Structural Z-score

This dataset accompanies the manuscript “A Structural Z-Score with Prudential Default and Memory: A Forward-Looking Measure of Bank Risk”. The research hypothesis is that the traditional accounting-based Z-score can be made more informative by redefining bank default as a forward-looking breach of regulatory capital requirements rather than as full accounting insolvency. The proposed structural Z-score preserves the distance-to-default intuition of the traditional measure, but anchors the default threshold in expected risk-weighted assets, expected expenses, and current total capital adequacy. The Data folder contains annual Excel files for eight large European banks: Banco Santander, BNP Paribas, Crédit Agricole, Deutsche Bank, Lloyds, Société Générale, UBS, and UniCredit. The files include the accounting and prudential variables required to compute the structural Z-score: income, expenses, profit, assets, equity, risk-weighted assets, total capital adequacy, and the regulatory solvency requirement. They also include macro-financial variables used in the Monte Carlo calibration and stress-classification exercises. “Solvency ratio” denotes the bank-year regulatory capital requirement, not the observed effective capital ratio. Total capital adequacy is used as a stock variable. The Scripts for original computations folder contains the Python routines used to compute the structural Z-score from the empirical bank data. These scripts implement bounded-support beta-kernel density estimation, rolling-window empirical distributions, the finite-memory prediction mechanism, the fixed-point procedure, and the graphical outputs. The memory parameter k governs the prediction step: k=0 corresponds to the structural Z-score without memory, whereas k>0 introduces finite-memory conditioning through the reference point, reference dispersion, and integration domain. The Scripts for Monte Carlo Simulations folder contains the Python scripts used to build and evaluate the representative-bank Monte Carlo framework. These scripts load the empirical panel, construct global and country macro-financial factors, partition observations into normal and stress regimes, calibrate a representative-bank data-generating process, simulate alternative stress environments, and compare the traditional and structural Z-scores. The Scripts for Monte Carlo Robustness folder contains the robustness pipeline. These scripts assess whether the main conclusions are stable under alternative empirical stress classifications, prudential requirements, and distress thresholds. They reproduce the robustness tables and appendix precision diagnostics reported in the manuscript. The dataset and scripts allow replication of the empirical application, Monte Carlo validation, robustness exercises, tables, and figures. Users can adapt the code to compute structural Z-scores for other banks, provided comparable annual accounting, prudential, and macro-financial data are available.