Quick Start

The public MPEC surface is the lower-level estimator API. Create an estimator, call estimate, and read the returned summary object.

from econirl.environments.rust_bus import RustBusEnvironment
from econirl.estimation.mpec import MPECEstimator, MPECConfig
from econirl.preferences.linear import LinearUtility
from econirl.simulation import simulate_panel

env = RustBusEnvironment(
    operating_cost=0.01,
    replacement_cost=2.0,
    num_mileage_bins=20,
    discount_factor=0.99,
    seed=42,
)
panel = simulate_panel(env, n_individuals=100, n_periods=50, seed=123)
utility = LinearUtility.from_environment(env)

model = MPECEstimator(config=MPECConfig(solver="sqp"), verbose=False)
summary = model.estimate(
    panel=panel,
    utility=utility,
    problem=env.problem_spec,
    transitions=env.transition_matrices,
)

print([float(x) for x in summary.parameters])
print([float(x) for x in summary.standard_errors])
print(summary.policy.shape)
print(summary.metadata["final_constraint_violation"])

Output

[0.017559314146637917, 2.244328022003174]
[0.003141143824905157, 0.07669873535633087]
(20, 2)
9.265894274079756e-09

The fitted summary exposes structural parameters, standard errors, a policy, a value function, a likelihood, and constrained-optimizer diagnostics.

Attribute	Meaning
`parameters`	Estimated structural reward parameters.
`standard_errors`	Standard errors for the structural parameters.
`policy`	Estimated action probabilities by state.
`value_function`	Estimated value function by state.
`log_likelihood`	Maximized constrained choice log likelihood.
`metadata`	Solver status and Bellman constraint diagnostics.

Counterfactual Example

MPEC does not currently expose the same one-call dataframe wrapper counterfactual method as NFXP and CCP. The counterfactual evidence is therefore produced through the simulation harness, which re-solves oracle Type A, Type B, and Type C counterfactuals against the recovered structural object.

See Counterfactuals for the reported counterfactual cases and Simulation Study for the generator and results file links.

Lower-Level Control

Use econirl.estimation.mpec.MPECEstimator when you need direct control over panel objects, utility objects, transition tensors, constrained optimizer options, Bellman constraint tolerance, or diagnostic metadata.