12:03 p.m.
Dear all
I use gretl 1.7.4 to estimate a GARCH model by using default menu function
(i.e., \Model\Time series\GARCH ) and by the command in console. But they
generate different results. The details of iterations are as follows. Any
idea?
==== results by menu=====
Automatic initialization of parameters
Regression coefficients:
theta[0] = 0.0511825
theta[1] = 1.08508
Variance parameters:
alpha[0] = 0.1
beta[0] = 0.9
Iteration 1: Log-likelihood = -168.929078042
Parameters: 0.11038 1.0851 0.10000 0.90000
Gradients: -26.738 17.074 481.64 14.243
Iteration 2: Log-likelihood = -139.605272823 (steplength = 0.0016)
Parameters: 0.067597 1.1124 0.87063 0.92279
Gradients: -4.4061 0.82275 -15.175 -8.1703
Iteration 3: Log-likelihood = -138.361721087 (steplength = 0.04)
Parameters: -0.13607 1.1652 0.87232 0.62335
Gradients: 14.823 -0.51238 -14.088 -6.9177
Iteration 4: Log-likelihood = -137.371894269 (steplength = 0.008)
Parameters: -0.12485 1.1848 0.87847 0.47302
Gradients: 13.892 -1.0676 -13.148 -5.2299
Iteration 5: Log-likelihood = -137.147777193 (steplength = 0.04)
Parameters: -0.12168 1.1207 0.88177 0.43192
Gradients: 13.651 0.40209 -12.854 -4.5731
Iteration 6: Log-likelihood = -136.065550003 (steplength = 0.2)
Parameters: -0.055939 1.1709 0.88165 0.29403
Gradients: 7.6020 -1.0743 -11.185 -0.64410
Iteration 7: Log-likelihood = -135.589722350 (steplength = 1)
Parameters: 0.054295 1.1305 0.86108 0.32088
Gradients: -3.2400 -0.48930 -11.296 -1.6096
Iteration 8: Log-likelihood = -135.332923963 (steplength = 1)
Parameters: 0.044387 1.1121 0.84171 0.26964
Gradients: -2.1320 -0.080313 -9.7947 0.99314
Iteration 9: Log-likelihood = -134.612248244 (steplength = 1)
Parameters: 0.014804 1.0933 0.76024 0.27015
Gradients: 0.83895 0.60996 -7.3230 3.3510
Iteration 10: Log-likelihood = -133.876720666 (steplength = 1)
Parameters: -0.034438 1.0938 0.46226 0.40605
Gradients: 6.1180 0.99752 10.528 6.9157
Iteration 11: Log-likelihood = -133.215756120 (steplength = 0.008)
Parameters: 0.014505 1.1018 0.54648 0.46137
Gradients: -0.55270 0.72059 -1.2327 0.43556
Iteration 12: Log-likelihood = -133.185323523 (steplength = 0.04)
Parameters: 0.010238 1.1316 0.52856 0.49641
Gradients: -0.25051 -0.14061 -0.71733 -0.28480
Iteration 13: Log-likelihood = -133.185060907 (steplength = 0.008)
Parameters: 0.011028 1.1308 0.52769 0.49709
Gradients: -0.36195 -0.11592 -0.65345 -0.28715
Iteration 14: Log-likelihood = -133.185045046 (steplength = 0.04)
Parameters: 0.0087095 1.1249 0.52597 0.50187
Gradients: -0.053499 0.085396 -0.65223 -0.41383
Iteration 15: Log-likelihood = -133.181789118 (steplength = 1)
Parameters: 0.0081963 1.1273 0.52065 0.49465
Gradients: -0.016645 -0.0072318 0.043542 -0.00045134
Iteration 16: Log-likelihood = -133.181774062 (steplength = 1)
Parameters: 0.0081232 1.1271 0.52128 0.49410
Gradients: -0.00074809 -0.00035241 -0.0012373 0.00028761
Iteration 17: Log-likelihood = -133.181774041 (steplength = 1)
Parameters: 0.0081164 1.1271 0.52126 0.49413
Gradients: 6.7268e-006 4.3038e-006-1.6833e-005-1.9424e-005
