Re: [Gretl-users] ARMA bug?
Hi,
I think I sent an incomplete e-mail. Well, estimating the model WITHOUT
CONSTANT I got:
Convergence achieved after 15 iterations
Model 1: ARMA estimates using the 30 observations 1971-2000
Dependent variable: sales
VARIABLE COEFFICIENT STDERROR T STAT P-VALUE
car(-1) 0.460111 0.928532 0.496 0.62410
e(-1) -0.673785 0.773696 -0.871 0.39123
The model has no constant term.
F is calculated as in Sect. 4.4 of Ramanathan's Introductory Econometrics.
R-squared is the square of the correlation between the observed and fitted
values of the dependent variable.
Mean of dependent variable = 0.0511333
Standard deviation of dep. var. = 0.330484
Mean of residuals = 0.0807281
Sum of squared residuals = 3.06531
Standard error of residuals = 0.33087
Unadjusted R-squared = 0.105481
Adjusted R-squared = 0.0735342
F-statistic (2, 28) = 0.803697 (p-value = 0.458)
Log-likelihood = -8.352
AIC = 22.705
BIC = 26.908
Real Imaginary Modulus Frequency
-----------------------------------------------------------
AR Root 1 2.1734 0.0000 2.1734 0.0000
MA Root 1 1.4842 0.0000 1.4842 0.0000
-----------------------------------------------------------
Using Stata I got, WITH CONSTANT:
ARIMA regression
Sample: 1970 to 2000 Number of obs =
31
Wald chi2(2) =
12.05
Log likelihood = -6.37294 Prob > chi2 =
0.0024
------------------------------------------------------------------------------
| OPG
car | Coef. Std. Err. z P>|z| [95% Conf.
Interval]
-------------+----------------------------------------------------------------
car |
_cons | .0440328 .0318114 1.38 0.166 -.0183163
.106382
-------------+----------------------------------------------------------------
ARMA |
ar |
L1. | .6318026 .3154554 2.00 0.045 .0135214
1.250084
ma |
L1. | -1.000008 2292.182 -0.00 1.000 -4493.595
4491.595
-------------+----------------------------------------------------------------
/sigma | .287396 329.4019 0.00 0.999 -645.3285
645.9032
------------------------------------------------------------------------------
However, the BHHH algorithm in Stata does not got convergence. It was
necessary change to the BFGS one.
I think the problem is due to teh MA term, which, in fact does not exist
(this is a second difference time-series!!!).
Did you tested the original series for one (or more ) unit roots? The
spectrum was not clear about this.
I hope this help.
Ricardo
----- Original Message -----
From: "John Paravantis" <paravantis(a)otenet.gr>
To: <gretl-users(a)ricardo.ecn.wfu.edu>
Sent: Monday, January 30, 2006 10:10 PM
Subject: [Gretl-users] ARMA bug?
Methinks I found a bug in ARMA.
Copy/paste the following data into a gretl column (possibly pasting into
Word and ALT-copying and pasting the column into Excel):
0.019
0.046
-0.16
0.328
0.077
0.408
-0.08
-0.166
-0.596
0.344
0.008
0.053
0.026
0.383
-0.196
-0.23
0.012
0.249
0.276
-0.881
0.089
0.742
-0.164
0.109
-0.006
0.2
0.115
0.718
0.138
-0.299
-0.009
(They are 2nd differences of car ownership per 100 people in Greece, from
1970 to 2000 appr.)
Then try to run an ARMA(1,1): you get a dialog that "The convergence
criterion is not met". Unfortunately, it is possible to fiddle with the
convergence criterios (# of iterations) on the ARMA dialog box!
Here comes worse. Try to estimate an ARMA(0,1) WITHOUT THE CONSTANT: you
get an "out of memory error"!
Incidentally, these estimate ok in Statgraphics 5.1.
BTW thanks to all for your kind responses to my graph querry:
- Peter (of Colorado), many thanks for your tip -- I would rather not go
through the PNG format because unlike vector graphics it deteriorates when
zoomed and is usually not accepted by research journals.
- Jean, your appropriate reference to Gnumeric made me delve deeper into
Excel 2003 (yuck!) only to find out that the reason I do not get the
format confirmation dialog anymore is that Excel 2003 has now a separate
menu (Data>Import External Data...) that imports other formats! Using that
menu worked fine!
Warmest regards to all,
John
University of Piraues
Greece
PS I am mailing a separate gretl file with the above data to Allin (I
cannot attach files to the user forum address, can I?)
_______________________________________________
Gretl-users mailing list
Gretl-users(a)ricardo.ecn.wfu.edu
http://ricardo.ecn.wfu.edu/mailman/listinfo/gretl-users