numerical_stability_exp

Penalizes numerical issues in \(W\): (i) deviation of each row sum from 1,
measured by the mean absolute deviation, and (ii) count of negative entries.
The score is \(\exp(-(a\,\mathrm{msd} + b\,\#\mathrm{neg}))\).

Implements a novel Bayesian disaggregation framework that combines
Principal Component Analysis (PCA) and Singular Value Decomposition (SVD)
dimension reduction of prior weight matrices with deterministic Bayesian
updating rules. The method provides Markov Chain Monte Carlo (MCMC) free
posterior estimation with built-in diagnostic metrics. While based on
established PCA (Jolliffe, 2002) <doi:10.1007/b98835> and Bayesian principles
(Gelman et al., 2013) <doi:10.1201/b16018>, the specific integration for
economic disaggregation represents an original methodological contribution.

José Mauricio Gómez Julián

BayesianDisaggregation

Bayesian Methods for Economic Data Disaggregation

numerical_stability_exp function

<dl><dt>W</dt>
<dd>Posterior matrix (\(T \times K\)).</dd>
<dt>a</dt>
<dd>Sensitivity for row-sum deviation (default <code>1000</code>).</dd>
<dt>b</dt>
<dd>Sensitivity for negative counts (default <code>10</code>).</dd></dl>

Arguments

Numerical stability (exponential penalty) — numerical_stability_exp

<dl>

<dt>W</dt>
<dd>Posterior matrix (\(T \times K\)).</dd>


<dt>a</dt>
<dd>Sensitivity for row-sum deviation (default <code>1000</code>).</dd>


<dt>b</dt>
<dd>Sensitivity for negative counts (default <code>10</code>).</dd>

</dl>

numerical_stability_exp: Numerical stability (exponential penalty)

Description

Usage

Value

Arguments

Examples