March 2011 - Gretl-users - gretlml.univpm.it

by Allin Cottrell

On Sat, 5 Mar 2011, Maria Dolores Montoya Diaz wrote: > I�d like to compare results of censored model (tobit) with OLS regression. > For this, I�ve prepared the following hypothetical data, that I�ve copied > below. The censoring point for y is 1000 [...] > > And I�d like to estimate intreg with upper limit of 1000 (equivalent to > Stata�s command: tobit YCENS X, ul(1000)). This doesn't sound like a case for interval regression. But you can easily get tobit to do the job; just transform Y so it's censored at 0 instead of at 1000. series TY = 1000 - Y TY = (TY < 0)? 0 : TY series TX = -X tobit TY 0 TX The coefficient reported for TX is the same as that reported by Stata for tobit ycens x, ul(1000). > Additional question: Is there any equivalent to Stata�s command: truncreg > YTRUNC XTRUNC, ul(1000)? There's no built-in equivalent to "truncreg", no. Allin Cottrell

14 years

3
2
0 / 0

Suggestion for GRETL website: Numerical Accuracy

by Charles Koss

Dear members: For your consideration, I would like to suggest the inclusion of Dr. Yalta's paper in GRETL's website. Yalta, A. Talha, A. Yasemin Yalta. 2010. Should Economists Use Open Source Software for Doing Research? Computational Economics. 35(4):371-394. Link: http://ideas.repec.org/a/kap/compec/v35y2010i4p371-394.html in the section about "Numerical accuracy" is the most appropiate to put it on. The paper contains valuable information that is worthy of sharing with our community. Cheers, CK -- Charles Koss http://charlesonnet.blogspot.com/ On Mon, Apr 19, 2010 at 1:53 AM, Talha Yalta <talhayalta(a)gmail.com> wrote: > Dear gretl users: > > You might be interested to hear about my new paper entitled "Should > Economists Use Open Source Software for Doing Research?" published > this month in Computational Economics. The paper investigates > econometric software reliability and advocates the use of open source > software by taking gretl as a case study and showing how responsive > and transparent its development process is. I think many people here > might find it an interesting read. > > More information and the download link is available here: > http://ideas.repec.org/a/kap/compec/v35y2010i4p371-394.html > > I can send a working paper version if you do not have access to the above. > Cheers > A. Talha Yalta > > > > -- > “Remember not only to say the right thing in the right place, but far > more difficult still, to leave unsaid the wrong thing at the tempting > moment.” - Benjamin Franklin (1706-1790) > -- > > _______________________________________________ > Gretl-users mailing list > Gretl-users(a)lists.wfu.edu > http://lists.wfu.edu/mailman/listinfo/gretl-users >

14 years

2
1
0 / 0

convergence tolerance

by Sam Sam

Dear all: If the convergence tolerance of 1.00E-2 is too high. What is the reasonable range of convergence tolerance that ARIMA model estimated is not misspecified? Thanks a lot

14 years

1
0
0 / 0

(no subject)

by & $

Date: Fri, 4 Mar 2011 18:44:24 -0500 (EST) From: Allin Cottrell < cottrell(a)wfu.edu > Subject: Re: [Gretl-users] ARIMA(0,0,1)(0,0,1) forecasts To: Gretl list < gretl-users(a)lists.wfu.edu > Message-ID: < Pine.A41.4.58.1103041841360.192740(a)f1n11.sp2net.wfu.edu

14 years

1
0
0 / 0

Re: [Gretl-users] convergence tolerance

by Allin Cottrell

On Sun, 6 Mar 2011, Sam Sam wrote: > When I estimate ARIMA by conditional maximum likelihood, > the convergence tolerance I choose is 1.00E-2 > (max interations is 500 ) > Is it too high? That is certainly an unsually "sloppy" tolerance level. If you don't get convergence with a tolerance a few orders of magnitude smaller, I would suspect misspecification. You can try the --verbose switch to the arma command to see what's going on in more detail. You can also try exact ML, with or without the --fcp switch. And there's always the --x-12-arima option for a second opinion. Allin Cottrell

14 years

1
0
0 / 0

convergence tolerance

by Sam Sam

Dear all: When I estimate ARIMA by conditional maximum likelihood, the convergence tolerance I choose is 1.00E-2 (max interations is 500 ) Is it too high? What range of converge tolence is too high and the model estimated may be misspecified?

