Dataset of Financial Risk Indicators for Chinese Universities (2015–2025)

This record contains three fully synthetic balanced panels and the materials required to reproduce a controlled methodological experiment in higher- education financial-risk modelling. Each panel contains 3,000 non-identifiable synthetic analytical units observed annually from 2015 through 2025 (33,000 unit--year records). The Weak, Moderate, and Strong panels use prespecified temporal-persistence coefficients of 0.20, 0.50, and 0.80, respectively, yielding 99,000 records in total. The deposit includes frozen parameters and random seeds, a data dictionary, MATLAB generation and independent integrity-audit programs and outputs, official Python programs and version-locked results for Task 1 and Task 2, realized split and rolling-fold manifests, selected hyperparameters, run logs, record-level predictions and aligned three-class probabilities, confusion matrices, and output-verification reports. Task 1 is a contemporaneous recovery audit of a disclosed deterministic rating rule. Task 2 uses predictors at t-1 to classify the synthetic rating at t under an eight-year rolling-origin evaluation. No deposited record represents or can be linked to an actual higher education institution. The outcome is a deterministic rule-based synthetic rating, not an observed financial-distress event or expert-adjudicated label. The deposit supports reproducibility of the disclosed synthetic experiment but not claims of real-world prevalence, national representativeness, external predictive validity, causal effects, or prospective deployment performance. ## Keywords - synthetic panel data - higher education finance - financial risk modelling - target leakage - temporal validation - rolling-origin evaluation - reproducible research - classification