Computer-implemented systems and methods for integrated model validation for compliance and credit risk

• US 8,515,862 B2
• Filed: 05/29/2009
• Issued: 08/20/2013
• Est. Priority Date: 05/29/2008
• Status: Active Grant

• Current Assignee: SAS Institute Incorporated
• Original Assignee: SAS Institute Incorporated
• Inventors: Zhang, Mingyuan, Abrahams, Clark Richard
• Primary Examiner(s): KANERVO, VIRPI H
• Application Number:
  US12/475,150
• Publication Number:
  US 20090299896A1
• Time in Patent Office: 1,544 Days
• Field of Search: None
• US Class Current: 705/38
• CPC Class Codes: G06Q 40/00 Finance; Insurance; Tax strategies; Processing of corporate or income taxes
• CPC Class Codes: G06Q 40/02 Banking, e.g. interest calculation or account maintenance credit or loans G06Q40/03
• CPC Class Codes: G06Q 40/03 Credit; Loans; Processing thereof
• CPC Class Codes: G06Q 40/06 Asset management; Financial planning or analysis
• CPC Class Codes: G06Q 40/08 Insurance

Abstract

Computer-implemented systems and methods are provided for model validation of a model for compliance and credit risk. Model input, output, and processing validation areas are contained on a computer system. A handle data structure connects the model validation areas with handles that comprise a unified metric. A handle represents combinations of covariate patterns and describes the joint distribution of risk characteristics.

Claims

1. A computer-implemented method for validating a credit risk model, comprising:

accessing, using one or more data processors, historical loan performance data including a plurality of records, wherein a record includes a plurality of borrower attributes and a default attribute associated with a loan, wherein the default attribute identifies whether a borrower defaulted on making payments associated with the loan, and wherein the plurality of borrower attributes includes both quantitative and qualitative components;

dividing, using the one or more data processors, each of the borrower attributes into a plurality of borrower attribute levels, wherein dividing the qualitative components includes quantizing the qualitative components into a discrete number of levels;

combining two or more borrower attribute levels;

generating, using the one or more data processors, a handle data structure including one record for each combination of borrower attribute levels, wherein the handle data structure represents a set of unified metrics including all possible combinations of covariate patterns, wherein the handle data structure is associated with a joint distribution of risk characteristics, and wherein a default probability is assigned to each record using default attributes from the historical loan performance data;

generating, using the one or more data processors, a credit risk model, wherein the credit risk model associates a risk level measurement with each record in the handle data structure; and

performing one or more validation operations on the credit risk model to determine whether the credit risk model is acceptable for use in a production environment.

2. The method of claim 1, wherein the one or more validation operations include a model performance validation operation or a model compliance validation operation.

3. The method of claim 1, wherein performing a validation operation further comprises:

determining a distribution of risk level measurements for two different groups of borrowers using the credit risk model, wherein the two different groups are mutually exclusive;

determining whether there is a statistically significant difference between the two distributions, wherein disparate treatment corresponding to one of the groups is identified when a statistically significant difference is determined; and

adjusting one or more of the risk level measurements of the credit risk model when disparate treatment corresponding to one of the groups is identified.

4. The method of claim 3, further comprising:

determining whether each distribution corresponding to each group'"'"'s handle data structure is ranked logically when adjusted by the corresponding risk level measurement.

5. The method of claim 4, further comprising:

logically ranking the distributions by sorting the distributions according to the corresponding risk level measurement.

6. The method of claim 3, further comprising:

determining whether one or more of the distributions corresponding to the handle data structure is internally inconsistent based on a logical criteria.

7. The method of claim 3, further comprising:

determining whether the distributions corresponding to each group'"'"'s handle data structure are ranked identically.

8. The method of claim 1, wherein the borrower attribute levels are subjective levels.

9. The method of claim 1, wherein a record in the handle data structure represents a covariate pattern that is predictive of the default probability.

10. The method of claim 1, wherein a record in the handle data structure represents a homogeneous segment in terms of risk, performance, and choice behavior.

11. The method of claim 1, further comprising:

segmenting borrowers based upon primary underwriting factors.

12. The method of claim 11, wherein the handle data structure allows for treatment of specific segments by embedding class priors with different thresholds and business policies.

13. The method of claim 1, wherein the risk level measurements predict the default probability.

14. The method of claim 13, wherein a risk level measurement is associated with a record in the handle data structure by a handle number.

15. The method of claim 14, wherein the association of the handle number provides a direct comparison between observed risk in the historical loan performance data and predicted risk in output data of the credit risk model.

16. The method of claim 1, wherein performance of the credit risk model is measured through credit risk and compliance risk perspectives.

