Scholarlink

<p>The growing level of household indebtedness amid macroeconomic instability increases the need for tools for early diagnosis of mortgage portfolio credit risk. Despite the widespread use of probability of default (PD) models, their limited sensitivity to borrower behaviour and digital characteristics reduces the effectiveness of targeted risk mitigation. This creates a research gap in terms of integrating digital indicators into credit ranking at the borrower and portfolio segment levels. The aim of this study is to develop and empirically test a hybrid composite model (HPCS) that combines traditional risk parameters (PD, LGD, EAD) and digital behavioural indices of the borrower (TRI, ARI) within the concept of a digital twin. The methodological basis includes weight calculation using the CRITIC method, robust normalisation, and empirical testing on a sample of 1,000,000 mortgage loans based on Fannie Mae data for 2022–2024. To assess the model’s effectiveness, we propose the EL-capture metric, which reflects the share of expected losses covered in the portfolio at a given level of intervention. The results show that with 10% coverage, the HPCS model covers 3.8 percentage points more expected losses than the PD model. The HPCS and EL heat maps demonstrate spatial consistency across high-risk segments in the LTV<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\times\)</EquationSource> </InlineEquation>DTI coordinates. The data obtained confirm the applicability of the model for both micro-level prioritisation of credit claims and macro-level portfolio management. The scientific novelty of the research lies in the formalisation of the borrower’s digital twin in a credit risk model that combines probabilistic, cost and behavioural components into an interpretable portfolio prioritisation index. The practical significance lies in the possibility of using HPCS for targeted risk mitigation under resource constraints. In the future, it is planned to expand the model by taking into account behavioural dynamics over time, climate and social sustainability factors, as well as conducting out-of-time validation and testing on alternative data.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Digital twin-enhanced credit risk prioritization in mortgage portfolios: a hybrid model approach

  • 期刊论文

  • 发表日期：

    2026年1月29日

查看全文
  • Sergey Barykin,
  • Aleksey Shulga,
  • Daria Dinets,
  • Alexey Mikhaylov

摘要

The growing level of household indebtedness amid macroeconomic instability increases the need for tools for early diagnosis of mortgage portfolio credit risk. Despite the widespread use of probability of default (PD) models, their limited sensitivity to borrower behaviour and digital characteristics reduces the effectiveness of targeted risk mitigation. This creates a research gap in terms of integrating digital indicators into credit ranking at the borrower and portfolio segment levels. The aim of this study is to develop and empirically test a hybrid composite model (HPCS) that combines traditional risk parameters (PD, LGD, EAD) and digital behavioural indices of the borrower (TRI, ARI) within the concept of a digital twin. The methodological basis includes weight calculation using the CRITIC method, robust normalisation, and empirical testing on a sample of 1,000,000 mortgage loans based on Fannie Mae data for 2022–2024. To assess the model’s effectiveness, we propose the EL-capture metric, which reflects the share of expected losses covered in the portfolio at a given level of intervention. The results show that with 10% coverage, the HPCS model covers 3.8 percentage points more expected losses than the PD model. The HPCS and EL heat maps demonstrate spatial consistency across high-risk segments in the LTV \(\times\) DTI coordinates. The data obtained confirm the applicability of the model for both micro-level prioritisation of credit claims and macro-level portfolio management. The scientific novelty of the research lies in the formalisation of the borrower’s digital twin in a credit risk model that combines probabilistic, cost and behavioural components into an interpretable portfolio prioritisation index. The practical significance lies in the possibility of using HPCS for targeted risk mitigation under resource constraints. In the future, it is planned to expand the model by taking into account behavioural dynamics over time, climate and social sustainability factors, as well as conducting out-of-time validation and testing on alternative data.