NK_CGG02AL not working (edit: NOT CGG99AL!)

[gboehl]

Describe the bug
NK_CGG99AL NK_CGG02AL gives problem with Cholesky decomposition. See #82 (comment) and #82 (comment) for details.

@AlexDece as far as I understand, oo_.dr.jacobia is created by the AL routines. Apparently it fails somewhere on the way.

@millazar @btatar13 does either of you have a deeper understanding of these routines? If nobody knows how to fix it I would just remove the model for now, as it appears that it was never working (hence not our responsibility).

[JTWendelborn]

To add to the confusion: it works for me (at least in the developer version) with both, Matlab 2019a and Octave 5.1 and Dynare 4.5.7...I used the default gain parameter setting of 0.01 and all states selected and tried all available rules with a MP shock.

[j2L4e]

We've encountered this before. (From conversation with fstrobel, sorry for citing the german text)

Jan
ich krieg die mmb 2.3 auch nicht ans laufen ('Reference to non-existent field 'jacobia'.')

Felix
Jau, das leigt an einem name-conflict. In der MMB 2.3-Version gibt es im Ordner ALTOOL ein paar für die Zwecke der AL-Simulationen modifizierte Standard-files von Dynare, die aber damals aus unerfindlichen Gründen nicht umbenannt wurden, so dass sie den gleichen Namen tragen, wie die nicht modifizierten Funktionen. Ehrlich gesagt, weiss ich nicht, warum das früher nicht zu Konflikten geführt hat. Ich schätze dieser name-conflict kommt nur bei Octave zu tragen. In der 3.0 Version habe ich die entsprechenden Files umbenannt (Suffix "_MMB") und die AL-Simulationen laufen wieder

[AlexDece]

This is what I encountered until now.

The model also works fine for me (in dev version, but also in the version for testing, using ctave 4.4.1, dynare 4.5.7, windows 10).

When I was looking into the code of MMB 2.3 I find the same ting Felix says.
So I focused on MB 3.1

Stoch_simul_MMB.m calls resol_MMB.m
resol_MMB calls dr_block.m (which I didnt find, but there exists dr1_AL.m)
or stochastic_solvers_MMB.m

stochastic_solvers_MMB.m has its main function called stochastic_solvers()
However, this shouldnt be a problem, since matlab takes the file name as function name and the file name is used when calling the function.

In line 227 of stochastic_solvers_MMB.m, it says
dr.jacobia=jacobia_; (remember)

stoch_simul_MMB.m calls TRFromBeta.m
In line 15 it says
jacobia = oo_.dr.jacobia;
But this does not exist (according to MMB 2.3, where 'Reference to non-existent field 'jacobia'.)

Is it possible that this problem can be solved by adding
oo_.dr.jacobia=dr.jacobia;
after line 227 in stochastic_solvers_MMB.m? But then, this problem should occur for every AL model, where options are specified such that stochastic_solvers_MM.m is called in resol_MMB.m.

It is very strange, since the model, like other AL models, is running through and delivers IRFs...

[gboehl]

Thanks for looking into that Alex. I also don't get it.

If it would be a naming issue ("ALTOOL" vs "ALTOOL_MMB") the problem would be with all models.

I renamed file references and folder anyways, the problem is then:

chol: input matrix must be positive definite
sim_AL_alt_gain.m, function 'sim_AL_alt_gain', line 40

stoch_simul_MMB.m, function 'stoch_simul_MMB', line 184

run_dynare_and_simulate.m, function 'run_dynare_and_simulate', line 15

run_model_with_rules.m, function 'run_model_with_rules', line 29

mmb.m, function 'mmb', line 35

Explicitely in sim_AL_alt_gain.m:

 Q = M_.Sigma_e + 1.e-14;
RQ = R * chol(Q);

Then its a problem with this Q object.
So I did not look at Q to see how ill conditioned it is, and I don't know what this M_.Sigma_e object is, but this looks much like a problem with the model now. Weird thing remains that it works for Jonas. As he is using the same version of Octave, the Cholesky decomposition should be the same.

I would just remove the model for now, as we don't have the time to have a closer look.

@AlexDece if you want to, you can have one last look at it. Otherwise I'd just put it in an alternative models_nonfunctional folder within mmci-cli.

[gboehl]

@JTWendelborn @AlexDece one quick thing to check is how this M_.Sigma_e looks for each of you. If its different it would be interesting to know why. If not, then 😠

Does Q look different for Matlab and Octave?

@JTWendelborn does the compatibility test run trough for you with Octave?

[JTWendelborn]

My M_.Sigma_e looks as follows and is the same for Matlab and Octave:

[j2L4e]

@JTWendelborn You can run the tests from the command line via

test('test/config.test.json', 'test/config.test-al.json');

after changing the dynare path in those test files.

[JTWendelborn]

Running this command from the command line strangely gives me a syntax error:
-bash: syntax error near unexpected token 'test/config.test.json'.
However, when I run the compatibility test (Octave 5.1, Dynare 4.5.7) from the developer GUI, it reports that one of the tests has failed.

[AlexDece]

My M_.Sigma_e looks like what Jonas sent above. Am Mi., 4. Dez. 2019 um 18:13 Uhr schrieb JTWendelborn < notifications@github.com>:

…

Running this command from the command line strangely gives me a syntax error. However, when I run the compatibility test (Octave 5.1, Dynare 4.5.7) from the developer GUI, it reports that one of the tests has failed. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#86?email_source=notifications&email_token=ANVSGFLIOZGIWKUZ3OROW6LQW7QKXA5CNFSM4JU4I672YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEF5YB3Y#issuecomment-561742063>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ANVSGFOTZVTOQLREQ3MN733QW7QKXANCNFSM4JU4I67Q> .

[gboehl]

@JTWendelborn you should run it from within octave, not bash. It is important that we get these details because it helps us to figure out what is wrong with the tests and/or octave (while the GUI is not very informative).

Running this command from the command line strangely gives me a syntax error:
-bash: syntax error near unexpected token 'test/config.test.json'.
However, when I run the compatibility test (Octave 5.1, Dynare 4.5.7) from the developer GUI, it reports that one of the tests has failed.

[JTWendelborn]

Sorry, my bad. I ran it from within Octave and got the following result, which seems to be the same that Alex reported in #82 :

       ---OUTPUT-START---

--- Autocorrelation ---
expected -0.071, actual -0.071
success
--- Unconditional variances ---
expected 1.1702, actual 1.1702
success
--- Input response functions ---
expected 0.0415, actual 0.0415
success
--- [AL] Autocorrelation ---
expected 0.7396, actual 0.777
failed
--- [AL] Unconditional variances ---
expected 8.2656, actual 15.9088
failed
--- [AL] Input response functions ---
expected -0.6807, actual -0.6807
success
4 out of 6 successful

I can try to install a different version of Octave/Dynare as my current setup is not officially compatible...

[gboehl]

Great, then we're all on the same page. Lets keep the compatibilty test in #82 and deal with NK_CGG99AL here (although it seems to me as if they are related).

@JTWendelborn Can you please confirm again that NK_CGG99AL works for you in Octave 5.1? This really puzzles me.