Page 1 of 1

building mean-fan charts using stochastic model

Posted: Wed Mar 31, 2021 2:56 am
by QIO
hi there
I am testing various ARIMAX auto-selected models and their applicability for producing benchmark models. For this I have also solved models via the stochastic option to produce fan charts (using simulation). To produce the forecast with standard error bounds one can simply add to the forecast trajectory (in the non panel worksheet) to the projected variable (see below)
YMP_REPIHSG_ML_US_M_0M
YMP_REPIHSG_ML_US_M
YMP_REPIHSG_ML_US_M_0M+2*YMP_REPIHSG_ML_US_M_0S
YMP_REPIHSG_ML_US_M_0M+YMP_REPIHSG_ML_US_M_0S
YMP_REPIHSG_ML_US_M_0M-YMP_REPIHSG_ML_US_M_0S
YMP_REPIHSG_ML_US_M_0M-2*YMP_REPIHSG_ML_US_M_0S
Now this can be replicated within the simulated panel with calling for graph>mean plus SD bounds>1std.dev or 2std.dev
However, if I select mean plus SD bounds>Fan chart the outcome is something totally different, it seems that the process has become more like an additive Brownian motion (making variables go negative, something not possible since I am using log processes for the ARIMA specification)

Do I get something wrong on what the FAN element is doing when applying it to mean? When using median the elements look to be correct (see attached word doc for comparing the charts)
Best
M

Re: building mean-fan charts using stochastic model

Posted: Mon May 10, 2021 10:20 am
by EViews Matt
Hello,

We discovered a bug in the panel mean fan chart display (option panel=meanfan) that caused the bounds to be drawn too wide. This issue will be fixed in the next patch.