search

IMFS-MMB / mmb-gui-electron

Electron-vue based GUI of the Macroeconomic Model Database
http://www.macromodelbase.com
Other
7 stars 16 forks source link

ACF of NK_IR04AL only few periods #121

Closed AlexDece closed 4 years ago

AlexDece commented 4 years ago

Describe the bug Autocorrelation function of AL model NK_IR04AL is shorter than the periods determined before simulation. This holds for different period lengths and ALL monetary policy rules. ACF of NK_IR04 is fine. ACF of other AL models is fine (I tested NK_LWW03AL)

To Reproduce Steps to reproduce the behavior:

  1. Simulate NK_IR04AL (check ACF)
  2. See error

Expected behavior ACF of all periods

Screenshots nk_al_1 nk_al_2 nk_al_3

Desktop (please complete the following information): Windows 10 Octave 4.4.1 Dynare 4.5.7

gboehl commented 4 years ago

Hm. True.

From the output.json:

...
    "data": {
        "AC": {
            "y": [1,0.940749575,0.8894850753,0.8450880773,0.803982271,0.7612268621,0.727444386,0.6945216243,0.6602946995,0.6305984967,0.604200661,0.5790707375,"_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_","_NaN_"],
...

It is clear that the AL method can lead to numerical problems/explosive behavior/etc depending on the setup (states & gain). For a different level of gain and/or differnt states everythink looks fine.

So this behavior is somewhat expected as it is left to the user to chose states + gain accordingly.