.compute_naive_cov

Estimates via shrinkage the covariance matrix of the residuals of the naive
or seasonal naive forecasts.
If the frequency of the time series is &gt; 1, the function chooses between the
two methods series-by-series according to a selection criterion.
If the frequency is 1 or is not provided, only the naive residuals are used.

internal

Provides methods for probabilistic reconciliation of hierarchical forecasts of time series.
The available methods include analytical Gaussian reconciliation (Corani et al., 2021)
<doi:10.1007/978-3-030-67664-3_13>,
MCMC reconciliation of count time series (Corani et al., 2024)
<doi:10.1016/j.ijforecast.2023.04.003>,
Bottom-Up Importance Sampling (Zambon et al., 2024)
<doi:10.1007/s11222-023-10343-y>,
methods for the reconciliation of mixed hierarchies (Mix-Cond and TD-cond) (Zambon et al., 2024)
<https://proceedings.mlr.press/v244/zambon24a.html>,
analytical reconciliation with Bayesian treatment of the covariance matrix (Carrara et al., 2025)
<doi: 10.48550/arXiv.2506.19554>.

Dario Azzimonti

bayesRecon

Probabilistic Reconciliation via Conditioning

Lorenzo Zambon

Stefano Damato

Nicolò Rubattu

Giorgio Corani

.compute_naive_cov function

<dl><dt>y_train</dt>
<dd>Multivariate time series object or numeric matrix of historical
observations, with dimensions <code>T x n</code> (rows are time points, columns are series).</dd>
<dt>freq</dt>
<dd>Positive integer seasonal frequency (optional).
If not provided and <code>y_train</code> is a multivariate time series, the frequency
of the data is used.</dd>
<dt>criterion</dt>
<dd>Character string used when <code>freq &gt; 1</code> to choose residuals.
Supported values are <code>"RSS"</code> (default) and <code>"seas-test"</code>.
<code>"RSS"</code> chooses the method with the lower residual sum of squares (RSS),
while <code>"seas-test"</code> uses a statistical test for seasonality (requires <code>forecast</code> package).</dd></dl>

Arguments

Estimate covariance from naive or seasonal-naive residuals — .compute_naive_cov

<dl>

<dt>y_train</dt>
<dd>Multivariate time series object or numeric matrix of historical
observations, with dimensions <code>T x n</code> (rows are time points, columns are series).</dd>


<dt>freq</dt>
<dd>Positive integer seasonal frequency (optional).
If not provided and <code>y_train</code> is a multivariate time series, the frequency
of the data is used.</dd>


<dt>criterion</dt>
<dd>Character string used when <code>freq &gt; 1</code> to choose residuals.
Supported values are <code>"RSS"</code> (default) and <code>"seas-test"</code>.
<code>"RSS"</code> chooses the method with the lower residual sum of squares (RSS),
while <code>"seas-test"</code> uses a statistical test for seasonality (requires <code>forecast</code> package).</dd>

</dl>

.compute_naive_cov: Estimate covariance from naive or seasonal-naive residuals

Description

Usage

Value

Arguments

See Also