Equilibrium Models Do Not Identify Causal Effects under Outcome Interdependence

This repository contains a fully reproducible Python script implementing a Monte Carlo simulation for equilibrium interaction models with outcome interdependence. The code generates synthetic data from a linear spatial autoregressive (SAR) model in which outcomes are jointly determined through an interaction network. Using the same fitted equilibrium model, the script computes and compares multiple causal estimands corresponding to different counterfactual adjustment regimes, including: partial-equilibrium effects, finite-order (local-interaction) effects, network-consistent (equilibrium) effects. The simulation evaluates how these estimands differ mechanically, prior to estimation, and how estimation bias is amplified under violations of exogeneity. The script produces summary tables reporting true values, Monte Carlo means, biases, and standard deviations, as well as publication-quality figures illustrating equilibrium amplification and the distribution of implied causal estimands under alternative assignment mechanisms. All results are generated from simulated data; no external or proprietary datasets are used. The code is self-contained, fully documented, and designed to support transparency and replication of the numerical results reported in the associated manuscript.