Core Estimators

EconIRL is a research build. These are the core estimators, the ones the project focuses on. NFXP is the reference: the exact maximum-likelihood estimator for tabular structural dynamic discrete choice, and the one with a verified paper-exact replication, matched to Rust (1987) Table IX. The rest of the core spans the structural and inverse-reinforcement-learning methods that carry the main identification stories and method lineages.

Read this page as the short menu for the main methods. If a method name is unfamiliar, open the combined chooser first, then return here for the specific method page.

For how to choose among them, including the side-by-side table, see Choosing and Comparing Estimators.

Estimator	Family	Best for
NFXP	Structural	Exact tabular DDC, replicated to Rust (1987) Table IX.
CCP	Structural	Hotz-Miller and NPL tabular DDC without a nested solve.
TD-CCP	Structural	Reward parameters without modeling the transition density.
MCE-IRL	IRL	Maximum causal entropy reward-feature matching.
Neural MCE-IRL	IRL	Unrestricted neural reward map under the MCE objective.
AIRL	IRL	Adversarial state-only reward recovery under the original AIRL transfer assumptions.
AIRL-Het	IRL	Segment-specific action-dependent rewards under exit and absorbing-state anchors.
GLADIUS	IRL	Neural Q and continuation reward recovery at scale.

GLADIUS is the core neural Q and continuation estimator.

• NFXP
• CCP
• TD-CCP
• MCE-IRL
• Neural MCE-IRL
• AIRL
• AIRL-Het
• GLADIUS