Iteration 17: Log-likelihood = -133.181774041 (steplength = 1)
Parameters: 0.0081164 1.1271 0.52126 0.49413
Gradients: 6.7268e-006 4.3038e-006-1.6833e-005-1.9424e-005
--- FINAL VALUES:
Log-likelihood = -133.181774041 (steplength = 6.4e-005)
Parameters: 0.0081164 1.1271 0.52126 0.49413
Gradients: 6.7268e-006 4.3038e-006-1.6833e-005-1.9424e-005
theta[0]: 0.00381941 (0.0406031)
theta[1]: 0.530368 (0.0834569)
theta[2]: 0.115430 (0.0264921)
theta[3]: 0.494127 (0.192625)
Function evaluations: 47
Evaluations of gradient: 17
Model 11: GARCH estimates using the 99 observations 1980:02-1988:04
Dependent variable: Y
Standard errors based on Hessian
VARIABLE COEFFICIENT STDERROR T STAT P-VALUE
const 0.00381941 0.0406031 0.094 0.92506
Y_1 0.530368 0.0834569 6.355 <0.00001 ***
alpha(0) 0.115430 0.0264921 4.357 0.00001 ***
alpha(1) 0.494127 0.192625 2.565 0.01031 **
Mean of dependent variable = 0.0519415
Standard deviation of dep. var. = 0.544934
Unconditional error variance = 0.22818
Log-likelihood = -58.5562
Akaike information criterion (AIC) = 127.112
Schwarz Bayesian criterion (BIC) = 140.088
Hannan-Quinn criterion (HQC) = 132.362
==== results by the command in console=====
? garch 0 1; Y Y(-1) --verbose
Automatic initialization of parameters
Regression coefficients:
theta[0] = 0.0511825
theta[1] = 1.08508
Variance parameters:
alpha[0] = 0.1
beta[0] = 0.9
Iteration 1: Log-likelihood = -270.911199903
Parameters: 0.11038 1.0851 0.10000 0.90000
Gradients: -125.16 -125.16 1010.7 30.388
Iteration 2: Log-likelihood = -170.537739660 (steplength = 0.0016)
Parameters: -0.089875 0.88483 1.7171 0.94862
Gradients: -25.749 -25.749 -8.7587 -8.4495
Iteration 3: Log-likelihood = -161.761522666 (steplength = 0.008)
Parameters: -0.29926 0.67545 1.6788 0.88214
Gradients: -11.183 -11.183 -10.840 -8.8715
Iteration 4: Log-likelihood = -158.391619816 (steplength = 0.04)
Parameters: -0.31167 0.66303 1.6896 0.51071
Gradients: -10.407 -10.407 -10.879 -8.1069
Iteration 5: Log-likelihood = -157.154680098 (steplength = 0.04)
Parameters: -0.31888 0.65583 1.6928 0.36062
Gradients: -10.000 -10.000 -10.621 -6.6638
Iteration 6: Log-likelihood = -155.742679661 (steplength = 0.2)
Parameters: -0.35479 0.61992 1.6858 0.22820
Gradients: -7.1755 -7.1755 -10.191 -3.8497
Iteration 7: Log-likelihood = -154.844791246 (steplength = 1)
Parameters: -0.48452 0.49018 1.6370 0.10641
Gradients: 5.3640 5.3640 -8.7104 3.1207
Iteration 8: Log-likelihood = -154.573226547 (steplength = 1)
Parameters: -0.45621 0.51850 1.6312 0.19907
Gradients: 2.1978 2.1978 -10.062 -2.3733
Iteration 9: Log-likelihood = -154.393792855 (steplength = 1)
Parameters: -0.41196 0.56275 1.6201 0.14415
Gradients: -2.0434 -2.0434 -9.3522 0.38397
Iteration 10: Log-likelihood = -153.968693027 (steplength = 1)
Parameters: -0.42290 0.55181 1.5786 0.15683
Gradients: -0.97944 -0.97944 -9.3680 0.21026
Iteration 11: Log-likelihood = -150.992406742 (steplength = 0.04)
Parameters: -0.46207 0.51263 1.2039 0.16524
Gradients: 3.4088 3.4088 -6.1548 7.6249
Iteration 12: Log-likelihood = -148.359103161 (steplength = 0.008)
