Dears, for a simulation study I want to repeat the estimate of a state space model /N/ times. Repeating the estimate is not a problem, but if I call the ksmooth function after each estimate (MLE block), gretl allows me it on the first iteration, but not the following ones. It appears that the function cannot find the bundle in subsequent iterations This happens if the repetitions occur in a single script run. Obviously it doesn't happen if every rep has its own run. It's really strange! Is it a bag or am I missing something?

> ksmooth(&mod)

On Tue, 25 Jan 2022, Riccardo (Jack) Lucchetti wrote: > On Tue, 25 Jan 2022, Paolo Chirico wrote: > >> Dears, >> for a simulation study I want to repeat the estimate of a state space model >> /N/ times. Repeating the estimate is not a problem, but if I call the >> ksmooth function after each estimate (MLE block), gretl allows me it on the >> first iteration, but not the following ones. >> It appears that the function cannot find the bundle in subsequent >> iterations >> This happens if the repetitions occur in a single script run. >> Obviously it doesn't happen if every rep has its own run. >> It's really strange! >> Is it a bag or am I missing something? > > It sounds like a bug, Paolo, thanks for the report. > > Could you send us an example script exposing the problem? This script runs without problems here, but I'm running current git: in the next version (due real soon) the whole Kalman apparatus has been revised quite thouroughly. Or maybe I misunderstood the problem. Paolo, could you check if this script runs on your machine? <hansl> set verbose off nulldata 100 setobs 1 1 --special-time-series loop i = 1 .. 4 # ma(1) example with known variance e = normal() theta = 0.8 y = e + theta * e(-1) smpl +1 ; par = 0.8 T = {0,0;1,0} Q = {1,0;0,0} mod = ksetup(y, {1; par}, T, Q) mle ll = misszero(llik) mod.obsymat[2] = par kfilter(&mod) series llik = mod.llt params par end mle --hessian ksmooth(&mod) series e$i = mod.state[,1] smpl full endloop </hansl> ------------------------------------------------------- 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)gretlml.univpm.it To unsubscribe send an email to gretl-users-leave(a)gretlml.univpm.it Website: https://gretlml.univpm.it/postorius/lists/gretl-users.gretlml.univpm.it/

Thanks, Artur, this is very helpful. It looks as if we had some sort of bug that got fixed as a side-effect by the recent Kalman rewrite. We were planning to have the new version out by the end of the week anyway. Could you please try my script on a recent snapshot to see if my conjecture holds?

Hi, I have installed the snapshot version. If I try to load an old simulation session, gretl suddenly quits. This only happens with old sessions where there was a bundle of a state space model.