Dear list members, I am trying to estimate several quantile regressions, each with a constant and a single regressor, on a rolling basis (497 steps). The quantile vector is set to q={0.1, 0.25, 0.5, 0.75, 0.90) yielding 5 quantile regressions per step and individual (1317). Using a simple weighting scheme, where the weights sum to 1, the estimated conditional quantiles ("y-hats") are aggregated so that this yield one estimate per step. Overall, I have 16 regressors and at the end, I intend to obtain one matrix for each regressor. At the end of the day, this should amount to 497 (steps) * 1317(indiviuals) * 16(regressors) * 5(quantiles) = 5,236,392,0 quantile regressions. Gretl works quite well for some time, but breaks down (i.e. does not react any more) somewhere during the process. I suppose this occurs at the end, prior to returning the script output, which is quite lengthy (I do use --quiet option after quantreg command). I tried to eliminate the 16^th regressors, after which Gretl broke down the first time and redid the estimation using the other 15 regressors. This time, I saved the matrices using mwrite()-function. After the 15^th round, Gretl broke down again. I opend the matrix number 15 in excel and saw that all rows and columns were filled with either data or, if the y-series was to short, with "nan". Is there an option to tell Gretl to suppress the entire script output or save it as .txt-file instead? I am not sure, maybe it is an issue with my data, because system performance (RAM 8GB) is not supposed to be the limiting factor, is it? Another possibility is that there is a mistake in my script. But you already were so kind, to hint, that, last time, I forget an endloop. This time I also counted my loop - endloop combinations (and they match). Thanks in advance. Best regard, Jan _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users

Dear list members, I am trying to estimate several quantile regressions, each with a constant and a single regressor, on a rolling basis (497 steps). The quantile vector is set to q={0.1, 0.25, 0.5, 0.75, 0.90) yielding 5 quantile regressions per step and individual (1317). Using a simple weighting scheme, where the weights sum to 1, the estimated conditional quantiles ("y-hats") are aggregated so that this yield one estimate per step. Overall, I have 16 regressors and at the end, I intend to obtain one matrix for each regressor. At the end of the day, this should amount to 497 (steps) * 1317(indiviuals) * 16(regressors) * 5(quantiles) = 5,236,392,0 quantile regressions.

scalar ikr=1+i endloop

Jack, thank you as well. I have executed my script with the 16th regressor only. It ran through all the loops (497 steps) for 1317 funds without freezing after finishing as well as opening the script output. This lends support to my first assumption, that it is the script output, that might cause the problem. I am not able, to check that instantaneously, because, estimating the regressions for the first 15 regressors, took the computer more than three hours. Jan -----Ursprüngliche Nachricht----- Von: Plus.line MailSystem [mailto:cyrus@mailer.plusline.de] Im Auftrag von Riccardo (Jack) Lucchetti Gesendet: Donnerstag, 7. Januar 2016 16:08 An: Gretl list <gretl-users(a)lists.wfu.edu&gt; Betreff: Re: [Gretl-users] gretl crashes during quantile regression On Thu, 7 Jan 2016, Jan Tille wrote: > Dear list members, > > I am trying to estimate several quantile regressions, each with a > constant and a single regressor, on a rolling basis (497 steps). The > quantile vector is set to q={0.1, 0.25, 0.5, 0.75, 0.90) yielding 5 > quantile regressions per step and individual (1317). Using a simple > weighting scheme, where the weights sum to 1, the estimated > conditional quantiles ("y-hats") are aggregated so that this yield > one estimate per step. > > Overall, I have 16 regressors and at the end, I intend to obtain one > matrix for each regressor. At the end of the day, this should amount > to > 497 (steps) * 1317(indiviuals) * 16(regressors) * 5(quantiles) = > 5,236,392,0 quantile regressions. [...] The only piece of advice I have for you at this stage, is: try reducing the size of your problem by cutting down the number of individuals and/or the number of steps, if possible. If the problem persists even with a much smaller combination, then the problem is most likely with gretl and we'll be able to diagnose what the problem is. If the problem goes away, perhaps the data you dropped contain some oddness. By some trial and error you should be able to isolate the offending observations and, again, we should be able to trace the cause(s) of your problem. ------------------------------------------------------- Riccardo (Jack) Lucchetti Dipartimento di Scienze Economiche e Sociali (DiSES) Università Politecnica delle Marche (formerly known as Università di Ancona) r.lucchetti(a)univpm.it http://www2.econ.univpm.it/servizi/hpp/lucchetti ------------------------------------------------------- _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users

