Replication package: business discontinuation in Spanish crop agriculture

This package reproduces the central results of the paper "From Financial Fragility to Business Discontinuation in Crop Agriculture: Evidence from Spanish Firm-Level Data." It contains a single self-contained Python script that estimates the discrete-time hazard, two-year lead and three-year trajectory models, the register-status and judicial-incident validation of the outcome, the register-validated re-estimation, the nested cumulative-versus-contemporaneous test, the sample-selection analysis, and the out-of-sample evaluation (cross-validated AUC and Brier score), together with a codebook and documentation. The underlying firm-level accounting data are drawn from the SABI database (Bureau van Dijk / Moody's Analytics), which is licensed and cannot be redistributed. The datasets included here are therefore fully synthetic: they reproduce the structure and the statistical relationships of the analytical data (the event-versus-active differences, the cross-variable correlations, the panel structure, and the register-status and judicial-incident distributions) without containing any real firm record. Running the script on the synthetic data reproduces the qualitative findings of the paper; the exact published coefficients require the licensed SABI data, which can be reconstructed by any researcher with SABI access using the extraction criteria described in the paper and the accompanying codebook. Requirements: Python 3.9+ with numpy, pandas, statsmodels, scikit-learn and scipy. See README.md for instructions.