Parameters: -0.43062 0.54409 0.75562 0.32000
Gradients: -2.3822 -2.3822 1.9263 6.8635
Iteration 13: Log-likelihood = -148.097032678 (steplength = 0.008)
Parameters: -0.45911 0.51560 0.71943 0.36973
Gradients: 0.75946 0.75946 1.8664 4.7097
Iteration 14: Log-likelihood = -147.949161098 (steplength = 1)
Parameters: -0.45122 0.52348 0.75142 0.44913
Gradients: -0.66911 -0.66911 -1.6394 -0.55287
Iteration 15: Log-likelihood = -147.921370789 (steplength = 1)
Parameters: -0.45617 0.51854 0.69072 0.47194
Gradients: -0.69320 -0.69320 0.78234 -0.15366
Iteration 16: Log-likelihood = -147.911096499 (steplength = 1)
Parameters: -0.46083 0.51388 0.71098 0.45901
Gradients: 0.15242 0.15242 -0.086460 0.020146
Iteration 17: Log-likelihood = -147.910849840 (steplength = 1)
Parameters: -0.45975 0.51495 0.70935 0.46012
Gradients: -0.0033672 -0.0033672 -0.010763 -0.0083438
Iteration 18: Log-likelihood = -147.910848410 (steplength = 1)
Parameters: -0.45979 0.51491 0.70914 0.46007
Gradients: 0.00068169 0.00068169 0.00046165 -0.00074704
Iteration 19: Log-likelihood = -147.910848385 (steplength = 1)
Parameters: -0.45979 0.51492 0.70917 0.46004
Gradients: 8.9567e-005 8.9567e-005 8.2642e-006 -0.00017120
Iteration 20: Log-likelihood = -147.910848385 (steplength = 0.008)
Parameters: -0.45979 0.51492 0.70917 0.46004
Gradients: 1.1075e-005 1.1075e-005 4.6432e-005 -0.00012421
Iteration 21: Log-likelihood = -147.910848385 (steplength = 0.008)
Parameters: -0.45979 0.51492 0.70917 0.46004
Gradients: 5.2761e-005 5.2761e-005 3.4521e-005-8.0059e-005
Iteration 22: Log-likelihood = -147.910848385 (steplength = 0.008)
Parameters: -0.45979 0.51492 0.70917 0.46004
Gradients: 4.1306e-005 4.1306e-005 5.3314e-006-9.2102e-005
Iteration 23: Log-likelihood = -147.910848385 (steplength = 1)
Parameters: -0.45978 0.51492 0.70917 0.46003
Gradients: -2.4463e-006-2.4463e-006 2.1778e-006 7.0214e-006
Iteration 23: Log-likelihood = -147.910848385 (steplength = 1)
Parameters: -0.45978 0.51492 0.70917 0.46003
Gradients: -2.4463e-006-2.4463e-006 2.1778e-006 7.0214e-006
--- FINAL VALUES:
Log-likelihood = -147.910848385 (steplength = 0.0016)
Parameters: -0.45978 0.51492 0.70917 0.46003
Gradients: -2.4463e-006-2.4463e-006 2.1778e-006 7.0214e-006
theta[0]: -0.216364 (0.171258)
theta[1]: 0.242311 (0.0635179)
theta[2]: 0.157041 (0.0421745)
theta[3]: 0.460033 (0.177473)
Function evaluations: 58
Evaluations of gradient: 23
Model 10: GARCH estimates using the 99 observations 1980:02-1988:04
Dependent variable: Y
Standard errors based on Hessian
VARIABLE COEFFICIENT STDERROR T STAT P-VALUE
const -0.216364 0.171258 -1.263 0.20645
Y_1 0.242311 0.0635179 3.815 0.00014 ***
alpha(0) 0.157041 0.0421745 3.724 0.00020 ***
alpha(1) 0.460033 0.177473 2.592 0.00954 ***
Mean of dependent variable = 0.0519415
Standard deviation of dep. var. = 0.544934
Unconditional error variance = 0.290835
Log-likelihood = -73.2853
Akaike information criterion (AIC) = 156.571
Schwarz Bayesian criterion (BIC) = 169.546
Hannan-Quinn criterion (HQC) = 161.82
12:32 p.m.
