Simulation Study

Read this page as an oracle-object simulation for the constrained-program route. The simulation checks whether MPEC reaches the same structural target as NFXP on a fully specified DGP.

MPEC runs on the canonical_low_action synthetic cell, the same low-dimensional action-dependent structural benchmark used for NFXP and CCP. The cell has 21 states, 3 actions, known linear reward, known transitions, and exact policy, value, Q, and Type A, Type B, and Type C counterfactual oracle objects, so every recovery claim is checked against fully specified truth.

The full result generator is mpec_run.py. It writes the results file mpec_results.json.

cd /path/to/econirl
PYTHONPATH=src:. python validation/estimators/mpec/run.py

The simulation DGP has action-dependent reward features and an exit action that anchors the reward level. The estimates are not exactly equal to truth because the panel is finite.

Evidence

MPEC appears in six simulation studies: bus engine, taxi gridworld, route choice, stockpiling, fleet maintenance, and vehicle scrappage. See the simulation studies index for what each study shows.