tsprofile: Model Simulation Based Profiling
In tsmodels/tsissm: Linear Innovations State Space Unobserved Components Model

tsprofile.tsissm.estimate

R Documentation

Model Simulation Based Profiling

Description

Profiling of model dynamics using simulation/estimation/prediction.

Usage

## S3 method for class 'tsissm.estimate'
tsprofile(
  object,
  h = 1,
  nsim = 100,
  seed = NULL,
  trace = FALSE,
  sigma_scale = 1,
  solver = "nlminb",
  autodiff = TRUE,
  ...
)

Arguments

`object`	an object of class “tsissm.estimate”.
`h`	the forecast horizon on which to evaluate performance metrics.
`nsim`	the number of paths to generate.
`seed`	an object specifying if and how the random number generator should be initialized. See the simulate documentation for more details.
`trace`	whether to show the progress bar. The user is expected to have set up appropriate handlers for this using the “progressr” package.
`sigma_scale`	a scaling factor for the innovations standard deviation.
`solver`	choice of solver to use for the estimation of the paths.
`autodiff`	whether to use automatic differentiation for estimation. This makes use of the tsissmad package.
`...`	not currently used.

Details

The function profiles an estimated model by simulating and then estimating multiple paths from the assumed DGP while leaving h values out for prediction evaluation. Each simulated path is equal to the size of the original dataset plus h additional values, and initialized with the initial state vector from the model. The resulting output contains the distribution of the MAPE, percent bias (BIAS) and mean squared log relative error (MSLRE) per horizon h. Since these matrices are of class “tsmodel.distribution” they can be readily plotted with the special purpose “plot” function for this class from the “tsmethods” package. Additionally, a data.table matrix is also returned with the distribution of the coefficients from each path estimation.