On Wed, 21 May 2008, yinung CYCU wrote:
Dear all
I use gretl 1.7.4 to estimate a GARCH model by using default menu
function (i.e., \Model\Time series\GARCH ) and by the command in
console. But they generate different results. The details of iterations
are as follows. Any idea?
Could you please check what happens by issuing the command
garch 0 1; Y const Y(-1) --verbose
via the console? If my guess is right, you should get the same estimates
as the menu function. The trick is: you must specify the constant among
the regressors for the mean equation. We don't support garch models withut
a constant term yet. If you don't put the constant in, we should flag an
error, which we don't.
Riccardo (Jack) Lucchetti
Dipartimento di Economia
Università Politecnica delle Marche
r.lucchetti(a)univpm.it
http://www.econ.univpm.it/lucchetti
12:49 p.m.
Dear Jack
You are right. I just tried in console:
garch 0 1; Y const Y(-1) --verbose
and I got the same results as what I had in default menu function.
Thanks
Yi-Nung Yang
2008/5/21 Riccardo (Jack) Lucchetti <r.lucchetti(a)univpm.it>:
On Wed, 21 May 2008, yinung CYCU wrote:
Dear all
>
>
> I use gretl 1.7.4 to estimate a GARCH model by using default menu function
> (i.e., \Model\Time series\GARCH ) and by the command in console. But they
> generate different results. The details of iterations are as follows. Any
> idea?
>
Could you please check what happens by issuing the command
garch 0 1; Y const Y(-1) --verbose
via the console? If my guess is right, you should get the same estimates as
the menu function. The trick is: you must specify the constant among the
regressors for the mean equation. We don't support garch models withut a
constant term yet. If you don't put the constant in, we should flag an
error, which we don't.
Riccardo (Jack) Lucchetti
Dipartimento di Economia
Università Politecnica delle Marche
r.lucchetti(a)univpm.it
http://www.econ.univpm.it/lucchetti
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
3:41 p.m.
On Wed, 21 May 2008, yinung CYCU wrote:
Dear Jack
You are right. I just tried in console:
garch 0 1; Y const Y(-1) --verbose
and I got the same results as what I had in default menu function.
Thanks
I just committed a fix to CVS: if missing, a constant is inserted
automatically, but in the wrong place. Now fixed.
Riccardo (Jack) Lucchetti
Dipartimento di Economia
Università Politecnica delle Marche
r.lucchetti(a)univpm.it
http://www.econ.univpm.it/lucchetti
7:49 a.m.
New subject: year and period
I would like to save in two strings the year and quarter of the last
observation of a quarterly series. I got to obtain the quarter but I don't
know how to save the year.
In the script below, the string @send contains the last observation = 1993:4
and the string @per contains the quarter =4. Any ideas about how to obtain the
year?
<script>
open data9-3
scalar endobs=lastobs(reskwh)
sprintf send "%s", date(endobs)
string per = @send + 5
print "@send"
print "@per"
</script>
Note for the programmers: I suggest two new commands, year() and period(),
similar to date() to obtain these strings.
--
Ignacio Diaz-Emparanza
DEPARTAMENTO DE ECONOMÍA APLICADA III (ECONOMETRÍA Y ESTADÍSTICA)
UPV/EHU
Avda. Lehendakari Aguirre, 83 | 48015 BILBAO
T.: +34 946013732 | F.: +34 946013754
www.et.bs.ehu.es
--
Para leer y escribir documentos en formato "open document"
(.ods, .odt, .odb, odp) utilice OpenOffice.org o también
MS Office, instalando el "plugin":
http://www.sun.com/software/star/odf_plugin
8:25 a.m.
