Information-Theoretic Methods for Econometric Estimation
infometrics is an R package implementing
Information-Theoretic (IT) estimators for econometric models, following
the unified framework of Golan (2008), Information and Entropy
Econometrics — A Review and Synthesis (doi:10.1561/0800000004).
The package focuses on the Generalized Maximum Entropy (GME) and
Generalized Cross-Entropy (GCE) family: each unknown is reparameterized
as the mean of a distribution on a bounded support, and the resulting
entropy problem is solved subject to the data.
Every estimator solves the dual (concentrated) model
— working in Lagrange-multiplier space via BFGS with analytic gradients
— which is computationally efficient and needs no compiled code. The
package is pure R and depends only on stats.
# install.packages("remotes")
remotes::install_github("GenMaxEnt/infometrics_R")One function per estimator family: uniform priors give GME,
user-supplied priors (p0, w0) give
GCE — there are no separate gce_* twins.
Regression
linreg() — lm-style GCE linear regression, with
predict(), an entropy-ratio (ER) coefficient test, and
prior-relative normalized entropy.linreg_iv() — GME instrumental-variables regression
(just- and over-identified), with sandwich / delta / bootstrap standard
errors.panel_gce() — one-way error-components panel model,
with standard errors for the coefficients, individual effects, and
support weights.Inverse problems
inverse_ce() — pure-moment ME/CE inverse problem
(formula interface), with information-matrix standard errors and Fano
error bounds.inverse_noise() — noisy-moment GME/GCE inverse problem,
with sandwich / delta / bootstrap standard errors and Fano bounds.Matrix balancing and Markov chains
matrix_ce() / matrix_gce() — exact and
noisy-moment matrix balancing.markov_ce() / markov_gce() — exact and
noisy-moment first-order Markov transition matrices, with marginal
effects and Fano bounds.Multinomial response
multinomial_gce() — nu-weighted GCE multinomial
(Golan–Judge–Perloff), with marginal effects (robust SEs) and Fano
bounds.mixed_gce() — doubly-reparameterized GCE mixed
model.Utilities — shannon_entropy(),
make_support(), default_supports(),
normalize_data(), plus the margins() and
fano_bounds() generics.
Every estimator returns an S3 object inheriting from
"infometrics" with print,
summary, coef, fitted, and
residuals methods (and, where applicable,
vcov, predict, margins, and
fano_bounds).
vignette("inverse-problems", package = "infometrics")
vignette("gme-gce-regression", package = "infometrics")linreg() and linreg_iv(): GME
under collinearity, GCE with an informative prior, the entropy weight
nu, entropy diagnostics and hypothesis tests, and an
instrumental-variables example.Support spaces are the user’s most consequential input. The dual is
bounded below only if the data are representable within the supports —
if they are not, the multipliers diverge and estimates collapse to a
support boundary. The estimators check the first-order condition at the
optimum (foc_residual) and warn when it is violated; the
remedy is a wider error support v (and/or
Z).
Golan, A. (2008). Information and Entropy Econometrics — A Review and Synthesis. Foundations and Trends in Econometrics, 2(1–2), 1–145. doi:10.1561/0800000004
GPL-3.