Financial instability is traditionally measured using indicators such as volatility levels, financial stress indices, or forecast errors, limiting the ability to capture the state-conditional and distributional properties of market dynamics. In this study, financial instability is reformulated as deviations from the conditional return distribution under the prevailing macro-financial state. To operationalize this formulation, a latent macro-financial state is estimated using a Dynamic Factor Model and integrated with KOSPI returns through an AI-based conditional density modeling framework consisting of a Conditional Time Variational Autoencoder combined with a state-conditional spline-flow density. Financial instability is then measured as the negative log-likelihood of the observed return under the estimated conditional density. The resulting index aligns with established benchmarks such as the CBOE Volatility Index and the South Korea Financial Instability Index, while capturing state-dependent distributional abnormalities that are not fully reflected in conventional volatility-based measures. It exhibits heightened sensitivity to periods of acute financial stress and identifies state-dependent anomalies that remain largely undetected by existing indicators. The proposed framework establishes a probabilistic and distribution-aware interpretation of financial instability, providing an interpretable foundation for sustainable financial risk management and long-term financial resilience beyond traditional volatility-based approaches.
Jeon et al. (Wed,) studied this question.