Alternatives when no model is acceptable • pickmdl3

When using the pickdmlselection procedure, it may be happen that none of the five models on the pickmdl list fulfills the three predefined criteria. In such cases, the default choiche is to choose the first model on the list, that is the AIRLINE model. It may however be that other models fits the data well. In such cases the pickmodel procedure selects a poorly fitted model, even though an acceptable model would have been available to the automodel procedure. To mitigate this risk the pickmdl3 package provides the alternative of falling back on the automodel approach when none of the five listed models proves adequate. This is done by setting the option pickmdl_method in x13_pickmdl.

pickmdl_method = first

Let us first show that the pickmdl procedure by default falls back on the AIRLINE model when none of the models on the list passes the pre-defined criteria. This is the default option pickmdl_method = "first". We select the standard specification rsa3and the Norwegian retail index for nace 47.2 to illustrate the case.



rti_472 <- pickmdl3::pickmdl_data("norwegian_rti")$rti_472
spec_now <- x13_spec("rsa3")

model_now <- pickmdl3::x13_pickmdl(rti_472,spec_now, pickmdl_method="first")
#> Warning in crit_selection(crit_tab_i, star = 0, when_star = when_star_here): No
#> model is ok according to criteria

We see that a warning is given that tells us that none of the models on the pickmdl list are acceptable. The ok function tells us further that the first model is selected, i.e. the AIRLINE model, but that this model is not acceptable according to the criteria.

pickmdl3::ok(model_now)
#> $ok
#> [1] FALSE
#> 
#> $ok_final
#> [1] FALSE
#> 
#> $mdl_nr
#> [1] 1

If the user wants to change the default model, this may be done by setting the star parameter. In the example below, the procedure now falls back on the third model on the list.


model_now <- pickmdl3::x13_pickmdl(rti_472,spec_now, pickmdl_method="first", star=3)
#> Warning in crit_selection(crit_tab_i, star = 0, when_star = when_star_here): No
#> model is ok according to criteria
pickmdl3::ok(model_now)
#> $ok
#> [1] FALSE
#> 
#> $ok_final
#> [1] FALSE
#> 
#> $mdl_nr
#> [1] 3

pickmdl_method = first_automdl

To let the x13_pickdml fall back on the automodel procedure when none of the models on the pickmdl list are acceptable, set the pickmdl_method = try_automdl. Below this is illustrated with the norwegian retail index for nace 47.51. A warning tells the user that the procedure has switched to the automdl procedure, which means that none of the models on the list passed the criteria. From the output of ok we now see that the selected model fulfills the criteria, however. The model number is 6, which means that this is a model selected by the automodel procedure. The model output shows that the model in question is of order $(1,0,2)(1,1,1)_s$ .

rti_4751 <- pickmdl3::pickmdl_data("norwegian_rti")$rti_4751
spec_now <- x13_spec("rsa4")|>
  set_outlier(outliers.type=NULL) 
model_now <-pickmdl3::x13_pickmdl(rti_4751, spec=spec_now,
                                  pickmdl_method = "first_automdl")
#> automdl since no pickmdl model ok

pickmdl3::ok(model_now)
#> $ok
#> [1] TRUE
#> 
#> $ok_final
#> [1] TRUE
#> 
#> $mdl_nr
#> [1] 6

model_now
#> Serie span: All 
#> 
#> Model: X-13
#> Log-transformation: yes 
#> SARIMA model: (1,0,2) (1,1,1)
#> 
#> SARIMA coefficients:
#>    phi(1)  theta(1)  theta(2)   bphi(1) btheta(1) 
#>   -0.8092   -0.4119   -0.1584   -0.4275   -1.0000 
#> 
#> Regression model:
#>    const   easter 
#>  0.03745 -0.06408 
#> 
#>  Seasonal filter: FILTER_S3X5;  Trend filter: H-23 terms
#>  M-Statistics: q Good (0.590); q-m2 Good (0.627)
#>  QS test on SA: Good (0.444);  F-test on SA: Good (0.995)
#> 
#> For a more detailed output, use the 'summary()' function.

pickmdl_method = first_tryautomdl

There are cases where neither the pickmdl procedure nor the automdl procedure will be able to identify a model that passes the criteria. In such cases one should consider falling back on a parsimonious default model, e.g. the AIRLINE model, although this model too provides seasonal adjustment of poor quality. This may still be considered a better strategy than selecting the optimally fitted model with the automodel procedure, as this approach now introduces the risk of model change in addition to the poor quality of an ill fitted model.

When the option pickmdl_method is set to first_tryautomdl, the x13_pickmdl function first checks the models on the pickmdl list. If none of these five models fulfills the criteria, a model is selected with the automodel procedure. If this model too does not fulfill the thre criteria, the default model defined by star is selected.

spec_now <- x13_spec("rsa3")

model_now <- pickmdl3::x13_pickmdl(rti_472,spec_now, pickmdl_method="first_tryautomdl")
#> Warning in crit_selection(crit_tab_i, star = 0, when_star = when_star_here): No
#> model is ok according to criteria
pickmdl3::ok(model_now)
#> $ok
#> [1] FALSE
#> 
#> $ok_final
#> [1] FALSE
#> 
#> $mdl_nr
#> [1] 1