Assessment of loan losses after default (Оценка потерь для ссуд после дефолта)

Battle Count: 80

The paper is primarily relevant to credit risk management and banking regulation rather than quantitative trading. However, LGD estimation is relevant for: (1) credit default swap (CDS) pricing, (2) loan portfolio trading and securitization (ABS/MBS), (3) credit risk factor models used in multi-asset strategies, (4) regulatory capital optimization for banks that participate in trading. The Bayesian recovery model could inform credit spread modeling but is not directly applicable to typical quantitative trading strategies.

The mathematical framework is moderately complex but well-defined. Key implementation steps include: (1) computing discounted recoveries with proportional costs, (2) estimating T via two-point or optimization method, (3) implementing the Bayesian update formula, (4) computing LGD(t) via the closed-form expression. The main complexity lies in data preparation (matching repayment events to defaulted loans, computing EAD at each date) and choosing appropriate discount rates. No ML training is required; the model is parametric and analytical.

The paper provides detailed formulas and methodology but lacks publicly available code or data. The empirical examples reference specific bank portfolios (housing: 2374 defaults, consumer: 29500 defaults) that are not publicly accessible. The Bayesian framework and exponential recovery model are well-specified mathematically, but practical implementation requires proprietary bank data on defaulted loans, repayment schedules, and collection costs.