A carregar...
Publicação
Credit Conversion Factor (CCF) segmentation within the region of instability
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 387.04 KB | Adobe PDF |
Orientador(es)
Resumo(s)
The aim of this project was to develop a segmentation solely for observations inside the region of instability for estimating the credit conversion factor (CCF) parameter. It involved two components: variable selection, including the removal of variables with over 50% missing values, variables dominated by a single value in more than 80% of observations, and the identification of highly correlated ation coefficient exceeding 0.95 for subsequent analysis, followed by the construction of a decision tree. Once the segmentation was developed, a set of tests was executed to assess the results and their impact. These included a discriminatory power test, which evaluated the performance of the segmentation, and three cliff effect analyses examining how the new segmentation would affect the exposure at default (EAD) values compared to the segmentation currently in use, which is based on a decision tree constructed exclusively from data outside the region of instability but applied to all observations. The results of these tests showed that the new segmentation differentiated observations with higher and lower CCF levels more effectively and mitigated the impact on EAD computation when compared to the original estimates (i.e. applying a segmentation developed solely from observations outside the region of instability to those inside). Several software tools were used during this project, including SAS®, Python, SAS® Enterprise Miner and Excel. SAS® was primarily used for data preparation, parameter estimation and postsegmentation testing, specifically for the cliff effect analyses. The variable selection step was developed in Python, while the decision tree was constructed using SAS® Enterprise Miner . An additional analysis to determine the maximum number of levels of the tree was also conducted in Python. Finally, auxiliary calculations for the discriminatory power test were performed in Excel, along with the verification of its correct implementation in SAS®.
Descrição
Trabalho de Projeto de Mestrado, Estatística e Investigação Operacional, 2026, Universidade de Lisboa, Faculdade de Ciências
Palavras-chave
Credit Conversion Factor Region of instability Segmentation Discriminatory power Cliff effect