Artur, yes, --quiet suppresses a lot. However, after running through a script, Gretl tries to open the script output, like the sample one ? loop for i=1..10 > scalar ikr=1+i > endloop loop: i = 1 ? scalar ikr=1+i loop: i = 2 ? scalar ikr=1+i loop: i = 3 [...] Number of iterations: 10 Nonetheless with 5 iterations for the quantreg: loop k=1..5 16 iterations for the models: loop m=1..16 497 iterations for the rolling steps: loop t=104..601 and 1317 iterations for the individuals i=1..1317 the above output would be quite lengthy. In this case, it would be great, if you could a command, prior to the rest of the script, that tells Gretl, to save that verbose text in a text-file, instead of opening it. To run the script, I usually mark all the lines and click on "run region". I am not sure if that answers your question. Best, Jan -----Ursprüngliche Nachricht----- Von: Plus.line MailSystem [mailto:cyrus@mailer.plusline.de] Im Auftrag von Artur Tarassow Gesendet: Donnerstag, 7. Januar 2016 16:24 An: gretl-users(a)lists.wfu.edu Betreff: Re: [Gretl-users] gretl crashes during quantile regression Which "output" do you actually mean? You are aware of "--quiet" option for the quantreg command, right? This already suppresses quite a lot of the estimation output... Best, Artur Am 07.01.2016 um 16:20 schrieb Jan Tille: > Jack, > > thank you as well. I have executed my script with the 16th regressor only. It ran through all the loops (497 steps) for 1317 funds without freezing after finishing as well as opening the script output. > > This lends support to my first assumption, that it is the script output, that might cause the problem. > I am not able, to check that instantaneously, because, estimating the regressions for the first 15 regressors, took the computer more than three hours. > > Jan > > -----Ursprüngliche Nachricht----- > Von: Plus.line MailSystem [mailto:cyrus@mailer.plusline.de] Im > Auftrag von Riccardo (Jack) Lucchetti > Gesendet: Donnerstag, 7. Januar 2016 16:08 > An: Gretl list <gretl-users(a)lists.wfu.edu&gt; > Betreff: Re: [Gretl-users] gretl crashes during quantile regression > > On Thu, 7 Jan 2016, Jan Tille wrote: > >> Dear list members, >> >> I am trying to estimate several quantile regressions, each with a >> constant and a single regressor, on a rolling basis (497 steps). The >> quantile vector is set to q={0.1, 0.25, 0.5, 0.75, 0.90) yielding 5 >> quantile regressions per step and individual (1317). Using a simple >> weighting scheme, where the weights sum to 1, the estimated >> conditional quantiles ("y-hats") are aggregated so that this yield >> one estimate per step. >> >> Overall, I have 16 regressors and at the end, I intend to obtain one >> matrix for each regressor. At the end of the day, this should amount >> to >> 497 (steps) * 1317(indiviuals) * 16(regressors) * 5(quantiles) = >> 5,236,392,0 quantile regressions. > > [...] > > The only piece of advice I have for you at this stage, is: try reducing the size of your problem by cutting down the number of individuals and/or the number of steps, if possible. If the problem persists even with a much smaller combination, then the problem is most likely with gretl and we'll be able to diagnose what the problem is. If the problem goes away, perhaps the data you dropped contain some oddness. By some trial and error you should be able to isolate the offending observations and, again, we should be able to trace the cause(s) of your problem. > > > ------------------------------------------------------- > Riccardo (Jack) Lucchetti > Dipartimento di Scienze Economiche e Sociali (DiSES) > > Università Politecnica delle Marche > (formerly known as Università di Ancona) > > r.lucchetti(a)univpm.it > http://www2.econ.univpm.it/servizi/hpp/lucchetti > ------------------------------------------------------- > > _______________________________________________ > Gretl-users mailing list > Gretl-users(a)lists.wfu.edu > http://lists.wfu.edu/mailman/listinfo/gretl-users > _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users

On Thu, 7 Jan 2016, Jan Tille wrote:

[...] The only piece of advice I have for you at this stage, is: try reducing the size of your problem by cutting down the number of individuals and/or the number of steps, if possible. If the problem persists even with a much smaller combination, then the problem is most likely with gretl and we'll be able to diagnose what the problem is. If the problem goes away, perhaps the data you dropped contain some oddness.