17. The method of claim 1, wherein the performance of the credit risk model is measured through an optimization process.

18. The method of claim 17, wherein a risk quadrants plot is used to balance trade-offs between compliance risk and credit risk in adjusting the risk level measurements.

19. The method of claim 17, wherein outputs from the optimization process facilitate model rebuilding according to available resources.

20. The method of claim 1, wherein the handle data structure integrates the quantitative and qualitative components.

21. The method of claim 1, wherein the credit risk model is developed through a model consensus session to provide model evaluation, specifications, and selection.

22. The method of claim 1, further comprising:

performing override analysis by matching override incidence to the records in the handle data structure, wherein results from the override analysis are used as feedback and in creation of records in the handle data structure and model performance evaluation.

23. The method of claim 1, wherein the qualitative components are based on subjective judgment assessments of conditions.

24. A system for validating a credit risk model, comprising:

one or more processors;

one or more computer-readable storage mediums containing instructions configured to cause the one or more processors to perform operations including;

accessing historical loan performance data including a plurality of records, wherein a record includes a plurality of borrower attributes and a default attribute associated with a loan, wherein the default attribute identifies whether a borrower defaulted on making payments associated with the loan, and wherein the plurality of borrower attributes includes both quantitative and qualitative components;

dividing each of the borrower attributes into a plurality of borrower attribute levels, wherein dividing the qualitative components includes quantizing the qualitative components into a discrete number of levels;

combining two or more borrower attribute levels;

generating, using the one or more data processors, a handle data structure including one record for each combination of borrower attribute levels, wherein the handle data structure represents a set of unified metrics including all possible combinations of covariate patterns, wherein the handle data structure is associated with a joint distribution of risk characteristics, and wherein a default probability is assigned to each record using default attributes from the historical loan performance data;

generating, using the one or more data processors, a credit risk model, wherein the credit risk model associates a risk level measurement with each record in the handle data structure; and

performing one or more validation operations on the credit risk model to determine whether the credit risk model is acceptable for use in a production environment.

25. The system of claim 24, wherein the one or more validation operations include a model performance validation operation or a model compliance validation operation.

26. The system of claim 24, wherein performing a validation operation further comprises:

determining a distribution of risk level measurements for two different groups of borrowers using the credit risk model, wherein the two different groups are mutually exclusive;

determining whether there is a statistically significant difference between the two distributions, wherein disparate treatment corresponding to one of the groups is identified when a statistically significant difference is determined; and

adjusting one or more of the risk level measurements of the credit risk model when disparate treatment corresponding to one of the groups is identified.

27. The system of claim 26, wherein the operations further include:

determining whether each distribution corresponding to each group'"'"'s handle data structure is ranked logically when adjusted by the corresponding risk level measurement.

28. The system of claim 27, wherein the operations further include:

logically ranking the distributions by sorting the distributions according to the corresponding risk level measurement.

29. The system of claim 26, wherein the operations further include:

determining whether one or more of the distributions corresponding to of handle data structure is internally inconsistent based on a logical criteria.

30. The system of claim 26, wherein the operations further include:

determining whether the distributions corresponding to each group'"'"'s handle data structure are ranked identically.

31. The system of claim 24, wherein the borrower attribute levels are subjective levels.

32. The system of claim 24, wherein a record in the handle data structure represents a covariate pattern that is predictive of the default probability.

33. The system of claim 24, wherein a record in the handle data structure represents a homogeneous segment in terms of risk, performance, and choice behavior.

34. The system of claim 24, wherein the operations further include:

segmenting borrowers based upon primary underwriting factors.

35. The system of claim 34, wherein the handle data structure allows for treatment of specific segments by embedding class priors with different thresholds and business policies.

36. The system of claim 24, wherein the risk level measurements predict the default probability.

37. The system of claim 36, wherein a risk level measurement is associated with a record in the handle data structure by a handle number.

38. The system of claim 37, wherein the association of the handle number provides a direct comparison between observed risk in the historical loan performance data and predicted risk in output data of the credit risk model.

39. The system of claim 24, wherein performance of the credit risk model is measured through credit risk and compliance risk perspectives.

40. The system of claim 24, wherein the performance of the credit risk model is measured through an optimization process.

41. The system of claim 40, wherein a risk quadrants plot is used to balance trade-offs between compliance risk and credit risk in adjusting the risk level measurements.

42. The system of claim 40, wherein outputs from the optimization process facilitate model rebuilding according to available resources.

43. The system of claim 34, wherein the handle data structure integrates the quantitative and qualitative components.

44. The system of claim 24, wherein the credit risk model is developed through a model consensus session to provide model evaluation, specifications, and selection.

45. The system of claim 24, wherein the operations further include:

performing override analysis by matching override incidence to the records in the handle data structure, wherein results from the override analysis are used as feedback and in creation of records in the handle data structure and model performance evaluation.

