Attach In-Sample Information to Fitted State Space Model

Attaches detailed in-sample information—such as predicted, filtered, and smoothed states and observations—to a model object fitted using any of the package’s supported SSM estimation methods. These quantities are not stored by default in model objects due to their potentially large memory footprint.

Usage

attach_insample_info(model)

Arguments

model: A fitted model object of class roams_SSM, classical_SSM, oracle_SSM, huber_robust_SSM, or trimmed_robust_SSM.

Value

A modified version of the input model object, with an additional class insample_info, and the following in-sample elements appended:

filtered_states: Filtered state estimates using data up to each time point.
predicted_states: One-step-ahead state predictions.
filtered_observations: Expected observations given data up to each time point.
predicted_observations: One-step-ahead forecasts of observations.
filtered_states_var: List of filtered state variance matrices.
predicted_states_var: List of one-step-ahead state prediction variances.
predicted_observations_var: List of one-step-ahead observation forecast variances.
mahalanobis_residuals: Vector of Mahalanobis distances of residuals from predicted observations.

For models of class roams_SSM, classical_SSM, or oracle_SSM, the following additional elements are also attached:

smoothed_states: Posterior means of hidden states using all data.
smoothed_observations: Posterior mean of the observed series based on smoothed states.
smoothed_states_var: List of smoothed state variance matrices.

These smoothed attributes are obtained using the RTS smoothing algorithm (Rauch et al. 1965).

Details

The attached outputs enable richer diagnostics, outlier inspection, and plotting. For huber_robust_SSM and trimmed_robust_SSM models, in-sample information is computed using a custom robust filtering function, and smoothed quantities (smoothed_states, smoothed_observations, and smoothed_states_var) are not available. This function should only be applied once to a model object.

References