New subject: year and period
On Fri, 23 May 2008, Ignacio Diaz-Emparanza wrote:
I would like to save in two strings the year and quarter of the last
observation of a quarterly series. I got to obtain the quarter but I don't
know how to save the year.
In the script below, the string @send contains the last observation = 1993:4
and the string @per contains the quarter =4. Any ideas about how to obtain the
year?
<script>
open data9-3
scalar endobs=lastobs(reskwh)
sprintf send "%s", date(endobs)
string per = @send + 5
print "@send"
print "@per"
</script>
If you're using CVS or the snapshot, you can use one of the latest bright
inventions by Allin, that is sscanf:
<script>
open data9-3
scalar endobs=lastobs(reskwh)
sprintf send "%s", date(endobs)
scalar yr
scalar qt
sscanf send "%d:%d", yr, qt
</script>
Note for the programmers: I suggest two new commands, year() and
period(),
similar to date() to obtain these strings.
I have considered this possibility many times, but there is a
complication: we don't always have data in the year:period format. We may
have cross-section data, daily data, panel data. I agree in principle that
functions like the ones you propose are useful, but the syntax will need
to be more general.
Riccardo (Jack) Lucchetti
Dipartimento di Economia
Università Politecnica delle Marche
r.lucchetti(a)univpm.it
http://www.econ.univpm.it/lucchetti
9:32 a.m.
New subject: year and period
El Friday 23 May 2008 14:25:41 Riccardo (Jack) Lucchetti escribió:
If you're using CVS or the snapshot, you can use one of the latest bright
inventions by Allin, that is sscanf:
<script>
open data9-3
scalar endobs=lastobs(reskwh)
sprintf send "%s", date(endobs)
scalar yr
scalar qt
sscanf send "%d:%d", yr, qt
</script>
Thank you, Jack. This works ok. I needed this variables as strings so I only
had to add the lines
sprintf year "%4f", yr
sprintf quarter "%2f", qt
to your script. Now the numbers are saved in the strings @year and @quarter
--
Ignacio Diaz-Emparanza
DEPARTAMENTO DE ECONOMÍA APLICADA III (ECONOMETRÍA Y ESTADÍSTICA)
UPV/EHU
Avda. Lehendakari Aguirre, 83 | 48015 BILBAO
T.: +34 946013732 | F.: +34 946013754
www.et.bs.ehu.es
9:43 a.m.
New subject: year and period
On Fri, 23 May 2008, Ignacio Diaz-Emparanza wrote:
Thank you, Jack. This works ok. I needed this variables as
strings so I only had to add the lines
sprintf year "%4f", yr
sprintf quarter "%2f", qt
If you just want year and quarter in string form:
<script>
open data9-3
scalar endobs=lastobs(reskwh)
sprintf send "%s", date(endobs)
sprintf year "%.4s", send
sprintf quarter "%s", send + 5
printf "year is %s and quarter is %s\n", year, quarter
</script>
or
<script>
scalar endobs=lastobs(reskwh)
sprintf send "%s", date(endobs)
string quarter, year
sscanf send, "%4s:%s", year, quarter
</script>
Allin
12:13 p.m.
New subject: year and period
El Friday 23 May 2008 15:43:24 Allin Cottrell escribió:
<script>
scalar endobs=lastobs(reskwh)
sprintf send "%s", date(endobs)
string quarter, year
sscanf send, "%4s:%s", year, quarter
</script>
Allin
Good!, you win the prize to the smallest size code.
Thanks,
--
Ignacio Diaz-Emparanza
DEPARTAMENTO DE ECONOMÍA APLICADA III (ECONOMETRÍA Y ESTADÍSTICA)
UPV/EHU
Avda. Lehendakari Aguirre, 83 | 48015 BILBAO
T.: +34 946013732 | F.: +34 946013754
www.et.bs.ehu.es
5908
days inactive
5911
days old
8 comments
4 participants
participants (4)
-
Allin Cottrell -
Ignacio Diaz-Emparanza -
Riccardo (Jack) Lucchetti -
yinung CYCU