Identifiability of Contagion Components amid Environmental Fluctuations in Aggregated Default Counts

Battle Count: 79

The paper is primarily relevant to credit risk management and systemic risk assessment rather than direct trading strategies. However, understanding whether default clustering is driven by contagion versus environmental factors has implications for credit portfolio risk management, CDS pricing, and tail-risk estimation. The VaR and ES comparisons are directly relevant to risk management. The identifiability framework could inform model selection for credit risk models used in quantitative finance.

The methodology involves maximum likelihood estimation of binomial mixture models with numerical integration over latent variables. The Davis-Lo model requires computing finite mixtures of shifted binomials, the Torri model involves threshold-type activation probabilities, and the environmental extensions require numerical integration of the Probit-Normal mixture. The KL divergence calculations and variance decompositions add additional computational steps. The code is provided on GitHub, reducing implementation burden.

Full analysis and simulation code is available on GitHub (https://github.com/shintaromori/contagion-vs-macro-defaults). However, the empirical Moody's default data are proprietary and not included in the repository. Researchers with access to comparable data sources can reproduce results. The methodology is well-documented with detailed derivations in appendices.