14 years

1
0
0 / 0

Paper advocating open source software and gretl

by Talha Yalta

Dear gretl users: You might be interested to hear about my new paper entitled "Should Economists Use Open Source Software for Doing Research?" published this month in Computational Economics. The paper investigates econometric software reliability and advocates the use of open source software by taking gretl as a case study and showing how responsive and transparent its development process is. I think many people here might find it an interesting read. More information and the download link is available here: http://ideas.repec.org/a/kap/compec/v35y2010i4p371-394.html I can send a working paper version if you do not have access to the above. Cheers A. Talha Yalta -- “Remember not only to say the right thing in the right place, but far more difficult still, to leave unsaid the wrong thing at the tempting moment.” - Benjamin Franklin (1706-1790) --

14 years

3
3
0 / 0

intreg doubt

by Maria Dolores Montoya Diaz

Hi everyone: I´d like to compare results of censored model (tobit) with OLS regression. For this, I´ve prepared the following hypothetical data, that I´ve copied below. The censoring point for y is 1000. I´d estimated the following models: ols Y const X ols YCENS const X And I´d like to estimate intreg with upper limit of 1000 (equivalent to Stata´s command: tobit YCENS X, ul(1000)). But, I´ve problems to define the variables minvar and maxvar. The model do not converge with Y_cens_minvar and Ycens_maxvar and X as independent variable. Are they correct? Additional question: Is there any equivalent to Stata´s command: truncreg YTRUNC XTRUNC, ul(1000)? Thanks in advance for any help. Maria Dolores Montoya Diaz The database is: Y X YCENS YTRUNC XTRUNC Ycens_minvar Ycens_maxvar 1 150 2 150 150 2 150 2 150 4 150 150 4 150 3 160 4 160 160 4 160 4 192 4 192 192 4 192 5 230.4 4 230.4 230.4 4 230.4 6 276.48 3 276.48 276.48 3 276.48 7 331.776 3 331.776 331.776 3 331.776 8 398.131 4 398.131 398.1312 4 398.1312 9 477.757 4 477.757 477.7574 4 477.7574 10 573.309 4 573.309 573.3089 4 573.3089 11 687.971 6 687.971 687.9707 6 687.9707 12 825.565 6 825.565 825.5649 6 825.5649 13 990.678 6 990.678 990.6778 6 990.6778 14 1188.813 8 1000 1000 15 1426.576 8 1000 1000 16 1711.891 8 1000 1000 17 1700 8 1000 1000 18 1700 8 1000 1000 19 1500 7 1000 1000 20 1200 6 1000 1000 21 1440 7 1000 1000 22 1728 8 1000 1000 23 2073.6 8 1000 1000 24 2488.32 8 1000 1000 25 2985.984 8 1000 1000 26 3583.181 18 1000 1000 27 4299.817 18 1000 1000 28 210 4 210 210 4 210 29 210 4 210 210 4 210 30 224 4 224 224 4 224 31 268.8 4 268.8 268.8 4 268.8 32 322.56 4 322.56 322.56 4 322.56 33 387.072 6 387.072 387.072 6 387.072 34 464.486 7 464.486 464.4864 7 464.4864 35 557.384 8 557.384 557.3837 8 557.3837 36 668.86 9 668.86 668.8604 9 668.8604 37 802.632 11 802.632 802.6325 11 802.6325 38 963.159 11 963.159 963.159 11 963.159 39 1155.791 15 1000 1000 40 1386.949 15 1000 1000 41 1664.339 15 1000 1000 42 1997.207 15 1000 1000 43 2396.648 15 1000 1000 44 2380 15 1000 1000 45 2380 14 1000 1000 46 2100 15 1000 1000 47 1680 15 1000 1000 48 2016 15 1000 1000 49 2419.2 15 1000 1000 50 2903.04 16 1000 1000

14 years

1
0
0 / 0

Re: [Gretl-users] ARIMA(0,0,1)(0,0,1) forecasts

by Allin Cottrell

On Fri, 4 Mar 2011, Allin Cottrell wrote: > On Fri, 4 Mar 2011, & $ wrote: > > > I'm having trouble understanding the way that Gretl computes > > forecasts for ARIMA(0,0,1)(0,0,1) models... OK, I think I have found the problem, and there's a fix in CVS and the current snapshots for Windows and OS X. Please test and see if it now appears to be right. Thanks to Chris ("& S") for the report: we don't want gretl to be giving false values in forecasts. Allin Cottrell

14 years

1
0
0 / 0

Re: [Gretl-users] ARIMA(0,0,1)(0,0,1) forecasts

by Allin Cottrell

On Fri, 4 Mar 2011, & $ wrote: > I'm having trouble understanding the way that Gretl computes > forecasts for ARIMA(0,0,1)(0,0,1) models. If I generate a random > data set and fit this model (using either algorithm) to get the > estimates theta_1 and Theta_1, and then generate a forecast, > then I would expect the forecast to be computed using the > polynomial > > 1 + theta_1*L + Theta_1*L^4 + theta_1*Theta_1*L^5 > > [... But] after a little experimenting, I have found that if I > use the polynomial > > 1 + theta_1*L + Theta_1*L^4 - theta_1*Theta_1*L^5 > > I get exactly the same forecast as Gretl. There seems to be something wrong with out-of-sample forecasts for seasonal MA models, you're right. I'll report back once I've figured out what's happening here. Allin Cottrell

14 years

1
0
0 / 0

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

Gretl-users March 2011