Thu, 21 Jan 2010, Sven Schreiber wrote: > Allin Cottrell schrieb: >> I've attempted to do this with stata, RATS 6.0 and R. Stata won't >> include the seasonals when a restricted constant is included (why >> not?) > My guess would be that it's because the seasonals are standard > (uncentered), and since those have a non-zero mean it would contradict > the choice of restricting the constant. (Just a guess though, stata > could of course easily include centered seasonals.) Hm, the relevant piece of syntax in stata is sindicators(varlist_si) which is annotated as follows: "sindicators(varlist_si) specifies the normalized seasonal indicator variables to include in the model. The indicator variables specified in this option must be normalized as discussed in Johansen (1995). If the indicators are not properly normalized, the estimator of the cointegrating vector does not converge to the asymptotic distribution derived by Johansen (1995). ... sindicators() cannot be specified with trend(none) or with trend(rconstant)."

If you include seasonal dummies in the "NO CONSTANT" or "restricted constant" it is unlikely that the sum of the seasonals will add to zero and you will then be imposing a trend on the undifferenced series in the vecm. In such a case the relevant statistics will not follow the "Johansen" distributions. Thus it may be more correct to not allow seasonal dummies in those cases.

John C Frain schrieb: > If you include seasonal dummies in the "NO CONSTANT" or "restricted > constant" it is unlikely that the sum of the seasonals will add to > zero and you will then be imposing a trend on the undifferenced series > in the vecm. Yeah but that's what centered seasonals are for which sum to zero by construction. We implicitly assumed that Stata's "normalized" s3 easonals concept means exactly this, because they refer to Johansen's discussion. So then there wouldn't be any reason to exclude them. But since Stata doesn't seem to have automatic seasonals, maybe they just don't trust their users to do it right and play it safe :-)

> > Stata in the option trend(trend) refers to a quadratic trend in the > undifferenced data it is correct as this corresponds to a linear trend > in the  differenced data. But that's not what they say, instead they say: >> >> "trend(trend)              include a linear trend in the cointegrating >>                                  equations

whereas Johansen's/Eviews' option number 5 has a quad trend in the cointegrating equations (and a linear trend in the differenced data, as you say). For other methods the Stata manual and help always seemed to me very illuminating, but not in this realm. But hey, they're newcomers there...

-sven _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users

Another oddity in stata. To get a look at the output with seasonals (which are not allowed in the "restricted constant" case), I specified a vecm with unrestricted constant: vec lrm lry ibo ide , rank(1) lags(2) \ trend(constant) sindicators(q1-q3) [In the docs, "trend(constant)" -> "include an unrestricted constant in model"] But stata then prints "_cons" values in both the per-equation output and the "beta" output, as in these snippets: First equation: --------------------------- | Coef. -------------+------------- D_lrm | _ce1 | L1. | -.1999204 [etc.]. _cons | .0012088 System info: ------------------------- beta | Coef. -------------+----------- _ce1 | lrm | 1 lry | -1.035892 ibo | 5.21588 ide | -4.226433 _cons | -6.048112 ------------------------- Eh?