46. The system of claim 24, wherein the qualitative components are based on subjective judgment assessments of conditions.

47. A computer-program product for validating a credit risk model, tangibly embodied in a non-transitory machine-readable storage medium, including instructions executable to cause a data processing apparatus to:

access historical loan performance data including a plurality of records, wherein a record includes a plurality of borrower attributes and a default attribute associated with a loan, wherein the default attribute identifies whether a borrower defaulted on making payments associated with the loan, and wherein the plurality of borrower attributes includes both quantitative and qualitative components;

divide each of the borrower attributes into a plurality of borrower attribute levels, wherein dividing the qualitative components includes quantizing the qualitative components into a discrete number of levels;

combine two or more borrower attribute levels;

generate, using the one or more data processors, a handle data structure including one record for each combination of borrower attribute levels, wherein the handle data structure represents a set of unified metrics including all possible combinations of covariate patterns, wherein the handle data structure is associated with a joint distribution of risk characteristics, and wherein a default probability is assigned to each record using default attributes from the historical loan performance data;

generate, using the one or more data processors, a credit risk model, wherein the credit risk model associates a risk level measurement with each record in the handle data structure; and

perform one or more validation operations on the credit risk model to determine whether the credit risk model is acceptable for use in a production environment.

48. The computer-program product of claim 47, wherein the one or more validation operations include a model performance validation operation or a model compliance validation operation.

49. The computer-program product of claim 47, wherein performing a validation operation further comprises:

determining a distribution of risk level measurements for two different groups of borrowers using the credit risk model, wherein the two different groups are mutually exclusive;

determining whether there is a statistically significant difference between the two distributions, wherein disparate treatment corresponding to one of the groups is identified when a statistically significant difference is determined; and

adjusting one or more of the risk level measurements of the credit risk model when disparate treatment corresponding to one of the groups is identified.

50. The computer-program product of claim 49, wherein the instructions are further executable to cause the data processing apparatus to:

determine whether each distribution corresponding to each group'"'"'s handle data structure is ranked logically when adjusted by the corresponding risk level measurement.

51. The computer-program product of claim 50, wherein the instructions are further executable to cause the data processing apparatus to:

logically rank the distributions by sorting the distributions according to the corresponding risk level measurement.

52. The computer-program product of claim 49, wherein the instructions are further executable to cause the data processing apparatus to:

whether one or more of the distributions corresponding to of handle data structure is internally inconsistent based on a logical criteria.

53. The computer-program product of claim 49, wherein the instructions are further executable to cause the data processing apparatus to:

determine whether the distributions corresponding to each group'"'"'s handle data structure are ranked identically.

54. The computer-program product of claim 47, wherein the borrower attribute levels are subjective levels.

55. The computer-program product of claim 47, wherein a record in the handle data structure represents a covariate pattern that is predictive of the default probability.

56. The computer-program product of claim 47, wherein a record in the handle data structure represents a homogeneous segment in terms of risk, performance, and choice behavior.

57. The computer-program product of claim 47, wherein the instructions are further executable to cause the data processing apparatus to:

segment borrowers based upon primary underwriting factors.

58. The computer-program product of claim 57, wherein the handle data structure allows for treatment of specific segments by embedding class priors with different thresholds and business policies.

59. The computer-program product of claim 47, wherein the risk level measurements predict the default probability.

60. The computer-program product of claim 59, wherein a risk level measurement is associated with a record in the handle data structure by a handle number.

61. The computer-program product of claim 60, wherein the association of the handle number provides a direct comparison between observed risk in the historical loan performance data and predicted risk in output data of the credit risk model.

62. The computer-program product of claim 47, wherein performance of the credit risk model is measured through credit risk and compliance risk perspectives.

63. The computer-program product of claim 47, wherein the performance of the credit risk model is measured through an optimization process.

64. The computer-program product of claim 63, wherein a risk quadrants plot is used to balance trade-offs between compliance risk and credit risk in adjusting the risk level measurements.

65. The computer-program product of claim 63, wherein outputs from the optimization process facilitate model rebuilding according to available resources.

66. The computer-program product of claim 47, wherein the handle data structure integrates the quantitative and qualitative components.

67. The computer-program product of claim 47, wherein the credit risk model is developed through a model consensus session to provide model evaluation, specifications, and selection.

68. The computer-program product of claim 47, wherein the instructions are further executable to cause the data processing apparatus to:

perform override analysis by matching override incidence to the records in the handle data structure, wherein results from the override analysis are used as feedback and in creation of records in the handle data structure and model performance evaluation.

69. The computer-program product of claim 47, wherein the qualitative components are based on subjective judgment assessments of conditions.