Use this function to determine which models should be skipped in the search process.
get.search.modelchecks(
estimation = TRUE,
maxConditionNumber = Inf,
minObsCount = 0,
minDof = 0,
minOutSim = 0,
minR2 = -Inf,
maxAic = Inf,
maxSic = Inf,
prediction = FALSE,
predictionBound = 10
)A list with the given options.
If TRUE, the model is estimated with all data and is ignored if this estimation fails. If FALSE, you might get a 'best model' that cannot be estimated.
A number used to ignore an estimation that has a high condition number (if implemented in the search).
An integer used to ignore an estimation where the number of observations (after dealing with NA) is low. Use 0 to disable this check.
An integer used to ignore an estimation with low degrees of freedom (equation-wise). Use 0 to disable this check.
An integer used to ignore estimations with a low number of out-of-sample simulations (if implemented in the search).
A number used to ignore estimations with a low value for 'R2' (if implemented in the search).
A number used to ignore estimations with a high 'AIC' (if implemented in the search).
A number used to ignore estimations with a high 'SIC' (if implemented in the search).
If TRUE, model data is predicted given all data and is ignored if this process fails. If FALSE, you might get a 'best model' that cannot be used for prediction.
A list containing two matrices: lower and upper, which represent the bounds for checking predictions. Each column corresponds to a target variable, and each row corresponds to a horizon. If the data has been transformed using a Box-Cox transformation, these bounds will be compared with the transformed data.
Alternatively, predictionBound can be a numeric value. In this case, the bounds are created by creating a confidence interval, assuming normality and using mean and standard errors of the growth rates.
Any model that produces a prediction outside of these bounds will be ignored. To disable this check, set predictionBound to NULL.